Changeset - 5da3dcf76c51
language_spec.md
Show inline comments
 
# Protocol Description Language
 

	
 
## Introduction
 

	
 
## Grammar
 

	
 
Beginning with the basics from which we'll construct the grammar, various characters and special variations thereof:
 

	
 
```
 
SP = " " // space 
 
HTAB = 0x09 // horizontal tab
 
VCHAR = 0x21-0x7E // visible ASCII character
 
VCHAR-ESCLESS = 0x20-0x5B | 0x5D-0x7E // visible ASCII character without "\"
 
WSP = SP | HTAB // whitespace
 
ALPHA = 0x41-0x5A | 0x61-0x7A // characters (lower and upper case)
 
DIGIT = 0x30-0x39 // digit
 
NEWLINE = (0x15 0x0A) | 0x0A // carriage return and line feed, or just line feed
 

	
 
// Classic backslash escaping to produce particular ASCII charcters
 
ESCAPE_CHAR = "\"
 
ESCAPED_CHARS = 
 
    ESCAPE_CHAR ESCAPE_CHAR |
 
    ESCAPE_CHAR "t" |
 
    ESCAPE_CHAR "r" |
 
    ESCAPE_CHAR "n" |
 
    ESCAPE_CHAR "0" |
 
    ESCAPE_CHAR "'" |
 
    ESCAPE_CHAR """
 
```
 

	
 
Which are composed into the following components of an input file that do not directly contribute towards the AST:
 

	
 
```
 
// asterisk followed by any ASCII char, excluding "/", or just any ASCII char without "*"
 
block-comment-contents = "*" (0x00-0x2E | 0x30-0x7E) | (0x20-0x29 | 0x2B-0x7E)
 
block-comment = "/*" block-comment-contents* "*/"
 
line-comment = "//" (WSP | VCHAR)* NEWLINE
 
comment = block-comment | line-comment
 
cw = (comment | WSP | NEWLINE)*
 
cwb = (comment | WSP | newline)+
 
```
 

	
 
Where it should be noted that the `cw` rule allows for not encountering any of the indicated characters, while the `cwb` rule expects at least one instance.
 

	
 
The following operators are defined:
 

	
 
```
 
binary-operator = "||" | "&&" | 
 
                  "!=" | "==" | "<=" | ">=" | "<" | ">" |
 
                  "|" | "&" | "^" | "<<" | ">>" |
 
                  "+" | "-" | "*" | "/" | "%"
 
assign-operator = "=" |
 
                  "|=" | "&=" | "^=" | "<<=" | ">>=" |
 
                  "+=" | "-=" | "*=" | "/=" | "%="
 
unary-operator = "++" | "--" | "+" | "-" | "~" | "!"
 
```
 

	
 
**QUESTION**: Do we include the pre/postfix "++" and "--" operators? They were introduced in C to reduce the amount of required characters. But is still necessary?
 

	
 
And to define various constants in the language, we allow for the following:
 

	
 
```
 
// Various integer constants, binary, octal, decimal, or hexadecimal, with a 
 
// utility underscore to enhance humans reading the characters. Allowing use to
 
// write something like 100_000_256 or 0xDEAD_BEEF
 
int-bin-char = "0" | "1"
 
int-bin-constant = "0b" int-bin-char (int-bin-char | "_")* // 0b0100_1110
 
int-oct-char = "0"-"7"
 
int-oct-constant = "0o" int-oct-char (int-oct-char | "_")* // 0o777
 
int-dec-constant = DIGIT (DIGIT | "_")* //
 
int-hex-char = DIGIT | "a"-"f" | "A"-"F" 
 
int-hex-constant = "0x" int-hex-char (int-hex-char | "_")* // 0xFEFE_1337
 
int-constant = int-bin-constant | int-oct-constant | int-dec-constant | int-hex-constant
 

	
 
// Floating point numbers
 
// TODO: Maybe support exponential notation? Seems silly for a networking
 
//  language, but might be useful? 
 
float-constant = DIGIT* "." DIGIT+
 

	
 
// Character constants: a single character. Its element may be an escaped 
 
// character or a VCHAR (excluding "'" and "\")
 
char-element = ESCAPED_CHARS | (0x20-0x26 | 0x28-0x5B | 0x5D-0x7E) 
 
char-constant = "'" char-element "'"
 

	
 
// Same thing for strings, but these may contain 0 or more characters
 
str-element = ESCAPED_CHARS | (0x20-0x21 | 0x23-0x5B | 0x5D-0x7E)
 
str-constant = """ str-element* """
 
```
 

	
 
Note that the integer characters are forced, somewhat arbitrarily without hampering the programmer's expressiveness, to start with a valid digit. Only then may one introduce the `_` character. And non-rigorously speaking characters may not contain an unescaped `'`-character, and strings may not contain an unescaped `"`-character.
 

	
 
We now introduce the various identifiers that exist within the language, we make a distinction between "any identifier" and "any identifier except for the builtin ones". Because we h
 

	
 
```
 
identifier-any = ALPHA | (ALPHA | DIGIT | "_")*
 
keyword = 
 
    "composite" | "primitive" |
 
    type-primitive | "true" | "false" | "null" |
 
    "struct" | "enum" |
 
    "if" | "else" |
 
    "while" | "break" | "continue" | "return" |
 
    "synchronous" | "assert" |
 
    "goto" | "skip" | "new" | "let"
 
builtin = "put" | "get" | "fires" | "create" | "assert"
 
identifier = identifier-any WITHOUT (keyword | builtin)
 

	
 
// Identifier with any number of prefixed namespaces
 
ns-identifier = (identifier "::")* identifier
 
```
 

	
 
We then start introducing the type system. Learning from the "mistake" of C/C++ of having types like `byte` and `short` with unspecified and compiler-dependent byte-sizes (followed by everyone using `stdint.h`), we use the Rust/Zig-like `u8`, `i16`, etc. Currently we will limit the programmer to not produce integers which take up more than 64 bits. Furthermore, as one is writing network code, it would be quite neat to be able to put non-byte-aligned integers into a struct in order to directly access meaningful bits. Hence, with restrictions introduced later, we will allow for types like `i4` or `u1`. When actually retrieving them or performing computations with them we will use the next-largest byte-size to operate on them in "registers".
 

	
 
**Question**: Difference between u1 and bool? Do we allow assignments between them? What about i1 and bool?
 

	
 
As the language semantics are value-based, we are prevented from returning information from functions through its arguments. We may only return information through its (single) return value. If we consider the common case of having to parse a series of bytes into a meaningful struct, we cannot return both the struct and a value as a success indicator. For this reason, we introduce algebraic datatypes (or: tagged unions, or: enums) as well.
 

	
 
Lastly, since functions are currently without internal side-effects (since functions cannot perform communication with components, and there is no functionality to interact "with the outside world" from within a function), it does not make sense to introduce the "void" type, as found in C/C++ to indicate that a function doesn't return anything of importance. However, internally we will allow for a "void" type, this will allow treating builtins such as "assert" and "put" like functions while constructing and evaluating the AST.
 

	
 
```
 
// The digits 1-64, without any leading zeros allowed, to allow specifying the
 
// signed and unsigned integer types
 
number-1-64 = NZ-DIGIT | (0x31-0x35 DIGIT) | ("6" 0x30-0x34)
 
type-signed-int = "i" number-1-64 // i1 through i64
 
type-unsigned-int = "u" number-1-64 // u1 through u64
 

	
 
// Standard floats and bools
 
type-float = "f32" | "f64"
 
type-bool = "bool"
 

	
 
// Messages, may be removed later
 
type-msg = "msg"
 

	
 
// Indicators of port types
 
type-port = "in" | "out"
 

	
 
// Unions and tagged unions, so we allow:
 
// enum SpecialBool { True, False }
 
// enum SpecialBool{True,False,}
 
// enum Tagged{Boolean(bool),SignedInt(i64),UnsignedInt(u64),Nothing}
 
type-union-element = identifier cw (("(" cw type cw ")") | ("=" cw int-constant))? 
 
type-union-def = "enum" cwb identifier cw "{" cw type-union-element (cw "," cw type-union-element)* (cw ",")? cw "}"
 

	
 
// Structs, so we allow:
 
// struct { u8 type, u2 flag0, u6 reserved }
 
type-struct-element = type cwb identifier
 
type-struct-def = "struct" cwb identifier cw "{" cw type-struct-element (cw "," cw type-struct-element)* (cw ",")? cw "}"
 

	
 
type-primitive = type-signed-int |
 
    type-unsigned-int |
 
    type-float |
 
    type-bool |
 
    type-msg |
 
    type-port
 
    
 
// A type may be a user-defined type (e.g. "struct Bla"), a namespaced
 
// user type (e.g. "Module::Bla"), or a non-namespaced primitive type. We 
 
// currently have no way (yet) to access nested modules, so we don't need to 
 
// care about identifier nesting.
 
type = type-primitive | ns-identifier
 
```
 

	
 
With these types, we need to introduce some extra constant types. Ones that are used to construct struct instances and ones that are used to construct/assign enums. These are constructed as:
 

	
 
```
 
// Struct literals
 
struct-constant-element = identifier cw ":" cw expr
 
struct-constant = ns-identifier cw "{" cw struct-constant-element (cw "," struct-constant-element)* cw "}"
 

	
 
enum-constant = ns-identifier "::" identifier cw "(" cw expr cw ")" 
 
```
 

	
 
Finally, we declare methods and field accessors as:
 

	
 
```
 
method = builtin | ns-identifier
 

	
 
field = "length" | identifier
 
```
 

	
 
**Question**: This requires some discussion. We allow for a "length" field on messages, and allow the definition of arrays. But if we wish to perform computation in a simple fashion, we need to allow for variable-length arrays of custom types. This requires builtin methods like "push", "pop", etc. But I suppose there is a much nicer way... In any case, this reminds me of programming in Fortran, which I definitely don't want to impose on other people (that, or I will force 72-character line lengths on them as well)
 

	
 
When we parse a particular source file, we may expect the following "pragmas" to be sprinkled at the top of the source
 
file. They may exist at any position in the global scope of a source file.
 

	
 
```
 
// A domain identifier is a dot-separated sequence of identifiers. As these are
 
// only used to identify modules we allow any identifier to be used in them. 
 
// The exception is the last identifier, which we, due to namespacing rules,
 
// force to be a non-reserved identifier.
 
domain-identifier = (identifier-any ".")* identifier
 

	
 
pragma-version = "#version" cwb int-constant cw ";" // e.g. #version 500
 
pragma-module = "#module" cwb domain-identifier cw ";" // e.g. #module hello.there
 

	
 
// Import, e.g.
 
// #import module.submodule // access through submodule::function(), or submodule::Type
 
// #import module.submodule as Sub // access through Sub::function(), or Sub::type
 
// #import module.submodule::* // access through function(), or Type
 
// #import module.submodule::{function} // access through function()
 
// #import module.submodule::{function as func, type} // access through func() or type
 

	
 
pragma-import-alias = cwb "as" cwb identifier
 
pragma-import-all = "::*"
 
pragma-import-single-symbol = "::" identifier pragma-import-alias?
 
pragma-import-multi-symbol = "::{" ...
 
    cw identifier pragma-import-alias? ...
 
    (cw "," cw identifier pragma-import-alias?)* ...
 
    (cw ",")? cw "}"
 
pragma-import = "#import" cwb domain-identifier ...
 
    (pragma-import-alias | pragma-import-all | pragma-import-single-symbol | pragma-import-multi-symbol)? 
 

	
 
// Custom pragmas for people which may be using (sometime, somewhere) 
 
// metaprogramming with pragmas
 
pragma-custom = "#" identifier-any (cwb VCHAR (VCHAR | WS)*) cw ";"
 

	
 
// Finally, a pragma may be any of the ones above
 
pragma = pragma-version | pragma-module | pragma-import | pragma-custom
 
```
 

	
 
Note that, different from C-like languages, we do require semicolons to exist at the end of a pragma statement. The reason is to prevent future hacks using the "\" character to indicate an end-of-line-but-not-really-end-of-line statements.
 

	
 
Apart from these pragmas, we can have component definitions, type definitions and function definitions within the source file. The grammar for these may be formulated as:
 

	
 
```
 
// Annotated types and function/component arguments
 
type-annotation = type (cw [])?
 
var-declaration = type-annotation cwb identifier
 
params-list = "(" cw (var-declaration (cw "," cw var-declaration)*)? cw ")"
 

	
 
// Functions and components
 
function-def = type-annotation cwb identifier cw params-list cw block
 
composite-def = "composite" cwb identifier cw params-list cw block
 
primitive-def = "primitive" cwb identifier cw params-list cw block
 
component-def = composite-def | primitive-def
 

	
 
// Symbol definitions now become
 
symbol-def = type-union-def | type-struct-def | function-def | component-def 
 
```
 

	
 
Using these rules, we can now describe the grammar of a single file as:
 

	
 
```
 
file = cw (pragma | symbol-def)* cw
 
```
 

	
 
Of course, we currently cannot do anything useful with our grammar, hence we have to describe blocks to let the functions and component definitions do something. To do so, we proceed as:
 

	
 
```
 
// channel a->b;, or channel a -> b;
 
channel-decl = channel cwb identifier cw "->" cw identifier cw ";"
 
// int a = 5, b = 2 + 3;
 
memory-decl = var-declaration cw "=" cw expression (cw "," cw identifier cw "=" cw expression)* cw ";"
 

	
 
stmt = block |
 
    identifier cw ":" cw stmt | // label
 
    "if" cw pexpr cw stmt (cw "else" cwb stmt)? |
 
    "while" cw pexpr cw stmt |
 
    "break" (cwb identifier)? cw ";" |
 
    "continue" (cwb identifier)? cw ";" |
 
    "synchronous" stmt |
 
    "return" cwb identifier cw ";" |
 
    "goto" cwb identifier cw ";" |
 
    "skip" cw ";" |
 
    "new" cwb method-expr cw ";" |
 
    expr cw ";"
 
    
 
// TODO: Add all the other expressions
 
// TODO: Also: add struct construction and enum construction
 
method-params-list = "(" cw (expr (cw "," cw expr)* )? cw ")"
 
method-expr = method cw method-params-list
 

	
 
enum-destructure-expr = "let" cw ns-identifier "::" identifier cw "(" cw identifier cw ")" cw "=" expr
 
enum-test-expr = ns-identifier "::" identifier cw "==" cw expr
 

	
 
block = "{" (cw (channel-decl | memory-decl | stmt))* cw "}"
 
```
 

	
 
Note that we have a potential collision of various expressions/statements. The following cases are of importance:
 

	
 
1. An empty block is written as `{}`, while an empty array construction is also written as `{}`.
 
2. Both function calls as enum constants feature the same construction syntax. That is: `foo::bar(expression)` may refer to a function call to `bar` in the namespace `foo`, but may also be the construction of enum `foo`'s `bar` variant (containing a value `expression`). These may be disambiguated using the type system.
 
3. The enumeration destructuring expression may collide with the constant enumeration literal. These may be disambiguated by looking at the inner value. If the inner value is an identifier and not yet defined as a variable, then it is a destructuring expression. Otherwise it must be interpreted as a constant enumeration. The enumeration destructuring expression must then be completed by it being a child of an binary equality operator. If not, then it is invalid syntax.
 

	
 
Finally, for consistency, there are additional rules to the enumeration destructuring. As a preamble: the language should allow programmers to express any kind of trickery they want, as long as it is correct. But programmers should be prevented from expressing something that is by definition incorrect/illogical. So enumeration destructuring (e.g. `Enum::Variant(bla) == expression`) should return a value with a special type (e.g. `EnumDestructureBool`) that may only reside within the testing expressions of `if` and `while` statements. Furthermore, this special boolean type only supports the logical-and (`&&`) operator. This way we prevent invalid expressions such as `if (Enum::Variant1(foo) == expr || Enum::Variant2(bar) == expr) { ... }`, but we do allow potentially valid expressions like `if (Enum::Variant1(foo) == expr_foo && Enum::Variant2(bar) == expr_bar) { ... }`.
 

	
 
**Question**: In the documentation for V1.0 we find the `synchronous cw (params-list cw stmt | block)` rule. Why the `params-list`?
 

	
 
**TODO**: Release constructions on memory declarations: as long as we have a write to it before a read we should be fine. Can be done once we add semantic analysis in order to optimize putting and getting port values.
 
**TODO**: Implement type inference, should be simpler once I figure out how to write a typechecker.
 
**TODO**: Add constants assigned in the global scope.
 
**TODO**: Add a runtime expression evaluator (probably before constants in global scope) to simplify expressions and/or remove impossible branches.
 
\ No newline at end of file
src/collections/freelist.rs
Show inline comments
 
use std::marker::PhantomData;
 
use alloc::raw_vec::RawVec;
 

	
 
/// Entry in a freelist. Contains a generation number to ensure silly mistakes
 
/// using an item's index after freeing it.
 
struct Entry<T> {
 
    generation: usize,
 
    item: T,
 
}
 

	
 
/// Key of an item in the freelist. Contains a generation number to prevent
 
/// use-after-free during development.
 
// TODO: Two usizes are probably overkill
 
#[derive(Copy, Clone)]
 
pub struct Key<T> {
 
    generation: usize,
 
    index: usize,
 
    _type: PhantomData<T>,
 
}
 

	
 
/// Generic freelist structure. Item insertion/retrieval/deletion works like a
 
/// HashMap through keys.
 
/// TODO: Use alloc::raw_vec::RawVec once stable and accessible
 
pub struct FreeList<T> {
 
    items: *mut Entry<T>,
 
    capacity: usize,
 
    length: usize,
 
    free: Vec<usize>,
 
}
 

	
 
impl<T> FreeList<T> {
 
    pub fn new() -> Self<T> {
 
        std::alloc::Layout::from_size_align()
 
        Self{
 
            items: std::ptr::null_mut(),
 
            capacity: 0,
 
            length: 0,
 
            free: Vec::new(),
 
        }
 
    }
 

	
 
    pub fn with_capacity(capacity: usize) -> Self {
 
        alloc::
 
        Self{
 
            items: std::,
 
            free: Vec::with_capacity(capacity),
 
            length: 0,
 
        }
 
    }
 

	
 
    /// Inserts a new item into the freelist. Will return a key that can be used
 
    /// to retrieve the item and delete it.
 
    pub fn insert(&mut self, item: T) -> Key<T> {
 
        let mut generation;
 
        let mut index;
 

	
 
        if self.free.is_empty() {
 
            // No free elements, make sure we have enough capacity
 
            if self.length == self.items.capacity() {
 
                self.items.reserve(self.length, 1);
 
            }
 

	
 
            // Now we do
 
            generation = 0;
 
            index = self.length;
 

	
 
            unsafe {
 
                let target = self.items.ptr().add(self.length);
 
                std::ptr::write(&mut target.item, item);
 
                self.length += 1;
 
            }
 
        } else {
 
            // We have a free spot. Note that the generation is incremented upon
 
            // freeing an item. So we can just take the current generation value
 
            // here.
 
            index = self.free.pop().unwrap();
 

	
 
            unsafe {
 
                let target = self.items.ptr().add(self.length);
 
                generation = target.generation;
 
                std::ptr::write(&mut target.item, item);
 
            }
 
        }
 

	
 
        Key { generation, index, _type: PhantomData::default() }
 
    }
 

	
 
    /// Removes the entry using the provided key. Will panic if the element was
 
    /// removed already.
 
    pub fn erase(&mut self, index: Key<T>) {
 
        // This should always be the case
 
        debug_assert!(index.index < self.length);
 

	
 
        // Retrieve element and make sure that the generation matches
 
        unsafe {
 
            let entry = self.items.ptr().add(index.index);
 
            assert_eq!(entry.generation, entry.generation);
 
            *entry.generation += 1;
 
            std::ptr::drop_in_place(&mut entry.item);
 
        }
 

	
 
        // Add the entry to the freelist.
 
        self.free.push(index.index);
 
    }
 
}
 

	
 
impl<T> std::ops::Index<Key<T>> for FreeList<T> {
 
    type Output = T;
 

	
 
    fn index(&self, index: &Key<T>) -> &Self::Output {
 
        debug_assert!(index.index < self.length);
 
        unsafe {
 
            let entry = self.items.ptr().add(index.index);
 
            assert_eq!(entry.generation, index.generation);
 
            return &entry.item;
 
        }
 
    }
 
}
 

	
 
impl<T> std::ops::IndexMut<Key<T>> for FreeList<T> {
 
    fn index_mut(&self, index: &Key<T>) -> &Self::Output {
 
        debug_assert!(index.index < self.length);
 
        unsafe {
 
            let entry = self.items.ptr().add(index.index);
 
            assert_eq!(entry.generation, index.generation);
 
            return &mut entry.item;
 
        }
 
    }
 
}
 

	
 
impl<T> Drop for FreeList<T> {
 
    fn drop(&mut self) {
 
        // Sort free indices to use them while traversing the allocated items
 
        self.free.sort_unstable();
 
        let free_length = self.free.len();
 
        let mut next_free_idx = 1;
 
        let mut next_free_item = usize::MAX;
 
        if free_length != 0 {
 
            next_free_item = self.free[0];
 
        }
 

	
 
        // Go through all items. If we didn't yet drop the item, then we do
 
        // so here.
 
        for item_idx in 0..self.length {
 
            if item_idx == next_free_item {
 
                if next_free_idx < free_length {
 
                    next_free_item = self.free[next_free_idx];
 
                    next_free_idx += 1;
 
                } else {
 
                    next_free_item = usize::MAX;
 
                }
 

	
 
                // Skipped the current item, go to the next one
 
                continue
 
            }
 

	
 
            // Need to deallocate the current item
 
            unsafe {
 
                let entry = self.items.ptr().add(item_idx);
 
                std::ptr::drop_in_place(&mut entry.item);
 
            }
 
        }
 
    }
 
}
 

	
 
#[cfg(test)]
 
mod tests {
 
    use super::*;
 
    use std::sync::Arc;
 
    use std::sync::atomic::{AtomicU32, Ordering};
 

	
 
    struct Counters {
 
        constructed: Arc<AtomicU32>,
 
        destructed: Arc<AtomicU32>,
 
    }
 

	
 
    impl Counters {
 
        pub fn new() -> Self {
 
            Self{
 
                constructed: Arc::new(AtomicU32::new(0)),
 
                destructed: Arc::new(AtomicU32::new(0)),
 
            }
 
        }
 
    }
 

	
 
    struct TestEntity {
 
        counters: Counters,
 
        pub value: u32,
 
    }
 

	
 
    impl TestEntity {
 
        pub fn new(counters: &Counters, value: u32) -> Self {
 
            counters.constructed.fetch_add(1, Ordering::SeqCst);
 
            return TestEntity{
 
                counters: Counters{
 
                    constructed: counters.constructed.clone(),
 
                    destructed: counters.destructed.clone(),
 
                },
 
                value,
 
            }
 
        }
 
    }
 

	
 
    impl Drop for TestEntity {
 
        fn drop(&mut self) {
 
            self.counters.destructed.fetch_add(1, Ordering::SeqCst);
 
        }
 
    }
 

	
 
    #[test]
 
    fn only_constructing() {
 
        const NUM_CREATED: u32 = 3;
 
        let counters = Counters::new();
 
        let mut list = FreeList::new();
 

	
 
        for i in 0..NUM_CREATED {
 
            list.insert(TestEntity::new(&counters, i));
 
            assert_eq!(counters.constructed.load(Ordering::AcqRel), i + 1);
 
        }
 

	
 
        // Everything is constructed, check freelist
 
        assert_eq!(counters.constructed.load(Ordering::AcqRel), NUM_CREATED);
 
        assert_eq!(list.length, 3);
 
        assert!(list.items.capacity() >= 3);
 
        assert!(list.free.is_empty());
 

	
 
        // Drop and check everything is properly dropped
 
        drop(list);
 
        assert_eq!(counters.destructed.load(Ordering::AcqRel), NUM_CREATED)
 
    }
 

	
 
    #[test]
 
    fn reusing_slots() {
 
        const NUM_ROUNDS: u32 = 10;
 
        const NUM_IN_USE: u32 = 10;
 

	
 
        let counters = Counters::new();
 
        let mut list = FreeList::new();
 
        let mut indices = Vec::with_capacity(NUM_IN_USE as usize);
 

	
 
        for round_idx in 0..NUM_ROUNDS {
 
            indices.clear();
 

	
 
            // Adding entries
 
            for i in 0..NUM_IN_USE {
 
                let new_index = list.insert(TestEntity::new(&counters, i));
 
                indices.push(new_index);
 
            }
 

	
 
            // Length should always remain the same as the total number of
 
            // entries in use
 
            assert_eq!(list.length, NUM_IN_USE as usize);
 

	
 
            // Removing entries, and making sure that everything is still
 
            // accessible
 
            for idx in 0..NUM_IN_USE {
 
                // Make sure we can still retrieve the item we're going to delete
 
                let idx_to_remove = NUM_IN_USE - 1 - idx;
 
                let pop_index = indices.pop().unwrap();
 
                let entry = &list[pop_index];
 
                assert_eq!(entry.value, idx_to_remove);
 

	
 
                // Remove the entry and make sure the other ones are still
 
                // accessible
 
                list.erase(pop_index);
 

	
 
                for remaining_idx in 0..idx_to_remove + 1 {
 
                    let remaining_key = &indices[remaining_idx as usize];
 
                    let remaining_entry = &list[*remaining_key];
 
                    assert_eq!(remaining_entry.value, remaining_idx);
 
                }
 
            }
 

	
 
            // Now that we're empty, our constructed and destructed counts
 
            // should match
 
            let expected_count = (round_idx + 1) * NUM_IN_USE;
 
            assert_eq!(counters.constructed.load(Ordering::AcqRel), expected_count);
 
            assert_eq!(counters.destructed.load(Ordering::AcqRel), expected_count);
 
        }
 

	
 
        // Make sure the capacity didn't grow out of bounds, maximum growth rate
 
        // I've ever encountered on `Vec`s is 2. So:
 
        assert!(list.items.capacity() >= NUM_IN_USE as usize && list.items.capacity() < 2*NUM_IN_USE as usize);
 

	
 
        // Finally, when we drop the list we shouldn't be destructing anything
 
        // anymore.
 
        drop(list);
 
        let final_count = NUM_ROUNDS * NUM_IN_USE;
 
        assert_eq!(counters.constructed.load(Ordering::AcqRel), final_count);
 
        assert_eq!(counters.destructed.load(Ordering::AcqRel), final_count);
 
    }
 

	
 
    #[test]
 
    #[should_panic]
 
    fn panic_on_reused_key_of_empty_slot() {
 
        let counters = Counters::new();
 
        let mut list = FreeList::new();
 
        let key = list.insert(TestEntity::new(&counters, 0));
 
        list.erase(key);
 
        let entry = &list[key];
 
    }
 

	
 
    #[test]
 
    #[should_panic]
 
    fn panic_on_reused_key_of_used_slot() {
 
        let counters = Counters::new();
 
        let mut list = FreeList::new();
 
        let key1 = list.insert(TestEntity::new(&counters, 0));
 
        list.erase(key1);
 
        let key2 = list.insert(TestEntity::new(&counters, 0));
 
        assert_eq!(key1.index, key2.index);
 
        assert_ne!(key1.generation, key2.generation);
 
        let entry = &list[key1];
 
    }
 
}
 
\ No newline at end of file
src/collections/mod.rs
Show inline comments
 
mod string_pool;
 
mod scoped_buffer;
 
mod sets;
 
mod raw_vec;
 

	
 
// TODO: Finish this later, use alloc::alloc and alloc::Layout
 
// mod freelist;
 

	
 
pub(crate) use string_pool::{StringPool, StringRef};
 
pub(crate) use scoped_buffer::{ScopedBuffer, ScopedSection};
 
pub(crate) use sets::{DequeSet, VecSet};
 
pub(crate) use raw_vec::RawVec;
 
\ No newline at end of file
src/collections/string_pool.rs
Show inline comments
 
use std::ptr::null_mut;
 
use std::hash::{Hash, Hasher};
 
use std::marker::PhantomData;
 
use std::fmt::{Debug, Display, Result as FmtResult};
 
use crate::common::Formatter;
 

	
 
const SLAB_SIZE: usize = u16::MAX as usize;
 

	
 
#[derive(Clone)]
 
pub struct StringRef<'a> {
 
    data: *const u8,
 
    length: usize,
 
    _phantom: PhantomData<&'a [u8]>,
 
}
 

	
 
// As the StringRef is an immutable thing:
 
unsafe impl Sync for StringRef<'_> {}
 
unsafe impl Send for StringRef<'_> {}
 

	
 
impl<'a> StringRef<'a> {
 
    /// `new` constructs a new StringRef whose data is not owned by the
 
    /// `StringPool`, hence cannot have a `'static` lifetime.
 
    pub(crate) fn new(data: &'a [u8]) -> StringRef<'a> {
 
        // This is an internal (compiler) function: so debug_assert that the
 
        // string is valid ascii. Most commonly the input will come from the
 
        // code's source file, which is checked for ASCII-ness anyway.
 
        debug_assert!(data.is_ascii());
 
        let length = data.len();
 
        let data = data.as_ptr();
 
        StringRef{ data, length, _phantom: PhantomData }
 
    }
 

	
 
    pub fn as_str(&self) -> &'a str {
 
        unsafe {
 
            let slice = std::slice::from_raw_parts::<'a, u8>(self.data, self.length);
 
            std::str::from_utf8_unchecked(slice)
 
        }
 
    }
 

	
 
    pub fn as_bytes(&self) -> &'a [u8] {
 
        unsafe {
 
            std::slice::from_raw_parts::<'a, u8>(self.data, self.length)
 
        }
 
    }
 
}
 

	
 
impl<'a> Debug for StringRef<'a> {
 
    fn fmt(&self, f: &mut Formatter<'_>) -> FmtResult {
 
        f.write_str("StringRef{ value: ")?;
 
        f.write_str(self.as_str())?;
 
        f.write_str(" }")
 
    }
 
}
 

	
 
impl<'a> Display for StringRef<'a> {
 
    fn fmt(&self, f: &mut Formatter<'_>) -> FmtResult {
 
        f.write_str(self.as_str())
 
    }
 
}
 

	
 
impl PartialEq for StringRef<'_> {
 
    fn eq(&self, other: &StringRef) -> bool {
 
        self.as_str() == other.as_str()
 
    }
 
}
 

	
 
impl Eq for StringRef<'_> {}
 

	
 
impl Hash for StringRef<'_> {
 
    fn hash<H: Hasher>(&self, state: &mut H) {
 
        state.write(self.as_bytes());
 
    }
 
}
 

	
 
struct StringPoolSlab {
 
    prev: *mut StringPoolSlab,
 
    data: Vec<u8>,
 
    remaining: usize,
 
}
 

	
 
impl StringPoolSlab {
 
    fn new(prev: *mut StringPoolSlab) -> Self {
 
        Self{ prev, data: Vec::with_capacity(SLAB_SIZE), remaining: SLAB_SIZE }
 
    }
 
}
 

	
 
/// StringPool is a ever-growing pool of strings. Strings have a maximum size
 
/// equal to the slab size. The slabs are essentially a linked list to maintain
 
/// pointer-stability of the strings themselves.
 
/// All `StringRef` instances are invalidated when the string pool is dropped
 
pub(crate) struct StringPool {
 
    last: *mut StringPoolSlab,
 
}
 

	
 
impl StringPool {
 
    pub(crate) fn new() -> Self {
 
        // To have some stability we just turn a box into a raw ptr.
 
        let initial_slab = Box::new(StringPoolSlab::new(null_mut()));
 
        let initial_slab = Box::into_raw(initial_slab);
 
        StringPool{
 
            last: initial_slab,
 
        }
 
    }
 

	
 
    /// Interns a string to the `StringPool`, returning a reference to it. The
 
    /// pointer owned by `StringRef` is `'static` as the `StringPool` doesn't
 
    /// reallocate/deallocate until dropped (which only happens at the end of
 
    /// the program.)
 
    pub(crate) fn intern(&mut self, data: &[u8]) -> StringRef<'static> {
 
        // TODO: Large string allocations, if ever needed.
 
        let data_len = data.len();
 
        assert!(data_len <= SLAB_SIZE, "string is too large for slab");
 
        assert!(data_len <= SLAB_SIZE, "string is too large for slab"); // if you hit this, create logic for large-string allocations
 
        debug_assert!(std::str::from_utf8(data).is_ok(), "string to intern is not valid UTF-8 encoded");
 
        
 
        let mut last = unsafe{&mut *self.last};
 
        if data.len() > last.remaining {
 
            // Doesn't fit: allocate new slab
 
            self.alloc_new_slab();
 
            last = unsafe{&mut *self.last};
 
        }
 

	
 
        // Must fit now, compute hash and put in buffer
 
        debug_assert!(data_len <= last.remaining);
 
        let range_start = last.data.len();
 
        last.data.extend_from_slice(data);
 
        last.remaining -= data_len;
 
        debug_assert_eq!(range_start + data_len, last.data.len());
 

	
 
        unsafe {
 
            let start = last.data.as_ptr().offset(range_start as isize);
 
            StringRef{ data: start, length: data_len, _phantom: PhantomData }
 
        }
 
    }
 

	
 
    fn alloc_new_slab(&mut self) {
 
        let new_slab = Box::new(StringPoolSlab::new(self.last));
 
        let new_slab = Box::into_raw(new_slab);
 
        self.last = new_slab;
 
    }
 
}
 

	
 
impl Drop for StringPool {
 
    fn drop(&mut self) {
 
        let mut new_slab = self.last;
 
        while !new_slab.is_null() {
 
            let cur_slab = new_slab;
 
            unsafe {
 
                new_slab = (*cur_slab).prev;
 
                Box::from_raw(cur_slab); // consume and deallocate
 
            }
 
        }
 
    }
 
}
 

	
 
// String pool cannot be cloned, and the created `StringRef` instances remain
 
// allocated until the end of the program, so it is always safe to send. It is
 
// also sync in the sense that it becomes an immutable thing after compilation,
 
// but lets not derive that if we would ever become a multithreaded compiler in
 
// the future.
 
unsafe impl Send for StringPool {}
 

	
 
#[cfg(test)]
 
mod tests {
 
    use super::*;
 

	
 
    #[test]
 
    fn test_string_just_fits() {
 
        let large = "0".repeat(SLAB_SIZE);
 
        let mut pool = StringPool::new();
 
        let interned = pool.intern(large.as_bytes());
 
        assert_eq!(interned.as_str(), large);
 
    }
 

	
 
    #[test]
 
    #[should_panic]
 
    fn test_string_too_large() {
 
        let large = "0".repeat(SLAB_SIZE + 1);
 
        let mut pool = StringPool::new();
 
        let _interned = pool.intern(large.as_bytes());
 
    }
 

	
 
    #[test]
 
    fn test_lots_of_small_allocations() {
 
        const NUM_PER_SLAB: usize = 32;
 
        const NUM_SLABS: usize = 4;
 

	
 
        let to_intern = "0".repeat(SLAB_SIZE / NUM_PER_SLAB);
 
        let mut pool = StringPool::new();
 

	
 
        let mut last_slab = pool.last;
 
        let mut all_refs = Vec::new();
 

	
 
        // Fill up first slab
 
        for _alloc_idx in 0..NUM_PER_SLAB {
 
            let interned = pool.intern(to_intern.as_bytes());
 
            all_refs.push(interned);
 
            assert!(std::ptr::eq(last_slab, pool.last));
 
        }
 

	
 
        for _slab_idx in 0..NUM_SLABS-1 {
 
            for alloc_idx in 0..NUM_PER_SLAB {
 
                let interned = pool.intern(to_intern.as_bytes());
 
                all_refs.push(interned);
 

	
 
                if alloc_idx == 0 {
 
                    // First allocation produces a new slab
 
                    assert!(!std::ptr::eq(last_slab, pool.last));
 
                    last_slab = pool.last;
 
                } else {
 
                    assert!(std::ptr::eq(last_slab, pool.last));
 
                }
 
            }
 
        }
 

	
 
        // All strings are still correct
 
        for string_ref in all_refs {
 
            assert_eq!(string_ref.as_str(), to_intern);
 
        }
 
    }
 
}
 
\ No newline at end of file
src/protocol/ast.rs
Show inline comments
 
// TODO: @cleanup, rigorous cleanup of dead code and silly object-oriented
 
//  trait impls where I deem them unfit.
 

	
 
use std::fmt;
 
use std::fmt::{Debug, Display, Formatter};
 
use std::ops::{Index, IndexMut};
 

	
 
use super::arena::{Arena, Id};
 
use crate::collections::StringRef;
 
use crate::protocol::input_source::InputSpan;
 

	
 
/// Helper macro that defines a type alias for a AST element ID. In this case 
 
/// only used to alias the `Id<T>` types.
 
macro_rules! define_aliased_ast_id {
 
    // Variant where we just defined the alias, without any indexing
 
    ($name:ident, $parent:ty) => {
 
        pub type $name = $parent;
 
    };
 
    // Variant where we define the type, and the Index and IndexMut traits
 
    (
 
        $name:ident, $parent:ty, 
 
        index($indexed_type:ty, $indexed_arena:ident)
 
    ) => {
 
        define_aliased_ast_id!($name, $parent);
 
        impl Index<$name> for Heap {
 
            type Output = $indexed_type;
 
            fn index(&self, index: $name) -> &Self::Output {
 
                &self.$indexed_arena[index]
 
            }
 
        }
 

	
 
        impl IndexMut<$name> for Heap {
 
            fn index_mut(&mut self, index: $name) -> &mut Self::Output {
 
                &mut self.$indexed_arena[index]
 
            }
 
        }
 
    };
 
    // Variant where we define type, Index(Mut) traits and an allocation function
 
    (
 
        $name:ident, $parent:ty,
 
        index($indexed_type:ty, $indexed_arena:ident),
 
        alloc($fn_name:ident)
 
    ) => {
 
        define_aliased_ast_id!($name, $parent, index($indexed_type, $indexed_arena));
 
        impl Heap {
 
            pub fn $fn_name(&mut self, f: impl FnOnce($name) -> $indexed_type) -> $name {
 
                self.$indexed_arena.alloc_with_id(|id| f(id))
 
            }
 
        }
 
    };
 
}
 

	
 
/// Helper macro that defines a wrapper type for a particular variant of an AST
 
/// element ID. Only used to define single-wrapping IDs.
 
macro_rules! define_new_ast_id {
 
    // Variant where we just defined the new type, without any indexing
 
    ($name:ident, $parent:ty) => {
 
        #[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
 
        pub struct $name (pub(crate) $parent);
 

	
 
        #[allow(dead_code)]
 
        impl $name {
 
            pub(crate) fn new_invalid() -> Self     { Self(<$parent>::new_invalid()) }
 
            pub(crate) fn is_invalid(&self) -> bool { self.0.is_invalid() }
 
            pub fn upcast(self) -> $parent          { self.0 }
 
        }
 
    };
 
    // Variant where we define the type, and the Index and IndexMut traits
 
    (
 
        $name:ident, $parent:ty, 
 
        index($indexed_type:ty, $wrapper_type:path, $indexed_arena:ident)
 
    ) => {
 
        define_new_ast_id!($name, $parent);
 
        impl Index<$name> for Heap {
 
            type Output = $indexed_type;
 
            fn index(&self, index: $name) -> &Self::Output {
 
                if let $wrapper_type(v) = &self.$indexed_arena[index.0] {
 
                    v
 
                } else {
 
                    unreachable!()
 
                }
 
            }
 
        }
 

	
 
        impl IndexMut<$name> for Heap {
 
            fn index_mut(&mut self, index: $name) -> &mut Self::Output {
 
                if let $wrapper_type(v) = &mut self.$indexed_arena[index.0] {
 
                    v
 
                } else {
 
                    unreachable!()
 
                }
 
            }
 
        }
 
    };
 
    // Variant where we define the type, the Index and IndexMut traits, and an allocation function
 
    (
 
        $name:ident, $parent:ty, 
 
        index($indexed_type:ty, $wrapper_type:path, $indexed_arena:ident),
 
        alloc($fn_name:ident)
 
    ) => {
 
        define_new_ast_id!($name, $parent, index($indexed_type, $wrapper_type, $indexed_arena));
 
        impl Heap {
 
            pub fn $fn_name(&mut self, f: impl FnOnce($name) -> $indexed_type) -> $name {
 
                $name(
 
                    self.$indexed_arena.alloc_with_id(|id| {
 
                        $wrapper_type(f($name(id)))
 
                    })
 
                )
 
            }
 
        }
 
    }
 
}
 

	
 
define_aliased_ast_id!(RootId, Id<Root>, index(Root, protocol_descriptions), alloc(alloc_protocol_description));
 
define_aliased_ast_id!(PragmaId, Id<Pragma>, index(Pragma, pragmas), alloc(alloc_pragma));
 
define_aliased_ast_id!(ImportId, Id<Import>, index(Import, imports), alloc(alloc_import));
 
define_aliased_ast_id!(VariableId, Id<Variable>, index(Variable, variables), alloc(alloc_variable));
 

	
 
define_aliased_ast_id!(DefinitionId, Id<Definition>, index(Definition, definitions));
 
define_new_ast_id!(StructDefinitionId, DefinitionId, index(StructDefinition, Definition::Struct, definitions), alloc(alloc_struct_definition));
 
define_new_ast_id!(EnumDefinitionId, DefinitionId, index(EnumDefinition, Definition::Enum, definitions), alloc(alloc_enum_definition));
 
define_new_ast_id!(UnionDefinitionId, DefinitionId, index(UnionDefinition, Definition::Union, definitions), alloc(alloc_union_definition));
 
define_new_ast_id!(ComponentDefinitionId, DefinitionId, index(ComponentDefinition, Definition::Component, definitions), alloc(alloc_component_definition));
 
define_new_ast_id!(FunctionDefinitionId, DefinitionId, index(FunctionDefinition, Definition::Function, definitions), alloc(alloc_function_definition));
 

	
 
define_aliased_ast_id!(StatementId, Id<Statement>, index(Statement, statements));
 
define_new_ast_id!(BlockStatementId, StatementId, index(BlockStatement, Statement::Block, statements), alloc(alloc_block_statement));
 
define_new_ast_id!(EndBlockStatementId, StatementId, index(EndBlockStatement, Statement::EndBlock, statements), alloc(alloc_end_block_statement));
 
define_new_ast_id!(LocalStatementId, StatementId, index(LocalStatement, Statement::Local, statements), alloc(alloc_local_statement));
 
define_new_ast_id!(MemoryStatementId, LocalStatementId);
 
define_new_ast_id!(ChannelStatementId, LocalStatementId);
 
define_new_ast_id!(LabeledStatementId, StatementId, index(LabeledStatement, Statement::Labeled, statements), alloc(alloc_labeled_statement));
 
define_new_ast_id!(IfStatementId, StatementId, index(IfStatement, Statement::If, statements), alloc(alloc_if_statement));
 
define_new_ast_id!(EndIfStatementId, StatementId, index(EndIfStatement, Statement::EndIf, statements), alloc(alloc_end_if_statement));
 
define_new_ast_id!(WhileStatementId, StatementId, index(WhileStatement, Statement::While, statements), alloc(alloc_while_statement));
 
define_new_ast_id!(EndWhileStatementId, StatementId, index(EndWhileStatement, Statement::EndWhile, statements), alloc(alloc_end_while_statement));
 
define_new_ast_id!(BreakStatementId, StatementId, index(BreakStatement, Statement::Break, statements), alloc(alloc_break_statement));
 
define_new_ast_id!(ContinueStatementId, StatementId, index(ContinueStatement, Statement::Continue, statements), alloc(alloc_continue_statement));
 
define_new_ast_id!(SynchronousStatementId, StatementId, index(SynchronousStatement, Statement::Synchronous, statements), alloc(alloc_synchronous_statement));
 
define_new_ast_id!(EndSynchronousStatementId, StatementId, index(EndSynchronousStatement, Statement::EndSynchronous, statements), alloc(alloc_end_synchronous_statement));
 
define_new_ast_id!(ForkStatementId, StatementId, index(ForkStatement, Statement::Fork, statements), alloc(alloc_fork_statement));
 
define_new_ast_id!(EndForkStatementId, StatementId, index(EndForkStatement, Statement::EndFork, statements), alloc(alloc_end_fork_statement));
 
define_new_ast_id!(ReturnStatementId, StatementId, index(ReturnStatement, Statement::Return, statements), alloc(alloc_return_statement));
 
define_new_ast_id!(GotoStatementId, StatementId, index(GotoStatement, Statement::Goto, statements), alloc(alloc_goto_statement));
 
define_new_ast_id!(NewStatementId, StatementId, index(NewStatement, Statement::New, statements), alloc(alloc_new_statement));
 
define_new_ast_id!(ExpressionStatementId, StatementId, index(ExpressionStatement, Statement::Expression, statements), alloc(alloc_expression_statement));
 

	
 
define_aliased_ast_id!(ExpressionId, Id<Expression>, index(Expression, expressions));
 
define_new_ast_id!(AssignmentExpressionId, ExpressionId, index(AssignmentExpression, Expression::Assignment, expressions), alloc(alloc_assignment_expression));
 
define_new_ast_id!(BindingExpressionId, ExpressionId, index(BindingExpression, Expression::Binding, expressions), alloc(alloc_binding_expression));
 
define_new_ast_id!(ConditionalExpressionId, ExpressionId, index(ConditionalExpression, Expression::Conditional, expressions), alloc(alloc_conditional_expression));
 
define_new_ast_id!(BinaryExpressionId, ExpressionId, index(BinaryExpression, Expression::Binary, expressions), alloc(alloc_binary_expression));
 
define_new_ast_id!(UnaryExpressionId, ExpressionId, index(UnaryExpression, Expression::Unary, expressions), alloc(alloc_unary_expression));
 
define_new_ast_id!(IndexingExpressionId, ExpressionId, index(IndexingExpression, Expression::Indexing, expressions), alloc(alloc_indexing_expression));
 
define_new_ast_id!(SlicingExpressionId, ExpressionId, index(SlicingExpression, Expression::Slicing, expressions), alloc(alloc_slicing_expression));
 
define_new_ast_id!(SelectExpressionId, ExpressionId, index(SelectExpression, Expression::Select, expressions), alloc(alloc_select_expression));
 
define_new_ast_id!(LiteralExpressionId, ExpressionId, index(LiteralExpression, Expression::Literal, expressions), alloc(alloc_literal_expression));
 
define_new_ast_id!(CastExpressionId, ExpressionId, index(CastExpression, Expression::Cast, expressions), alloc(alloc_cast_expression));
 
define_new_ast_id!(CallExpressionId, ExpressionId, index(CallExpression, Expression::Call, expressions), alloc(alloc_call_expression));
 
define_new_ast_id!(VariableExpressionId, ExpressionId, index(VariableExpression, Expression::Variable, expressions), alloc(alloc_variable_expression));
 

	
 
#[derive(Debug)]
 
pub struct Heap {
 
    // Root arena, contains the entry point for different modules. Each root
 
    // contains lists of IDs that correspond to the other arenas.
 
    pub(crate) protocol_descriptions: Arena<Root>,
 
    // Contents of a file, these are the elements the `Root` elements refer to
 
    pragmas: Arena<Pragma>,
 
    pub(crate) imports: Arena<Import>,
 
    pub(crate) variables: Arena<Variable>,
 
    pub(crate) definitions: Arena<Definition>,
 
    pub(crate) statements: Arena<Statement>,
 
    pub(crate) expressions: Arena<Expression>,
 
}
 

	
 
impl Heap {
 
    pub fn new() -> Heap {
 
        Heap {
 
            // string_alloc: StringAllocator::new(),
 
            protocol_descriptions: Arena::new(),
 
            pragmas: Arena::new(),
 
            imports: Arena::new(),
 
            variables: Arena::new(),
 
            definitions: Arena::new(),
 
            statements: Arena::new(),
 
            expressions: Arena::new(),
 
        }
 
    }
 
    pub fn alloc_memory_statement(
 
        &mut self,
 
        f: impl FnOnce(MemoryStatementId) -> MemoryStatement,
 
    ) -> MemoryStatementId {
 
        MemoryStatementId(LocalStatementId(self.statements.alloc_with_id(|id| {
 
            Statement::Local(LocalStatement::Memory(
 
                f(MemoryStatementId(LocalStatementId(id)))
 
            ))
 
        })))
 
    }
 
    pub fn alloc_channel_statement(
 
        &mut self,
 
        f: impl FnOnce(ChannelStatementId) -> ChannelStatement,
 
    ) -> ChannelStatementId {
 
        ChannelStatementId(LocalStatementId(self.statements.alloc_with_id(|id| {
 
            Statement::Local(LocalStatement::Channel(
 
                f(ChannelStatementId(LocalStatementId(id)))
 
            ))
 
        })))
 
    }
 
}
 

	
 
impl Index<MemoryStatementId> for Heap {
 
    type Output = MemoryStatement;
 
    fn index(&self, index: MemoryStatementId) -> &Self::Output {
 
        &self.statements[index.0.0].as_memory()
 
    }
 
}
 

	
 
impl Index<ChannelStatementId> for Heap {
 
    type Output = ChannelStatement;
 
    fn index(&self, index: ChannelStatementId) -> &Self::Output {
 
        &self.statements[index.0.0].as_channel()
 
    }
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct Root {
 
    pub this: RootId,
 
    // Phase 1: parser
 
    // pub position: InputPosition,
 
    pub pragmas: Vec<PragmaId>,
 
    pub imports: Vec<ImportId>,
 
    pub definitions: Vec<DefinitionId>,
 
}
 

	
 
impl Root {
 
    pub fn get_definition_ident(&self, h: &Heap, id: &[u8]) -> Option<DefinitionId> {
 
        for &def in self.definitions.iter() {
 
            if h[def].identifier().value.as_bytes() == id {
 
                return Some(def);
 
            }
 
        }
 
        None
 
    }
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub enum Pragma {
 
    Version(PragmaVersion),
 
    Module(PragmaModule),
 
}
 

	
 
impl Pragma {
 
    pub(crate) fn as_module(&self) -> &PragmaModule {
 
        match self {
 
            Pragma::Module(pragma) => pragma,
 
            _ => unreachable!("Tried to obtain {:?} as PragmaModule", self),
 
        }
 
    }
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct PragmaVersion {
 
    pub this: PragmaId,
 
    // Phase 1: parser
 
    pub span: InputSpan, // of full pragma
 
    pub version: u64,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct PragmaModule {
 
    pub this: PragmaId,
 
    // Phase 1: parser
 
    pub span: InputSpan, // of full pragma
 
    pub value: Identifier,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub enum Import {
 
    Module(ImportModule),
 
    Symbols(ImportSymbols)
 
}
 

	
 
impl Import {
 
    pub(crate) fn span(&self) -> InputSpan {
 
        match self {
 
            Import::Module(v) => v.span,
 
            Import::Symbols(v) => v.span,
 
        }
 
    }
 

	
 
    pub(crate) fn as_module(&self) -> &ImportModule {
 
        match self {
 
            Import::Module(m) => m,
 
            _ => unreachable!("Unable to cast 'Import' to 'ImportModule'")
 
        }
 
    }
 
    pub(crate) fn as_symbols(&self) -> &ImportSymbols {
 
        match self {
 
            Import::Symbols(m) => m,
 
            _ => unreachable!("Unable to cast 'Import' to 'ImportSymbols'")
 
        }
 
    }
 
    pub(crate) fn as_symbols_mut(&mut self) -> &mut ImportSymbols {
 
        match self {
 
            Import::Symbols(m) => m,
 
            _ => unreachable!("Unable to cast 'Import' to 'ImportSymbols'")
 
        }
 
    }
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct ImportModule {
 
    pub this: ImportId,
 
    // Phase 1: parser
 
    pub span: InputSpan,
 
    pub module: Identifier,
 
    pub alias: Identifier,
 
    pub module_id: RootId,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct AliasedSymbol {
 
    pub name: Identifier,
 
    pub alias: Option<Identifier>,
 
    pub definition_id: DefinitionId,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct ImportSymbols {
 
    pub this: ImportId,
 
    // Phase 1: parser
 
    pub span: InputSpan,
 
    pub module: Identifier,
 
    pub module_id: RootId,
 
    pub symbols: Vec<AliasedSymbol>,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct Identifier {
 
    pub span: InputSpan,
 
    pub value: StringRef<'static>,
 
}
 

	
 
impl PartialEq for Identifier {
 
    fn eq(&self, other: &Self) -> bool {
 
        return self.value == other.value
 
    }
 
}
 

	
 
impl Display for Identifier {
 
    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
 
        write!(f, "{}", self.value.as_str())
 
    }
 
}
 

	
 
#[derive(Debug, Clone, PartialEq, Eq)]
 
pub enum ParserTypeVariant {
 
    // Special builtin, only usable by the compiler and not constructable by the
 
    // programmer
 
    Void,
 
    InputOrOutput,
 
    ArrayLike,
 
    IntegerLike,
 
    // Basic builtin
 
    Message,
 
    Bool,
 
    UInt8, UInt16, UInt32, UInt64,
 
    SInt8, SInt16, SInt32, SInt64,
 
    Character, String,
 
    // Literals (need to get concrete builtin type during typechecking)
 
    IntegerLiteral,
 
    // Marker for inference
 
    Inferred,
 
    // Builtins expecting one subsequent type
 
    Array,
 
    Input,
 
    Output,
 
    // User-defined types
 
    PolymorphicArgument(DefinitionId, u32), // u32 = index into polymorphic variables
 
    Definition(DefinitionId, u32), // u32 = number of subsequent types in the type tree.
 
}
 

	
 
impl ParserTypeVariant {
 
    pub(crate) fn num_embedded(&self) -> usize {
 
        use ParserTypeVariant::*;
 

	
 
        match self {
 
            Void | IntegerLike |
 
            Message | Bool |
 
            UInt8 | UInt16 | UInt32 | UInt64 |
 
            SInt8 | SInt16 | SInt32 | SInt64 |
 
            Character | String | IntegerLiteral |
 
            Inferred | PolymorphicArgument(_, _) =>
 
                0,
 
            ArrayLike | InputOrOutput | Array | Input | Output =>
 
                1,
 
            Definition(_, num) => *num as usize,
 
        }
 
    }
 
}
 

	
 
/// ParserTypeElement is an element of the type tree. An element may be
 
/// implicit, meaning that the user didn't specify the type, but it was set by
 
/// the compiler.
 
#[derive(Debug, Clone)]
 
pub struct ParserTypeElement {
 
    // TODO: @Fix span
 
    pub element_span: InputSpan, // span of this element, not including the child types
 
    pub variant: ParserTypeVariant,
 
}
 

	
 
/// ParserType is a specification of a type during the parsing phase and initial
 
/// linker/validator phase of the compilation process. These types may be
 
/// (partially) inferred or represent literals (e.g. a integer whose bytesize is
 
/// not yet determined).
 
///
 
/// Its contents are the depth-first serialization of the type tree. Each node
 
/// is a type that may accept polymorphic arguments. The polymorphic arguments
 
/// are then the children of the node.
 
#[derive(Debug, Clone)]
 
pub struct ParserType {
 
    pub elements: Vec<ParserTypeElement>,
 
    pub full_span: InputSpan,
 
}
 

	
 
impl ParserType {
 
    pub(crate) fn iter_embedded(&self, parent_idx: usize) -> ParserTypeIter {
 
        ParserTypeIter::new(&self.elements, parent_idx)
 
    }
 
}
 

	
 
/// Iterator over the embedded elements of a specific element.
 
pub struct ParserTypeIter<'a> {
 
    pub elements: &'a [ParserTypeElement],
 
    pub cur_embedded_idx: usize,
 
}
 

	
 
impl<'a> ParserTypeIter<'a> {
 
    fn new(elements: &'a [ParserTypeElement], parent_idx: usize) -> Self {
 
        debug_assert!(parent_idx < elements.len(), "parent index exceeds number of elements in ParserType");
 
        if elements[0].variant.num_embedded() == 0 {
 
            // Parent element does not have any embedded types, place
 
            // `cur_embedded_idx` at end so we will always return `None`
 
            Self{ elements, cur_embedded_idx: elements.len() }
 
        } else {
 
            // Parent element has an embedded type
 
            Self{ elements, cur_embedded_idx: parent_idx + 1 }
 
        }
 
    }
 
}
 

	
 
impl<'a> Iterator for ParserTypeIter<'a> {
 
    type Item = &'a [ParserTypeElement];
 

	
 
    fn next(&mut self) -> Option<Self::Item> {
 
        let elements_len = self.elements.len();
 
        if self.cur_embedded_idx >= elements_len {
 
            return None;
 
        }
 

	
 
        // Seek to the end of the subtree
 
        let mut depth = 1;
 
        let start_element = self.cur_embedded_idx;
 
        while self.cur_embedded_idx < elements_len {
 
            let cur_element = &self.elements[self.cur_embedded_idx];
 
            let depth_change = cur_element.variant.num_embedded() as i32 - 1;
 
            depth += depth_change;
 
            debug_assert!(depth >= 0, "illegally constructed ParserType: {:?}", self.elements);
 

	
 
            self.cur_embedded_idx += 1;
 
            if depth == 0 {
 
                break;
 
            }
 
        }
 

	
 
        debug_assert!(depth == 0, "illegally constructed ParserType: {:?}", self.elements);
 
        return Some(&self.elements[start_element..self.cur_embedded_idx]);
 
    }
 
}
 

	
 
/// ConcreteType is the representation of a type after the type inference and
 
/// checker is finished. These are fully typed.
 
#[derive(Debug, Clone, Copy, Eq, PartialEq)]
 
pub enum ConcreteTypePart {
 
    // Special types (cannot be explicitly constructed by the programmer)
 
    Void,
 
    // Builtin types without nested types
 
    Message,
 
    Bool,
 
    UInt8, UInt16, UInt32, UInt64,
 
    SInt8, SInt16, SInt32, SInt64,
 
    Character, String,
 
    // Builtin types with one nested type
 
    Array,
 
    Slice,
 
    Input,
 
    Output,
 
    // User defined type with any number of nested types
 
    Instance(DefinitionId, u32),    // instance of data type
 
    Function(DefinitionId, u32),    // instance of function
 
    Component(DefinitionId, u32),   // instance of a connector
 
}
 

	
 
impl ConcreteTypePart {
 
    fn num_embedded(&self) -> u32 {
 
        use ConcreteTypePart::*;
 

	
 
        match self {
 
            Void | Message | Bool |
 
            UInt8 | UInt16 | UInt32 | UInt64 |
 
            SInt8 | SInt16 | SInt32 | SInt64 |
 
            Character | String =>
 
                0,
 
            Array | Slice | Input | Output =>
 
                1,
 
            Instance(_, num_embedded) => *num_embedded,
 
            Function(_, num_embedded) => *num_embedded,
 
            Component(_, num_embedded) => *num_embedded,
 
        }
 
    }
 
}
 

	
 
#[derive(Debug, Clone, Eq, PartialEq)]
 
pub struct ConcreteType {
 
    pub(crate) parts: Vec<ConcreteTypePart>
 
}
 

	
 
impl Default for ConcreteType {
 
    fn default() -> Self {
 
        Self{ parts: Vec::new() }
 
    }
 
}
 

	
 
impl ConcreteType {
 
    /// Returns an iterator over the subtrees that are type arguments (e.g. an
 
    /// array element's type, or a polymorphic type's arguments) to the
 
    /// provided parent type (specified by its index in the `parts` array).
 
    pub(crate) fn embedded_iter<'a>(&'a self, parent_part_idx: usize) -> ConcreteTypeIter<'a> {
 
        let num_embedded = self.parts[parent_part_idx].num_embedded();
 
        return ConcreteTypeIter{
 
            concrete: self,
 
            idx_embedded: 0,
 
            num_embedded,
 
            part_idx: parent_part_idx + 1,
 
        }
 
    }
 

	
 
    /// Given the starting position of a type tree, determine the exclusive
 
    /// ending index.
 
    pub(crate) fn subtree_end_idx(&self, start_idx: usize) -> usize {
 
        let mut depth = 1;
 
        let num_parts = self.parts.len();
 
        debug_assert!(start_idx < num_parts);
 

	
 
        for part_idx in start_idx..self.parts.len() {
 
            let depth_change = self.parts[part_idx].num_embedded() as i32 - 1;
 
            depth += depth_change;
 
            debug_assert!(depth >= 0);
 

	
 
            if depth == 0 {
 
                return part_idx + 1;
 
            }
 
        }
 

	
 
        debug_assert!(false, "incorrectly constructed ConcreteType instance");
 
        return 0;
 
    }
 

	
 
    /// Construct a human-readable name for the type. Because this performs
 
    /// a string allocation don't use it for anything else then displaying the
 
    /// type to the user.
 
    pub(crate) fn display_name(&self, heap: &Heap) -> String {
 
        fn display_part(parts: &[ConcreteTypePart], heap: &Heap, mut idx: usize, target: &mut String) -> usize {
 
            use ConcreteTypePart as CTP;
 
            use crate::protocol::parser::token_parsing::*;
 

	
 
            let cur_idx = idx;
 
            idx += 1; // increment by 1, because it always happens
 

	
 
            match parts[cur_idx] {
 
                CTP::Void => { target.push_str("void"); },
 
                CTP::Message => { target.push_str(KW_TYPE_MESSAGE_STR); },
 
                CTP::Bool => { target.push_str(KW_TYPE_BOOL_STR); },
 
                CTP::UInt8 => { target.push_str(KW_TYPE_UINT8_STR); },
 
                CTP::UInt16 => { target.push_str(KW_TYPE_UINT16_STR); },
 
                CTP::UInt32 => { target.push_str(KW_TYPE_UINT32_STR); },
 
                CTP::UInt64 => { target.push_str(KW_TYPE_UINT64_STR); },
 
                CTP::SInt8 => { target.push_str(KW_TYPE_SINT8_STR); },
 
                CTP::SInt16 => { target.push_str(KW_TYPE_SINT16_STR); },
 
                CTP::SInt32 => { target.push_str(KW_TYPE_SINT32_STR); },
 
                CTP::SInt64 => { target.push_str(KW_TYPE_SINT64_STR); },
 
                CTP::Character => { target.push_str(KW_TYPE_CHAR_STR); },
 
                CTP::String => { target.push_str(KW_TYPE_STRING_STR); },
 
                CTP::Array | CTP::Slice => {
 
                    idx = display_part(parts, heap, idx, target);
 
                    target.push_str("[]");
 
                },
 
                CTP::Input => {
 
                    target.push_str(KW_TYPE_IN_PORT_STR);
 
                    target.push('<');
 
                    idx = display_part(parts, heap, idx, target);
 
                    target.push('>');
 
                },
 
                CTP::Output => {
 
                    target.push_str(KW_TYPE_OUT_PORT_STR);
 
                    target.push('<');
 
                    idx = display_part(parts, heap, idx, target);
 
                    target.push('>');
 
                },
 
                CTP::Instance(definition_id, num_poly_args) |
 
                CTP::Function(definition_id, num_poly_args) |
 
                CTP::Component(definition_id, num_poly_args) => {
 
                    let definition = &heap[definition_id];
 
                    target.push_str(definition.identifier().value.as_str());
 

	
 
                    if num_poly_args != 0 {
 
                        target.push('<');
 
                        for poly_arg_idx in 0..num_poly_args {
 
                            if poly_arg_idx != 0 {
 
                                target.push(',');
 
                                idx = display_part(parts, heap, idx, target);
 
                            }
 
                        }
 
                        target.push('>');
 
                    }
 
                }
 
            }
 

	
 
            idx
 
        }
 

	
 
        let mut name = String::with_capacity(128);
 
        let _final_idx = display_part(&self.parts, heap, 0, &mut name);
 
        debug_assert_eq!(_final_idx, self.parts.len());
 

	
 
        return name;
 
    }
 
}
 

	
 
#[derive(Debug)]
 
pub struct ConcreteTypeIter<'a> {
 
    concrete: &'a ConcreteType,
 
    idx_embedded: u32,
 
    num_embedded: u32,
 
    part_idx: usize,
 
}
 

	
 
impl<'a> Iterator for ConcreteTypeIter<'a> {
 
    type Item = &'a [ConcreteTypePart];
 

	
 
    fn next(&mut self) -> Option<Self::Item> {
 
        if self.idx_embedded == self.num_embedded {
 
            return None;
 
        }
 

	
 
        // Retrieve the subtree of interest
 
        let start_idx = self.part_idx;
 
        let end_idx = self.concrete.subtree_end_idx(start_idx);
 

	
 
        self.idx_embedded += 1;
 
        self.part_idx = end_idx;
 

	
 
        return Some(&self.concrete.parts[start_idx..end_idx]);
 
    }
 
}
 

	
 
#[derive(Debug, Clone, Copy)]
 
pub enum Scope {
 
    Definition(DefinitionId),
 
    Regular(BlockStatementId),
 
    Synchronous((SynchronousStatementId, BlockStatementId)),
 
}
 

	
 
impl Scope {
 
    pub fn is_block(&self) -> bool {
 
        match &self {
 
            Scope::Definition(_) => false,
 
            Scope::Regular(_) => true,
 
            Scope::Synchronous(_) => true,
 
        }
 
    }
 
    pub fn to_block(&self) -> BlockStatementId {
 
        match &self {
 
            Scope::Regular(id) => *id,
 
            Scope::Synchronous((_, id)) => *id,
 
            _ => panic!("unable to get BlockStatement from Scope")
 
        }
 
    }
 
}
 

	
 
/// `ScopeNode` is a helper that links scopes in two directions. It doesn't
 
/// actually contain any information associated with the scope, this may be
 
/// found on the AST elements that `Scope` points to.
 
#[derive(Debug, Clone)]
 
pub struct ScopeNode {
 
    pub parent: Scope,
 
    pub nested: Vec<Scope>,
 
}
 

	
 
impl ScopeNode {
 
    pub(crate) fn new_invalid() -> Self {
 
        ScopeNode{
 
            parent: Scope::Definition(DefinitionId::new_invalid()),
 
            nested: Vec::new(),
 
        }
 
    }
 
}
 

	
 
#[derive(Debug, Clone, PartialEq, Eq)]
 
pub enum VariableKind {
 
    Parameter,      // in parameter list of function/component
 
    Local,          // declared in function/component body
 
    Binding,        // may be bound to in a binding expression (determined in validator/linker)
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct Variable {
 
    pub this: VariableId,
 
    // Parsing
 
    pub kind: VariableKind,
 
    pub parser_type: ParserType,
 
    pub identifier: Identifier,
 
    // Validator/linker
 
    pub relative_pos_in_block: u32,
 
    pub unique_id_in_scope: i32, // Temporary fix until proper bytecode/asm is generated
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub enum Definition {
 
    Struct(StructDefinition),
 
    Enum(EnumDefinition),
 
    Union(UnionDefinition),
 
    Component(ComponentDefinition),
 
    Function(FunctionDefinition),
 
}
 

	
 
impl Definition {
 
    pub fn is_struct(&self) -> bool {
 
        match self {
 
            Definition::Struct(_) => true,
 
            _ => false
 
        }
 
    }
 
    pub(crate) fn as_struct(&self) -> &StructDefinition {
 
        match self {
 
            Definition::Struct(result) => result,
 
            _ => panic!("Unable to cast 'Definition' to 'StructDefinition'"),
 
        }
 
    }
 
    pub(crate) fn as_struct_mut(&mut self) -> &mut StructDefinition {
 
        match self {
 
            Definition::Struct(result) => result,
 
            _ => panic!("Unable to cast 'Definition' to 'StructDefinition'"),
 
        }
 
    }
 
    pub fn is_enum(&self) -> bool {
 
        match self {
 
            Definition::Enum(_) => true,
 
            _ => false,
 
        }
 
    }
 
    pub(crate) fn as_enum(&self) -> &EnumDefinition {
 
        match self {
 
            Definition::Enum(result) => result,
 
            _ => panic!("Unable to cast 'Definition' to 'EnumDefinition'"),
 
        }
 
    }
 
    pub(crate) fn as_enum_mut(&mut self) -> &mut EnumDefinition {
 
        match self {
 
            Definition::Enum(result) => result,
 
            _ => panic!("Unable to cast 'Definition' to 'EnumDefinition'"),
 
        }
 
    }
 
    pub fn is_union(&self) -> bool {
 
        match self {
 
            Definition::Union(_) => true,
 
            _ => false,
 
        }
 
    }
 
    pub(crate) fn as_union(&self) -> &UnionDefinition {
 
        match self {
 
            Definition::Union(result) => result, 
 
            _ => panic!("Unable to cast 'Definition' to 'UnionDefinition'"),
 
        }
 
    }
 
    pub(crate) fn as_union_mut(&mut self) -> &mut UnionDefinition {
 
        match self {
 
            Definition::Union(result) => result,
 
            _ => panic!("Unable to cast 'Definition' to 'UnionDefinition'"),
 
        }
 
    }
 
@@ -1052,737 +1048,736 @@ impl Statement {
 
            _ => panic!("Unable to cast `Statement` to `BlockStatement`"),
 
        }
 
    }
 
    pub fn as_local(&self) -> &LocalStatement {
 
        match self {
 
            Statement::Local(result) => result,
 
            _ => panic!("Unable to cast `Statement` to `LocalStatement`"),
 
        }
 
    }
 
    pub fn as_memory(&self) -> &MemoryStatement {
 
        self.as_local().as_memory()
 
    }
 
    pub fn as_channel(&self) -> &ChannelStatement {
 
        self.as_local().as_channel()
 
    }
 

	
 
    pub fn as_new(&self) -> &NewStatement {
 
        match self {
 
            Statement::New(result) => result,
 
            _ => panic!("Unable to cast `Statement` to `NewStatement`"),
 
        }
 
    }
 

	
 
    pub fn span(&self) -> InputSpan {
 
        match self {
 
            Statement::Block(v) => v.span,
 
            Statement::Local(v) => v.span(),
 
            Statement::Labeled(v) => v.label.span,
 
            Statement::If(v) => v.span,
 
            Statement::While(v) => v.span,
 
            Statement::Break(v) => v.span,
 
            Statement::Continue(v) => v.span,
 
            Statement::Synchronous(v) => v.span,
 
            Statement::Fork(v) => v.span,
 
            Statement::Return(v) => v.span,
 
            Statement::Goto(v) => v.span,
 
            Statement::New(v) => v.span,
 
            Statement::Expression(v) => v.span,
 
            Statement::EndBlock(_) | Statement::EndIf(_) | Statement::EndWhile(_) | Statement::EndSynchronous(_) | Statement::EndFork(_) => unreachable!(),
 
        }
 
    }
 
    pub fn link_next(&mut self, next: StatementId) {
 
        match self {
 
            Statement::Block(stmt) => stmt.next = next,
 
            Statement::EndBlock(stmt) => stmt.next = next,
 
            Statement::Local(stmt) => match stmt {
 
                LocalStatement::Channel(stmt) => stmt.next = next,
 
                LocalStatement::Memory(stmt) => stmt.next = next,
 
            },
 
            Statement::EndIf(stmt) => stmt.next = next,
 
            Statement::EndWhile(stmt) => stmt.next = next,
 
            Statement::EndSynchronous(stmt) => stmt.next = next,
 
            Statement::EndFork(stmt) => stmt.next = next,
 
            Statement::New(stmt) => stmt.next = next,
 
            Statement::Expression(stmt) => stmt.next = next,
 
            Statement::Return(_)
 
            | Statement::Break(_)
 
            | Statement::Continue(_)
 
            | Statement::Synchronous(_)
 
            | Statement::Fork(_)
 
            | Statement::Goto(_)
 
            | Statement::While(_)
 
            | Statement::Labeled(_)
 
            | Statement::If(_) => unreachable!(),
 
        }
 
    }
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct BlockStatement {
 
    pub this: BlockStatementId,
 
    // Phase 1: parser
 
    pub is_implicit: bool,
 
    pub span: InputSpan, // of the complete block
 
    pub statements: Vec<StatementId>,
 
    pub end_block: EndBlockStatementId,
 
    // Phase 2: linker
 
    pub scope_node: ScopeNode,
 
    pub first_unique_id_in_scope: i32, // Temporary fix until proper bytecode/asm is generated
 
    pub next_unique_id_in_scope: i32, // Temporary fix until proper bytecode/asm is generated
 
    pub relative_pos_in_parent: u32,
 
    pub locals: Vec<VariableId>,
 
    pub labels: Vec<LabeledStatementId>,
 
    pub next: StatementId,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct EndBlockStatement {
 
    pub this: EndBlockStatementId,
 
    // Parser
 
    pub start_block: BlockStatementId,
 
    // Validation/Linking
 
    pub next: StatementId,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub enum LocalStatement {
 
    Memory(MemoryStatement),
 
    Channel(ChannelStatement),
 
}
 

	
 
impl LocalStatement {
 
    pub fn this(&self) -> LocalStatementId {
 
        match self {
 
            LocalStatement::Memory(stmt) => stmt.this.upcast(),
 
            LocalStatement::Channel(stmt) => stmt.this.upcast(),
 
        }
 
    }
 
    pub fn as_memory(&self) -> &MemoryStatement {
 
        match self {
 
            LocalStatement::Memory(result) => result,
 
            _ => panic!("Unable to cast `LocalStatement` to `MemoryStatement`"),
 
        }
 
    }
 
    pub fn as_channel(&self) -> &ChannelStatement {
 
        match self {
 
            LocalStatement::Channel(result) => result,
 
            _ => panic!("Unable to cast `LocalStatement` to `ChannelStatement`"),
 
        }
 
    }
 
    pub fn span(&self) -> InputSpan {
 
        match self {
 
            LocalStatement::Channel(v) => v.span,
 
            LocalStatement::Memory(v) => v.span,
 
        }
 
    }
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct MemoryStatement {
 
    pub this: MemoryStatementId,
 
    // Phase 1: parser
 
    pub span: InputSpan,
 
    pub variable: VariableId,
 
    // Phase 2: linker
 
    pub next: StatementId,
 
}
 

	
 
/// ChannelStatement is the declaration of an input and output port associated
 
/// with the same channel. Note that the polarity of the ports are from the
 
/// point of view of the component. So an output port is something that a
 
/// component uses to send data over (i.e. it is the "input end" of the
 
/// channel), and vice versa.
 
#[derive(Debug, Clone)]
 
pub struct ChannelStatement {
 
    pub this: ChannelStatementId,
 
    // Phase 1: parser
 
    pub span: InputSpan, // of the "channel" keyword
 
    pub from: VariableId, // output
 
    pub to: VariableId,   // input
 
    // Phase 2: linker
 
    pub relative_pos_in_block: u32,
 
    pub next: StatementId,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct LabeledStatement {
 
    pub this: LabeledStatementId,
 
    // Phase 1: parser
 
    pub label: Identifier,
 
    pub body: StatementId,
 
    // Phase 2: linker
 
    pub relative_pos_in_block: u32,
 
    pub in_sync: SynchronousStatementId, // may be invalid
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct IfStatement {
 
    pub this: IfStatementId,
 
    // Phase 1: parser
 
    pub span: InputSpan, // of the "if" keyword
 
    pub test: ExpressionId,
 
    pub true_body: BlockStatementId,
 
    pub false_body: Option<BlockStatementId>,
 
    pub end_if: EndIfStatementId,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct EndIfStatement {
 
    pub this: EndIfStatementId,
 
    pub start_if: IfStatementId,
 
    pub next: StatementId,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct WhileStatement {
 
    pub this: WhileStatementId,
 
    // Phase 1: parser
 
    pub span: InputSpan, // of the "while" keyword
 
    pub test: ExpressionId,
 
    pub body: BlockStatementId,
 
    pub end_while: EndWhileStatementId,
 
    pub in_sync: SynchronousStatementId, // may be invalid
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct EndWhileStatement {
 
    pub this: EndWhileStatementId,
 
    pub start_while: WhileStatementId,
 
    // Phase 2: linker
 
    pub next: StatementId,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct BreakStatement {
 
    pub this: BreakStatementId,
 
    // Phase 1: parser
 
    pub span: InputSpan, // of the "break" keyword
 
    pub label: Option<Identifier>,
 
    // Phase 2: linker
 
    pub target: Option<EndWhileStatementId>,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct ContinueStatement {
 
    pub this: ContinueStatementId,
 
    // Phase 1: parser
 
    pub span: InputSpan, // of the "continue" keyword
 
    pub label: Option<Identifier>,
 
    // Phase 2: linker
 
    pub target: Option<WhileStatementId>,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct SynchronousStatement {
 
    pub this: SynchronousStatementId,
 
    // Phase 1: parser
 
    pub span: InputSpan, // of the "sync" keyword
 
    pub body: BlockStatementId,
 
    pub end_sync: EndSynchronousStatementId,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct EndSynchronousStatement {
 
    pub this: EndSynchronousStatementId,
 
    pub start_sync: SynchronousStatementId,
 
    // Phase 2: linker
 
    pub next: StatementId,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct ForkStatement {
 
    pub this: ForkStatementId,
 
    // Phase 1: parser
 
    pub span: InputSpan, // of the "fork" keyword
 
    pub left_body: BlockStatementId,
 
    pub right_body: Option<BlockStatementId>,
 
    pub end_fork: EndForkStatementId,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct EndForkStatement {
 
    pub this: EndForkStatementId,
 
    pub start_fork: ForkStatementId,
 
    pub next: StatementId,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct ReturnStatement {
 
    pub this: ReturnStatementId,
 
    // Phase 1: parser
 
    pub span: InputSpan, // of the "return" keyword
 
    pub expressions: Vec<ExpressionId>,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct GotoStatement {
 
    pub this: GotoStatementId,
 
    // Phase 1: parser
 
    pub span: InputSpan, // of the "goto" keyword
 
    pub label: Identifier,
 
    // Phase 2: linker
 
    pub target: Option<LabeledStatementId>,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct NewStatement {
 
    pub this: NewStatementId,
 
    // Phase 1: parser
 
    pub span: InputSpan, // of the "new" keyword
 
    pub expression: CallExpressionId,
 
    // Phase 2: linker
 
    pub next: StatementId,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct ExpressionStatement {
 
    pub this: ExpressionStatementId,
 
    // Phase 1: parser
 
    pub span: InputSpan,
 
    pub expression: ExpressionId,
 
    // Phase 2: linker
 
    pub next: StatementId,
 
}
 

	
 
#[derive(Debug, PartialEq, Eq, Clone, Copy)]
 
pub enum ExpressionParent {
 
    None, // only set during initial parsing
 
    If(IfStatementId),
 
    While(WhileStatementId),
 
    Return(ReturnStatementId),
 
    New(NewStatementId),
 
    ExpressionStmt(ExpressionStatementId),
 
    Expression(ExpressionId, u32) // index within expression (e.g LHS or RHS of expression)
 
}
 

	
 
impl ExpressionParent {
 
    pub fn is_new(&self) -> bool {
 
        match self {
 
            ExpressionParent::New(_) => true,
 
            _ => false,
 
        }
 
    }
 

	
 
    pub fn as_expression(&self) -> ExpressionId {
 
        match self {
 
            ExpressionParent::Expression(id, _) => *id,
 
            _ => panic!("called as_expression() on {:?}", self),
 
        }
 
    }
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub enum Expression {
 
    Assignment(AssignmentExpression),
 
    Binding(BindingExpression),
 
    Conditional(ConditionalExpression),
 
    Binary(BinaryExpression),
 
    Unary(UnaryExpression),
 
    Indexing(IndexingExpression),
 
    Slicing(SlicingExpression),
 
    Select(SelectExpression),
 
    Literal(LiteralExpression),
 
    Cast(CastExpression),
 
    Call(CallExpression),
 
    Variable(VariableExpression),
 
}
 

	
 
impl Expression {
 
    pub fn as_variable(&self) -> &VariableExpression {
 
        match self {
 
            Expression::Variable(result) => result,
 
            _ => panic!("Unable to cast `Expression` to `VariableExpression`"),
 
        }
 
    }
 

	
 
    /// Returns operator span, function name, a binding's "let" span, etc. An
 
    /// indicator for the kind of expression that is being applied.
 
    pub fn operation_span(&self) -> InputSpan {
 
        match self {
 
            Expression::Assignment(expr) => expr.operator_span,
 
            Expression::Binding(expr) => expr.operator_span,
 
            Expression::Conditional(expr) => expr.operator_span,
 
            Expression::Binary(expr) => expr.operator_span,
 
            Expression::Unary(expr) => expr.operator_span,
 
            Expression::Indexing(expr) => expr.operator_span,
 
            Expression::Slicing(expr) => expr.slicing_span,
 
            Expression::Select(expr) => expr.operator_span,
 
            Expression::Literal(expr) => expr.span,
 
            Expression::Cast(expr) => expr.cast_span,
 
            Expression::Call(expr) => expr.func_span,
 
            Expression::Variable(expr) => expr.identifier.span,
 
        }
 
    }
 

	
 
    /// Returns the span covering the entire expression (i.e. including the
 
    /// spans of the arguments as well).
 
    pub fn full_span(&self) -> InputSpan {
 
        match self {
 
            Expression::Assignment(expr) => expr.full_span,
 
            Expression::Binding(expr) => expr.full_span,
 
            Expression::Conditional(expr) => expr.full_span,
 
            Expression::Binary(expr) => expr.full_span,
 
            Expression::Unary(expr) => expr.full_span,
 
            Expression::Indexing(expr) => expr.full_span,
 
            Expression::Slicing(expr) => expr.full_span,
 
            Expression::Select(expr) => expr.full_span,
 
            Expression::Literal(expr) => expr.span,
 
            Expression::Cast(expr) => expr.full_span,
 
            Expression::Call(expr) => expr.full_span,
 
            Expression::Variable(expr) => expr.identifier.span,
 
        }
 
    }
 

	
 
    // TODO: @cleanup
 
    pub fn parent(&self) -> &ExpressionParent {
 
        match self {
 
            Expression::Assignment(expr) => &expr.parent,
 
            Expression::Binding(expr) => &expr.parent,
 
            Expression::Conditional(expr) => &expr.parent,
 
            Expression::Binary(expr) => &expr.parent,
 
            Expression::Unary(expr) => &expr.parent,
 
            Expression::Indexing(expr) => &expr.parent,
 
            Expression::Slicing(expr) => &expr.parent,
 
            Expression::Select(expr) => &expr.parent,
 
            Expression::Literal(expr) => &expr.parent,
 
            Expression::Cast(expr) => &expr.parent,
 
            Expression::Call(expr) => &expr.parent,
 
            Expression::Variable(expr) => &expr.parent,
 
        }
 
    }
 
    // TODO: @cleanup
 

	
 
    pub fn parent_expr_id(&self) -> Option<ExpressionId> {
 
        if let ExpressionParent::Expression(id, _) = self.parent() {
 
            Some(*id)
 
        } else {
 
            None
 
        }
 
    }
 

	
 
    pub fn get_unique_id_in_definition(&self) -> i32 {
 
        match self {
 
            Expression::Assignment(expr) => expr.unique_id_in_definition,
 
            Expression::Binding(expr) => expr.unique_id_in_definition,
 
            Expression::Conditional(expr) => expr.unique_id_in_definition,
 
            Expression::Binary(expr) => expr.unique_id_in_definition,
 
            Expression::Unary(expr) => expr.unique_id_in_definition,
 
            Expression::Indexing(expr) => expr.unique_id_in_definition,
 
            Expression::Slicing(expr) => expr.unique_id_in_definition,
 
            Expression::Select(expr) => expr.unique_id_in_definition,
 
            Expression::Literal(expr) => expr.unique_id_in_definition,
 
            Expression::Cast(expr) => expr.unique_id_in_definition,
 
            Expression::Call(expr) => expr.unique_id_in_definition,
 
            Expression::Variable(expr) => expr.unique_id_in_definition,
 
        }
 
    }
 
}
 

	
 
#[derive(Debug, Clone, Copy)]
 
pub enum AssignmentOperator {
 
    Set,
 
    Concatenated,
 
    Multiplied,
 
    Divided,
 
    Remained,
 
    Added,
 
    Subtracted,
 
    ShiftedLeft,
 
    ShiftedRight,
 
    BitwiseAnded,
 
    BitwiseXored,
 
    BitwiseOred,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct AssignmentExpression {
 
    pub this: AssignmentExpressionId,
 
    // Parsing
 
    pub operator_span: InputSpan,
 
    pub full_span: InputSpan,
 
    pub left: ExpressionId,
 
    pub operation: AssignmentOperator,
 
    pub right: ExpressionId,
 
    // Validator/Linker
 
    pub parent: ExpressionParent,
 
    pub unique_id_in_definition: i32,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct BindingExpression {
 
    pub this: BindingExpressionId,
 
    // Parsing
 
    pub operator_span: InputSpan,
 
    pub full_span: InputSpan,
 
    pub bound_to: ExpressionId,
 
    pub bound_from: ExpressionId,
 
    // Validator/Linker
 
    pub parent: ExpressionParent,
 
    pub unique_id_in_definition: i32,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct ConditionalExpression {
 
    pub this: ConditionalExpressionId,
 
    // Parsing
 
    pub operator_span: InputSpan,
 
    pub full_span: InputSpan,
 
    pub test: ExpressionId,
 
    pub true_expression: ExpressionId,
 
    pub false_expression: ExpressionId,
 
    // Validator/Linking
 
    pub parent: ExpressionParent,
 
    pub unique_id_in_definition: i32,
 
}
 

	
 
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
 
pub enum BinaryOperator {
 
    Concatenate,
 
    LogicalOr,
 
    LogicalAnd,
 
    BitwiseOr,
 
    BitwiseXor,
 
    BitwiseAnd,
 
    Equality,
 
    Inequality,
 
    LessThan,
 
    GreaterThan,
 
    LessThanEqual,
 
    GreaterThanEqual,
 
    ShiftLeft,
 
    ShiftRight,
 
    Add,
 
    Subtract,
 
    Multiply,
 
    Divide,
 
    Remainder,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct BinaryExpression {
 
    pub this: BinaryExpressionId,
 
    // Parsing
 
    pub operator_span: InputSpan,
 
    pub full_span: InputSpan,
 
    pub left: ExpressionId,
 
    pub operation: BinaryOperator,
 
    pub right: ExpressionId,
 
    // Validator/Linker
 
    pub parent: ExpressionParent,
 
    pub unique_id_in_definition: i32,
 
}
 

	
 
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
 
pub enum UnaryOperator {
 
    Positive,
 
    Negative,
 
    BitwiseNot,
 
    LogicalNot,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct UnaryExpression {
 
    pub this: UnaryExpressionId,
 
    // Parsing
 
    pub operator_span: InputSpan,
 
    pub full_span: InputSpan,
 
    pub operation: UnaryOperator,
 
    pub expression: ExpressionId,
 
    // Validator/Linker
 
    pub parent: ExpressionParent,
 
    pub unique_id_in_definition: i32,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct IndexingExpression {
 
    pub this: IndexingExpressionId,
 
    // Parsing
 
    pub operator_span: InputSpan,
 
    pub full_span: InputSpan,
 
    pub subject: ExpressionId,
 
    pub index: ExpressionId,
 
    // Validator/Linker
 
    pub parent: ExpressionParent,
 
    pub unique_id_in_definition: i32,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct SlicingExpression {
 
    pub this: SlicingExpressionId,
 
    // Parsing
 
    pub slicing_span: InputSpan, // from '[' to ']'
 
    pub full_span: InputSpan, // includes subject
 
    pub subject: ExpressionId,
 
    pub from_index: ExpressionId,
 
    pub to_index: ExpressionId,
 
    // Validator/Linker
 
    pub parent: ExpressionParent,
 
    pub unique_id_in_definition: i32,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct SelectExpression {
 
    pub this: SelectExpressionId,
 
    // Parsing
 
    pub operator_span: InputSpan, // of the '.'
 
    pub full_span: InputSpan, // includes subject and field
 
    pub subject: ExpressionId,
 
    pub field_name: Identifier,
 
    // Validator/Linker
 
    pub parent: ExpressionParent,
 
    pub unique_id_in_definition: i32,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct CastExpression {
 
    pub this: CastExpressionId,
 
    // Parsing
 
    pub cast_span: InputSpan, // of the "cast" keyword,
 
    pub full_span: InputSpan, // includes the cast subject
 
    pub to_type: ParserType,
 
    pub subject: ExpressionId,
 
    // Validator/linker
 
    pub parent: ExpressionParent,
 
    pub unique_id_in_definition: i32,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct CallExpression {
 
    pub this: CallExpressionId,
 
    // Parsing
 
    pub func_span: InputSpan, // of the function name
 
    pub full_span: InputSpan, // includes the arguments and parentheses
 
    pub parser_type: ParserType, // of the function call, not the return type
 
    pub method: Method,
 
    pub arguments: Vec<ExpressionId>,
 
    pub definition: DefinitionId,
 
    // Validator/Linker
 
    pub parent: ExpressionParent,
 
    pub unique_id_in_definition: i32,
 
}
 

	
 
#[derive(Debug, Clone, PartialEq, Eq)]
 
pub enum Method {
 
    // Builtin
 
    Get,
 
    Put,
 
    Fires,
 
    Create,
 
    Length,
 
    Assert,
 
    Print,
 
    UserFunction,
 
    UserComponent,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct MethodSymbolic {
 
    pub(crate) parser_type: ParserType,
 
    pub(crate) definition: DefinitionId
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct LiteralExpression {
 
    pub this: LiteralExpressionId,
 
    // Parsing
 
    pub span: InputSpan,
 
    pub value: Literal,
 
    // Validator/Linker
 
    pub parent: ExpressionParent,
 
    pub unique_id_in_definition: i32,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub enum Literal {
 
    Null, // message
 
    True,
 
    False,
 
    Character(char),
 
    String(StringRef<'static>),
 
    Integer(LiteralInteger),
 
    Struct(LiteralStruct),
 
    Enum(LiteralEnum),
 
    Union(LiteralUnion),
 
    Array(Vec<ExpressionId>),
 
}
 

	
 
impl Literal {
 
    pub(crate) fn as_struct(&self) -> &LiteralStruct {
 
        if let Literal::Struct(literal) = self{
 
            literal
 
        } else {
 
            unreachable!("Attempted to obtain {:?} as Literal::Struct", self)
 
        }
 
    }
 

	
 
    pub(crate) fn as_enum(&self) -> &LiteralEnum {
 
        if let Literal::Enum(literal) = self {
 
            literal
 
        } else {
 
            unreachable!("Attempted to obtain {:?} as Literal::Enum", self)
 
        }
 
    }
 

	
 
    pub(crate) fn as_union(&self) -> &LiteralUnion {
 
        if let Literal::Union(literal) = self {
 
            literal
 
        } else {
 
            unreachable!("Attempted to obtain {:?} as Literal::Union", self)
 
        }
 
    }
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct LiteralInteger {
 
    pub(crate) unsigned_value: u64,
 
    pub(crate) negated: bool, // for constant expression evaluation, TODO: @Int
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct LiteralStructField {
 
    // Phase 1: parser
 
    pub(crate) identifier: Identifier,
 
    pub(crate) value: ExpressionId,
 
    // Phase 2: linker
 
    pub(crate) field_idx: usize, // in struct definition
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct LiteralStruct {
 
    // Phase 1: parser
 
    pub(crate) parser_type: ParserType,
 
    pub(crate) fields: Vec<LiteralStructField>,
 
    pub(crate) definition: DefinitionId,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct LiteralEnum {
 
    // Phase 1: parser
 
    pub(crate) parser_type: ParserType,
 
    pub(crate) variant: Identifier,
 
    pub(crate) definition: DefinitionId,
 
    // Phase 2: linker
 
    pub(crate) variant_idx: usize, // as present in the type table
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct LiteralUnion {
 
    // Phase 1: parser
 
    pub(crate) parser_type: ParserType,
 
    pub(crate) variant: Identifier,
 
    pub(crate) values: Vec<ExpressionId>,
 
    pub(crate) definition: DefinitionId,
 
    // Phase 2: linker
 
    pub(crate) variant_idx: usize, // as present in type table
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub struct VariableExpression {
 
    pub this: VariableExpressionId,
 
    // Parsing
 
    pub identifier: Identifier,
 
    // Validator/Linker
 
    pub declaration: Option<VariableId>,
 
    pub used_as_binding_target: bool,
 
    pub parent: ExpressionParent,
 
    pub unique_id_in_definition: i32,
 
}
 
\ No newline at end of file
src/protocol/ast_printer.rs
Show inline comments
 
@@ -506,460 +506,459 @@ impl ASTWriter {
 
                self.kv(indent2).with_s_key("EndSync").with_disp_val(&stmt.end_sync.0.index);
 
                self.kv(indent2).with_s_key("Body");
 
                self.write_stmt(heap, stmt.body.upcast(), indent3);
 
            },
 
            Statement::EndSynchronous(stmt) => {
 
                self.kv(indent).with_id(PREFIX_ENDSYNC_STMT_ID, stmt.this.0.index)
 
                    .with_s_key("EndSynchronous");
 
                self.kv(indent2).with_s_key("StartSync").with_disp_val(&stmt.start_sync.0.index);
 
                self.kv(indent2).with_s_key("Next").with_disp_val(&stmt.next.index);
 
            },
 
            Statement::Fork(stmt) => {
 
                self.kv(indent).with_id(PREFIX_FORK_STMT_ID, stmt.this.0.index)
 
                    .with_s_key("Fork");
 
                self.kv(indent2).with_s_key("EndFork").with_disp_val(&stmt.end_fork.0.index);
 
                self.kv(indent2).with_s_key("LeftBody");
 
                self.write_stmt(heap, stmt.left_body.upcast(), indent3);
 

	
 
                if let Some(right_body_id) = stmt.right_body {
 
                    self.kv(indent2).with_s_key("RightBody");
 
                    self.write_stmt(heap, right_body_id.upcast(), indent3);
 
                }
 
            },
 
            Statement::EndFork(stmt) => {
 
                self.kv(indent).with_id(PREFIX_END_FORK_STMT_ID, stmt.this.0.index)
 
                    .with_s_key("EndFork");
 
                self.kv(indent2).with_s_key("StartFork").with_disp_val(&stmt.start_fork.0.index);
 
                self.kv(indent2).with_s_key("Next").with_disp_val(&stmt.next.index);
 
            }
 
            Statement::Return(stmt) => {
 
                self.kv(indent).with_id(PREFIX_RETURN_STMT_ID, stmt.this.0.index)
 
                    .with_s_key("Return");
 
                self.kv(indent2).with_s_key("Expressions");
 
                for expr_id in &stmt.expressions {
 
                    self.write_expr(heap, *expr_id, indent3);
 
                }
 
            },
 
            Statement::Goto(stmt) => {
 
                self.kv(indent).with_id(PREFIX_GOTO_STMT_ID, stmt.this.0.index)
 
                    .with_s_key("Goto");
 
                self.kv(indent2).with_s_key("Label").with_identifier_val(&stmt.label);
 
                self.kv(indent2).with_s_key("Target")
 
                    .with_opt_disp_val(stmt.target.as_ref().map(|v| &v.0.index));
 
            },
 
            Statement::New(stmt) => {
 
                self.kv(indent).with_id(PREFIX_NEW_STMT_ID, stmt.this.0.index)
 
                    .with_s_key("New");
 
                self.kv(indent2).with_s_key("Expression");
 
                self.write_expr(heap, stmt.expression.upcast(), indent3);
 
                self.kv(indent2).with_s_key("Next").with_disp_val(&stmt.next.index);
 
            },
 
            Statement::Expression(stmt) => {
 
                self.kv(indent).with_id(PREFIX_EXPR_STMT_ID, stmt.this.0.index)
 
                    .with_s_key("ExpressionStatement");
 
                self.write_expr(heap, stmt.expression, indent2);
 
                self.kv(indent2).with_s_key("Next").with_disp_val(&stmt.next.index);
 
            }
 
        }
 
    }
 

	
 
    fn write_expr(&mut self, heap: &Heap, expr_id: ExpressionId, indent: usize) {
 
        let expr = &heap[expr_id];
 
        let indent2 = indent + 1;
 
        let indent3 = indent2 + 1;
 

	
 
        match expr {
 
            Expression::Assignment(expr) => {
 
                self.kv(indent).with_id(PREFIX_ASSIGNMENT_EXPR_ID, expr.this.0.index)
 
                    .with_s_key("AssignmentExpr");
 
                self.kv(indent2).with_s_key("Operation").with_debug_val(&expr.operation);
 
                self.kv(indent2).with_s_key("Left");
 
                self.write_expr(heap, expr.left, indent3);
 
                self.kv(indent2).with_s_key("Right");
 
                self.write_expr(heap, expr.right, indent3);
 
                self.kv(indent2).with_s_key("Parent")
 
                    .with_custom_val(|v| write_expression_parent(v, &expr.parent));
 
            },
 
            Expression::Binding(expr) => {
 
                self.kv(indent).with_id(PREFIX_BINARY_EXPR_ID, expr.this.0.index)
 
                    .with_s_key("BindingExpr");
 
                self.kv(indent2).with_s_key("BindToExpression");
 
                self.write_expr(heap, expr.bound_to, indent3);
 
                self.kv(indent2).with_s_key("BindFromExpression");
 
                self.write_expr(heap, expr.bound_from, indent3);
 
                self.kv(indent2).with_s_key("Parent")
 
                    .with_custom_val(|v| write_expression_parent(v, &expr.parent));
 
            },
 
            Expression::Conditional(expr) => {
 
                self.kv(indent).with_id(PREFIX_CONDITIONAL_EXPR_ID, expr.this.0.index)
 
                    .with_s_key("ConditionalExpr");
 
                self.kv(indent2).with_s_key("Condition");
 
                self.write_expr(heap, expr.test, indent3);
 
                self.kv(indent2).with_s_key("TrueExpression");
 
                self.write_expr(heap, expr.true_expression, indent3);
 
                self.kv(indent2).with_s_key("FalseExpression");
 
                self.write_expr(heap, expr.false_expression, indent3);
 
                self.kv(indent2).with_s_key("Parent")
 
                    .with_custom_val(|v| write_expression_parent(v, &expr.parent));
 
            },
 
            Expression::Binary(expr) => {
 
                self.kv(indent).with_id(PREFIX_BINARY_EXPR_ID, expr.this.0.index)
 
                    .with_s_key("BinaryExpr");
 
                self.kv(indent2).with_s_key("Operation").with_debug_val(&expr.operation);
 
                self.kv(indent2).with_s_key("Left");
 
                self.write_expr(heap, expr.left, indent3);
 
                self.kv(indent2).with_s_key("Right");
 
                self.write_expr(heap, expr.right, indent3);
 
                self.kv(indent2).with_s_key("Parent")
 
                    .with_custom_val(|v| write_expression_parent(v, &expr.parent));
 
            },
 
            Expression::Unary(expr) => {
 
                self.kv(indent).with_id(PREFIX_UNARY_EXPR_ID, expr.this.0.index)
 
                    .with_s_key("UnaryExpr");
 
                self.kv(indent2).with_s_key("Operation").with_debug_val(&expr.operation);
 
                self.kv(indent2).with_s_key("Argument");
 
                self.write_expr(heap, expr.expression, indent3);
 
                self.kv(indent2).with_s_key("Parent")
 
                    .with_custom_val(|v| write_expression_parent(v, &expr.parent));
 
            },
 
            Expression::Indexing(expr) => {
 
                self.kv(indent).with_id(PREFIX_INDEXING_EXPR_ID, expr.this.0.index)
 
                    .with_s_key("IndexingExpr");
 
                self.kv(indent2).with_s_key("Subject");
 
                self.write_expr(heap, expr.subject, indent3);
 
                self.kv(indent2).with_s_key("Index");
 
                self.write_expr(heap, expr.index, indent3);
 
                self.kv(indent2).with_s_key("Parent")
 
                    .with_custom_val(|v| write_expression_parent(v, &expr.parent));
 
            },
 
            Expression::Slicing(expr) => {
 
                self.kv(indent).with_id(PREFIX_SLICING_EXPR_ID, expr.this.0.index)
 
                    .with_s_key("SlicingExpr");
 
                self.kv(indent2).with_s_key("Subject");
 
                self.write_expr(heap, expr.subject, indent3);
 
                self.kv(indent2).with_s_key("FromIndex");
 
                self.write_expr(heap, expr.from_index, indent3);
 
                self.kv(indent2).with_s_key("ToIndex");
 
                self.write_expr(heap, expr.to_index, indent3);
 
                self.kv(indent2).with_s_key("Parent")
 
                    .with_custom_val(|v| write_expression_parent(v, &expr.parent));
 
            },
 
            Expression::Select(expr) => {
 
                self.kv(indent).with_id(PREFIX_SELECT_EXPR_ID, expr.this.0.index)
 
                    .with_s_key("SelectExpr");
 
                self.kv(indent2).with_s_key("Subject");
 
                self.write_expr(heap, expr.subject, indent3);
 

	
 
                self.kv(indent2).with_s_key("Field").with_identifier_val(&expr.field_name);
 
                self.kv(indent2).with_s_key("Parent")
 
                    .with_custom_val(|v| write_expression_parent(v, &expr.parent));
 
            },
 
            Expression::Literal(expr) => {
 
                self.kv(indent).with_id(PREFIX_LITERAL_EXPR_ID, expr.this.0.index)
 
                    .with_s_key("LiteralExpr");
 

	
 
                let val = self.kv(indent2).with_s_key("Value");
 
                match &expr.value {
 
                    Literal::Null => { val.with_s_val("null"); },
 
                    Literal::True => { val.with_s_val("true"); },
 
                    Literal::False => { val.with_s_val("false"); },
 
                    Literal::Character(data) => { val.with_disp_val(data); },
 
                    Literal::String(data) => {
 
                        // Stupid hack
 
                        let string = String::from(data.as_str());
 
                        val.with_disp_val(&string);
 
                    },
 
                    Literal::Integer(data) => { val.with_debug_val(data); },
 
                    Literal::Struct(data) => {
 
                        val.with_s_val("Struct");
 
                        let indent4 = indent3 + 1;
 

	
 
                        self.kv(indent3).with_s_key("ParserType")
 
                            .with_custom_val(|t| write_parser_type(t, heap, &data.parser_type));
 
                        self.kv(indent3).with_s_key("Definition").with_disp_val(&data.definition.index);
 

	
 
                        for field in &data.fields {
 
                            self.kv(indent3).with_s_key("Field");
 
                            self.kv(indent4).with_s_key("Name").with_identifier_val(&field.identifier);
 
                            self.kv(indent4).with_s_key("Index").with_disp_val(&field.field_idx);
 
                            self.kv(indent4).with_s_key("ParserType");
 
                            self.write_expr(heap, field.value, indent4 + 1);
 
                        }
 
                    },
 
                    Literal::Enum(data) => {
 
                        val.with_s_val("Enum");
 

	
 
                        self.kv(indent3).with_s_key("ParserType")
 
                            .with_custom_val(|t| write_parser_type(t, heap, &data.parser_type));
 
                        self.kv(indent3).with_s_key("Definition").with_disp_val(&data.definition.index);
 
                        self.kv(indent3).with_s_key("VariantIdx").with_disp_val(&data.variant_idx);
 
                    },
 
                    Literal::Union(data) => {
 
                        val.with_s_val("Union");
 
                        let indent4 = indent3 + 1;
 

	
 
                        self.kv(indent3).with_s_key("ParserType")
 
                            .with_custom_val(|t| write_parser_type(t, heap, &data.parser_type));
 
                        self.kv(indent3).with_s_key("Definition").with_disp_val(&data.definition.index);
 
                        self.kv(indent3).with_s_key("VariantIdx").with_disp_val(&data.variant_idx);
 

	
 
                        for value in &data.values {
 
                            self.kv(indent3).with_s_key("Value");
 
                            self.write_expr(heap, *value, indent4);
 
                        }
 
                    }
 
                    Literal::Array(data) => {
 
                        val.with_s_val("Array");
 
                        let indent4 = indent3 + 1;
 

	
 
                        self.kv(indent3).with_s_key("Elements");
 
                        for expr_id in data {
 
                            self.write_expr(heap, *expr_id, indent4);
 
                        }
 
                    }
 
                }
 

	
 
                self.kv(indent2).with_s_key("Parent")
 
                    .with_custom_val(|v| write_expression_parent(v, &expr.parent));
 
            },
 
            Expression::Cast(expr) => {
 
                self.kv(indent).with_id(PREFIX_CAST_EXPR_ID, expr.this.0.index)
 
                    .with_s_key("CallExpr");
 
                self.kv(indent2).with_s_key("ToType")
 
                    .with_custom_val(|t| write_parser_type(t, heap, &expr.to_type));
 
                self.kv(indent2).with_s_key("Subject");
 
                self.write_expr(heap, expr.subject, indent3);
 
                self.kv(indent2).with_s_key("Parent")
 
                    .with_custom_val(|v| write_expression_parent(v, &expr.parent));
 
            }
 
            Expression::Call(expr) => {
 
                self.kv(indent).with_id(PREFIX_CALL_EXPR_ID, expr.this.0.index)
 
                    .with_s_key("CallExpr");
 

	
 
                let definition = &heap[expr.definition];
 
                match definition {
 
                    Definition::Component(definition) => {
 
                        self.kv(indent2).with_s_key("BuiltIn").with_disp_val(&false);
 
                        self.kv(indent2).with_s_key("Variant").with_debug_val(&definition.variant);
 
                    },
 
                    Definition::Function(definition) => {
 
                        self.kv(indent2).with_s_key("BuiltIn").with_disp_val(&definition.builtin);
 
                        self.kv(indent2).with_s_key("Variant").with_s_val("Function");
 
                    },
 
                    _ => unreachable!()
 
                }
 
                self.kv(indent2).with_s_key("MethodName").with_identifier_val(definition.identifier());
 
                self.kv(indent2).with_s_key("ParserType")
 
                    .with_custom_val(|t| write_parser_type(t, heap, &expr.parser_type));
 

	
 
                // Arguments
 
                self.kv(indent2).with_s_key("Arguments");
 
                for arg_id in &expr.arguments {
 
                    self.write_expr(heap, *arg_id, indent3);
 
                }
 

	
 
                // Parent
 
                self.kv(indent2).with_s_key("Parent")
 
                    .with_custom_val(|v| write_expression_parent(v, &expr.parent));
 
            },
 
            Expression::Variable(expr) => {
 
                self.kv(indent).with_id(PREFIX_VARIABLE_EXPR_ID, expr.this.0.index)
 
                    .with_s_key("VariableExpr");
 
                self.kv(indent2).with_s_key("Name").with_identifier_val(&expr.identifier);
 
                self.kv(indent2).with_s_key("Definition")
 
                    .with_opt_disp_val(expr.declaration.as_ref().map(|v| &v.index));
 
                self.kv(indent2).with_s_key("Parent")
 
                    .with_custom_val(|v| write_expression_parent(v, &expr.parent));
 
            }
 
        }
 
    }
 

	
 
    fn write_variable(&mut self, heap: &Heap, variable_id: VariableId, indent: usize) {
 
        let var = &heap[variable_id];
 
        let indent2 = indent + 1;
 

	
 
        self.kv(indent).with_id(PREFIX_VARIABLE_ID, variable_id.index)
 
            .with_s_key("Variable");
 

	
 
        self.kv(indent2).with_s_key("Name").with_identifier_val(&var.identifier);
 
        self.kv(indent2).with_s_key("Kind").with_debug_val(&var.kind);
 
        self.kv(indent2).with_s_key("ParserType")
 
            .with_custom_val(|w| write_parser_type(w, heap, &var.parser_type));
 
        self.kv(indent2).with_s_key("RelativePos").with_disp_val(&var.relative_pos_in_block);
 
        self.kv(indent2).with_s_key("UniqueScopeID").with_disp_val(&var.unique_id_in_scope);
 
    }
 

	
 
    //--------------------------------------------------------------------------
 
    // Printing Utilities
 
    //--------------------------------------------------------------------------
 

	
 
    fn kv(&mut self, indent: usize) -> KV {
 
        KV::new(&mut self.buffer, &mut self.temp1, &mut self.temp2, indent)
 
    }
 

	
 
    fn flush<W: IOWrite>(&mut self, w: &mut W) {
 
        w.write(self.buffer.as_bytes()).unwrap();
 
        self.buffer.clear()
 
    }
 
}
 

	
 
fn write_option<V: Display>(target: &mut String, value: Option<V>) {
 
    target.clear();
 
    match &value {
 
        Some(v) => target.push_str(&format!("Some({})", v)),
 
        None => target.push_str("None")
 
    };
 
}
 

	
 
fn write_parser_type(target: &mut String, heap: &Heap, t: &ParserType) {
 
    use ParserTypeVariant as PTV;
 

	
 
    fn write_element(target: &mut String, heap: &Heap, t: &ParserType, mut element_idx: usize) -> usize {
 
        let element = &t.elements[element_idx];
 
        match &element.variant {
 
            PTV::Void => target.push_str("void"),
 
            PTV::InputOrOutput => {
 
                target.push_str("portlike<");
 
                element_idx = write_element(target, heap, t, element_idx + 1);
 
                target.push('>');
 
            },
 
            PTV::ArrayLike => {
 
                element_idx = write_element(target, heap, t, element_idx + 1);
 
                target.push_str("[???]");
 
            },
 
            PTV::IntegerLike => target.push_str("integerlike"),
 
            PTV::Message => { target.push_str(KW_TYPE_MESSAGE_STR); },
 
            PTV::Bool => { target.push_str(KW_TYPE_BOOL_STR); },
 
            PTV::UInt8 => { target.push_str(KW_TYPE_UINT8_STR); },
 
            PTV::UInt16 => { target.push_str(KW_TYPE_UINT16_STR); },
 
            PTV::UInt32 => { target.push_str(KW_TYPE_UINT32_STR); },
 
            PTV::UInt64 => { target.push_str(KW_TYPE_UINT64_STR); },
 
            PTV::SInt8 => { target.push_str(KW_TYPE_SINT8_STR); },
 
            PTV::SInt16 => { target.push_str(KW_TYPE_SINT16_STR); },
 
            PTV::SInt32 => { target.push_str(KW_TYPE_SINT32_STR); },
 
            PTV::SInt64 => { target.push_str(KW_TYPE_SINT64_STR); },
 
            PTV::Character => { target.push_str(KW_TYPE_CHAR_STR); },
 
            PTV::String => { target.push_str(KW_TYPE_STRING_STR); },
 
            PTV::IntegerLiteral => { target.push_str("int_literal"); },
 
            PTV::Inferred => { target.push_str(KW_TYPE_INFERRED_STR); },
 
            PTV::Array => {
 
                element_idx = write_element(target, heap, t, element_idx + 1);
 
                target.push_str("[]");
 
            },
 
            PTV::Input => {
 
                target.push_str(KW_TYPE_IN_PORT_STR);
 
                target.push('<');
 
                element_idx = write_element(target, heap, t, element_idx + 1);
 
                target.push('>');
 
            },
 
            PTV::Output => {
 
                target.push_str(KW_TYPE_OUT_PORT_STR);
 
                target.push('<');
 
                element_idx = write_element(target, heap, t, element_idx + 1);
 
                target.push('>');
 
            },
 
            PTV::PolymorphicArgument(definition_id, arg_idx) => {
 
                let definition = &heap[*definition_id];
 
                let poly_var = &definition.poly_vars()[*arg_idx as usize].value;
 
                target.push_str(poly_var.as_str());
 
            },
 
            PTV::Definition(definition_id, num_embedded) => {
 
                let definition = &heap[*definition_id];
 
                let definition_ident = definition.identifier().value.as_str();
 
                target.push_str(definition_ident);
 

	
 
                let num_embedded = *num_embedded;
 
                if num_embedded != 0 {
 
                    target.push('<');
 
                    for embedded_idx in 0..num_embedded {
 
                        if embedded_idx != 0 {
 
                            target.push(',');
 
                        }
 
                        element_idx = write_element(target, heap, t, element_idx + 1);
 
                    }
 
                    target.push('>');
 
                }
 
            }
 
        }
 

	
 
        element_idx
 
    }
 

	
 
    write_element(target, heap, t, 0);
 
}
 

	
 
// TODO: @Cleanup, this is littered at three places in the codebase
 
fn write_concrete_type(target: &mut String, heap: &Heap, def_id: DefinitionId, t: &ConcreteType) {
 
    use ConcreteTypePart as CTP;
 

	
 
    fn write_concrete_part(target: &mut String, heap: &Heap, def_id: DefinitionId, t: &ConcreteType, mut idx: usize) -> usize {
 
        if idx >= t.parts.len() {
 
            return idx;
 
        }
 

	
 
        match &t.parts[idx] {
 
            CTP::Void => target.push_str("void"),
 
            CTP::Message => target.push_str("msg"),
 
            CTP::Bool => target.push_str("bool"),
 
            CTP::UInt8 => target.push_str(KW_TYPE_UINT8_STR),
 
            CTP::UInt16 => target.push_str(KW_TYPE_UINT16_STR),
 
            CTP::UInt32 => target.push_str(KW_TYPE_UINT32_STR),
 
            CTP::UInt64 => target.push_str(KW_TYPE_UINT64_STR),
 
            CTP::SInt8 => target.push_str(KW_TYPE_SINT8_STR),
 
            CTP::SInt16 => target.push_str(KW_TYPE_SINT16_STR),
 
            CTP::SInt32 => target.push_str(KW_TYPE_SINT32_STR),
 
            CTP::SInt64 => target.push_str(KW_TYPE_SINT64_STR),
 
            CTP::Character => target.push_str(KW_TYPE_CHAR_STR),
 
            CTP::String => target.push_str(KW_TYPE_STRING_STR),
 
            CTP::Array => {
 
                idx = write_concrete_part(target, heap, def_id, t, idx + 1);
 
                target.push_str("[]");
 
            },
 
            CTP::Slice => {
 
                idx = write_concrete_part(target, heap, def_id, t, idx + 1);
 
                target.push_str("[..]");
 
            }
 
            CTP::Input => {
 
                target.push_str("in<");
 
                idx = write_concrete_part(target, heap, def_id, t, idx + 1);
 
                target.push('>');
 
            },
 
            CTP::Output => {
 
                target.push_str("out<");
 
                idx = write_concrete_part(target, heap, def_id, t, idx + 1);
 
                target.push('>')
 
            },
 
            CTP::Instance(definition_id, num_embedded) => {
 
                let identifier = heap[*definition_id].identifier();
 
                target.push_str(identifier.value.as_str());
 
                target.push('<');
 
                for idx_embedded in 0..*num_embedded {
 
                    if idx_embedded != 0 {
 
                        target.push_str(", ");
 
                    }
 
                    idx = write_concrete_part(target, heap, def_id, t, idx + 1);
 
                }
 
                target.push('>');
 
            },
 
            CTP::Function(_, _) => todo!("AST printer for ConcreteTypePart::Function"),
 
            CTP::Component(_, _) => todo!("AST printer for ConcreteTypePart::Component"),
 
        }
 

	
 
        idx + 1
 
    }
 

	
 
    write_concrete_part(target, heap, def_id, t, 0);
 
}
 

	
 
fn write_expression_parent(target: &mut String, parent: &ExpressionParent) {
 
    use ExpressionParent as EP;
 

	
 
    *target = match parent {
 
        EP::None => String::from("None"),
 
        EP::If(id) => format!("IfStmt({})", id.0.index),
 
        EP::While(id) => format!("WhileStmt({})", id.0.index),
 
        EP::Return(id) => format!("ReturnStmt({})", id.0.index),
 
        EP::New(id) => format!("NewStmt({})", id.0.index),
 
        EP::ExpressionStmt(id) => format!("ExprStmt({})", id.0.index),
 
        EP::Expression(id, idx) => format!("Expr({}, {})", id.index, idx)
 
    };
 
}
 
\ No newline at end of file
src/protocol/eval/value.rs
Show inline comments
 
@@ -56,769 +56,768 @@ pub enum Value {
 

	
 
macro_rules! impl_union_unpack_as_value {
 
    ($func_name:ident, $variant_name:path, $return_type:ty) => {
 
        impl Value {
 
            pub(crate) fn $func_name(&self) -> $return_type {
 
                match self {
 
                    $variant_name(v) => *v,
 
                    _ => panic!(concat!("called ", stringify!($func_name()), " on {:?}"), self),
 
                }
 
            }
 
        }
 
    }
 
}
 

	
 
impl_union_unpack_as_value!(as_stack_boundary, Value::PrevStackBoundary, isize);
 
impl_union_unpack_as_value!(as_ref,     Value::Ref,     ValueId);
 
impl_union_unpack_as_value!(as_input,   Value::Input,   PortId);
 
impl_union_unpack_as_value!(as_output,  Value::Output,  PortId);
 
impl_union_unpack_as_value!(as_message, Value::Message, HeapPos);
 
impl_union_unpack_as_value!(as_bool,    Value::Bool,    bool);
 
impl_union_unpack_as_value!(as_char,    Value::Char,    char);
 
impl_union_unpack_as_value!(as_string,  Value::String,  HeapPos);
 
impl_union_unpack_as_value!(as_uint8,   Value::UInt8,   u8);
 
impl_union_unpack_as_value!(as_uint16,  Value::UInt16,  u16);
 
impl_union_unpack_as_value!(as_uint32,  Value::UInt32,  u32);
 
impl_union_unpack_as_value!(as_uint64,  Value::UInt64,  u64);
 
impl_union_unpack_as_value!(as_sint8,   Value::SInt8,   i8);
 
impl_union_unpack_as_value!(as_sint16,  Value::SInt16,  i16);
 
impl_union_unpack_as_value!(as_sint32,  Value::SInt32,  i32);
 
impl_union_unpack_as_value!(as_sint64,  Value::SInt64,  i64);
 
impl_union_unpack_as_value!(as_array,   Value::Array,   HeapPos);
 
impl_union_unpack_as_value!(as_enum,    Value::Enum,    i64);
 
impl_union_unpack_as_value!(as_struct,  Value::Struct,  HeapPos);
 

	
 
impl Value {
 
    pub(crate) fn as_union(&self) -> (i64, HeapPos) {
 
        match self {
 
            Value::Union(tag, v) => (*tag, *v),
 
            _ => panic!("called as_union on {:?}", self),
 
        }
 
    }
 

	
 
    pub(crate) fn is_integer(&self) -> bool {
 
        match self {
 
            Value::UInt8(_) | Value::UInt16(_) | Value::UInt32(_) | Value::UInt64(_) |
 
            Value::SInt8(_) | Value::SInt16(_) | Value::SInt32(_) | Value::SInt64(_) => true,
 
            _ => false
 
        }
 
    }
 

	
 
    pub(crate) fn is_unsigned_integer(&self) -> bool {
 
        match self {
 
            Value::UInt8(_) | Value::UInt16(_) | Value::UInt32(_) | Value::UInt64(_) => true,
 
            _ => false
 
        }
 
    }
 

	
 
    pub(crate) fn is_signed_integer(&self) -> bool {
 
        match self {
 
            Value::SInt8(_) | Value::SInt16(_) | Value::SInt32(_) | Value::SInt64(_) => true,
 
            _ => false
 
        }
 
    }
 

	
 
    pub(crate) fn as_unsigned_integer(&self) -> u64 {
 
        match self {
 
            Value::UInt8(v)  => *v as u64,
 
            Value::UInt16(v) => *v as u64,
 
            Value::UInt32(v) => *v as u64,
 
            Value::UInt64(v) => *v as u64,
 
            _ => unreachable!("called as_unsigned_integer on {:?}", self),
 
        }
 
    }
 

	
 
    pub(crate) fn as_signed_integer(&self) -> i64 {
 
        match self {
 
            Value::SInt8(v)  => *v as i64,
 
            Value::SInt16(v) => *v as i64,
 
            Value::SInt32(v) => *v as i64,
 
            Value::SInt64(v) => *v as i64,
 
            _ => unreachable!("called as_signed_integer on {:?}", self)
 
        }
 
    }
 

	
 
    /// Returns the heap position associated with the value. If the value
 
    /// doesn't store anything in the heap then we return `None`.
 
    pub(crate) fn get_heap_pos(&self) -> Option<HeapPos> {
 
        match self {
 
            Value::Message(v) => Some(*v),
 
            Value::String(v) => Some(*v),
 
            Value::Array(v) => Some(*v),
 
            Value::Union(_, v) => Some(*v),
 
            Value::Struct(v) => Some(*v),
 
            _ => None
 
        }
 
    }
 
}
 

	
 
/// When providing arguments to a new component, or when transferring values
 
/// from one component's store to a newly instantiated component, one has to
 
/// transfer stack and heap values. This `ValueGroup` represents such a
 
/// temporary group of values with potential heap allocations.
 
///
 
/// Constructing such a ValueGroup manually requires some extra care to make
 
/// sure all elements of `values` point to valid elements of `regions`.
 
///
 
/// Again: this is a temporary thing, hopefully removed once we move to a
 
/// bytecode interpreter.
 
#[derive(Clone, Debug)]
 
pub struct ValueGroup {
 
    pub(crate) values: Vec<Value>,
 
    pub(crate) regions: Vec<Vec<Value>>
 
}
 

	
 
impl ValueGroup {
 
    pub(crate) fn new_stack(values: Vec<Value>) -> Self {
 
        debug_assert!(values.iter().all(|v| v.get_heap_pos().is_none()));
 
        Self{
 
            values,
 
            regions: Vec::new(),
 
        }
 
    }
 
    pub(crate) fn from_store(store: &Store, values: &[Value]) -> Self {
 
        let mut group = ValueGroup{
 
            values: Vec::with_capacity(values.len()),
 
            regions: Vec::with_capacity(values.len()), // estimation
 
        };
 

	
 
        for value in values {
 
            let transferred = group.retrieve_value(value, store);
 
            group.values.push(transferred);
 
        }
 

	
 
        group
 
    }
 

	
 
    /// Transfers a provided value from a store into a local value with its
 
    /// heap allocations (if any) stored in the ValueGroup. Calling this
 
    /// function will not store the returned value in the `values` member.
 
    fn retrieve_value(&mut self, value: &Value, from_store: &Store) -> Value {
 
        let value = from_store.maybe_read_ref(value);
 
        if let Some(heap_pos) = value.get_heap_pos() {
 
            // Value points to a heap allocation, so transfer the heap values
 
            // internally.
 
            let from_region = &from_store.heap_regions[heap_pos as usize].values;
 
            let mut new_region = Vec::with_capacity(from_region.len());
 
            for value in from_region {
 
                let transferred = self.retrieve_value(value, from_store);
 
                new_region.push(transferred);
 
            }
 

	
 
            // Region is constructed, store internally and return the new value.
 
            let new_region_idx = self.regions.len() as HeapPos;
 
            self.regions.push(new_region);
 

	
 
            return match value {
 
                Value::Message(_)    => Value::Message(new_region_idx),
 
                Value::String(_)     => Value::String(new_region_idx),
 
                Value::Array(_)      => Value::Array(new_region_idx),
 
                Value::Union(tag, _) => Value::Union(*tag, new_region_idx),
 
                Value::Struct(_)     => Value::Struct(new_region_idx),
 
                _ => unreachable!(),
 
            };
 
        } else {
 
            return value.clone();
 
        }
 
    }
 

	
 
    /// Transfers the heap values and the stack values into the store. Stack
 
    /// values are pushed onto the Store's stack in the order in which they
 
    /// appear in the value group.
 
    pub(crate) fn into_store(self, store: &mut Store) {
 
        for value in &self.values {
 
            let transferred = self.provide_value(value, store);
 
            store.stack.push(transferred);
 
        }
 
    }
 

	
 
    /// Transfers the heap values into the store, but will put the stack values
 
    /// into the provided `VecDeque`. This is mainly used to merge `ValueGroup`
 
    /// instances retrieved by the code by `get` calls into the expression
 
    /// stack.
 
    pub(crate) fn into_stack(self, stack: &mut VecDeque<Value>, store: &mut Store) {
 
        for value in &self.values {
 
            let transferred = self.provide_value(value, store);
 
            stack.push_back(transferred);
 
        }
 
    }
 

	
 
    fn provide_value(&self, value: &Value, to_store: &mut Store) -> Value {
 
        if let Some(from_heap_pos) = value.get_heap_pos() {
 
            let from_heap_pos = from_heap_pos as usize;
 
            let to_heap_pos = to_store.alloc_heap();
 
            let to_heap_pos_usize = to_heap_pos as usize;
 
            to_store.heap_regions[to_heap_pos_usize].values.reserve(self.regions[from_heap_pos].len());
 

	
 
            for value in &self.regions[from_heap_pos as usize] {
 
                let transferred = self.provide_value(value, to_store);
 
                to_store.heap_regions[to_heap_pos_usize].values.push(transferred);
 
            }
 

	
 
            return match value {
 
                Value::Message(_)    => Value::Message(to_heap_pos),
 
                Value::String(_)     => Value::String(to_heap_pos),
 
                Value::Array(_)      => Value::Array(to_heap_pos),
 
                Value::Union(tag, _) => Value::Union(*tag, to_heap_pos),
 
                Value::Struct(_)     => Value::Struct(to_heap_pos),
 
                _ => unreachable!(),
 
            };
 
        } else {
 
            return value.clone();
 
        }
 
    }
 
}
 

	
 
impl Default for ValueGroup {
 
    /// Returns an empty ValueGroup
 
    fn default() -> Self {
 
        Self { values: Vec::new(), regions: Vec::new() }
 
    }
 
}
 

	
 
enum ValueKind { Message, String, Array }
 

	
 
pub(crate) fn apply_assignment_operator(store: &mut Store, lhs: ValueId, op: AssignmentOperator, rhs: Value) {
 
    use AssignmentOperator as AO;
 

	
 
    macro_rules! apply_int_op {
 
        ($lhs:ident, $assignment_tokens:tt, $operator:ident, $rhs:ident) => {
 
            match $lhs {
 
                Value::UInt8(v)  => { *v $assignment_tokens $rhs.as_uint8();  },
 
                Value::UInt16(v) => { *v $assignment_tokens $rhs.as_uint16(); },
 
                Value::UInt32(v) => { *v $assignment_tokens $rhs.as_uint32(); },
 
                Value::UInt64(v) => { *v $assignment_tokens $rhs.as_uint64(); },
 
                Value::SInt8(v)  => { *v $assignment_tokens $rhs.as_sint8();  },
 
                Value::SInt16(v) => { *v $assignment_tokens $rhs.as_sint16(); },
 
                Value::SInt32(v) => { *v $assignment_tokens $rhs.as_sint32(); },
 
                Value::SInt64(v) => { *v $assignment_tokens $rhs.as_sint64(); },
 
                _ => unreachable!("apply_assignment_operator {:?} on lhs {:?} and rhs {:?}", $operator, $lhs, $rhs),
 
            }
 
        }
 
    }
 

	
 
    let lhs = store.read_mut_ref(lhs);
 

	
 
    let mut to_dealloc = None;
 
    match op {
 
        AO::Set => {
 
            match lhs {
 
                Value::Unassigned => { *lhs = rhs; },
 
                Value::Input(v)  => { *v = rhs.as_input(); },
 
                Value::Output(v) => { *v = rhs.as_output(); },
 
                Value::Message(v)  => { to_dealloc = Some(*v); *v = rhs.as_message(); },
 
                Value::Bool(v)    => { *v = rhs.as_bool(); },
 
                Value::Char(v) => { *v = rhs.as_char(); },
 
                Value::String(v) => { *v = rhs.as_string().clone(); },
 
                Value::UInt8(v) => { *v = rhs.as_uint8(); },
 
                Value::UInt16(v) => { *v = rhs.as_uint16(); },
 
                Value::UInt32(v) => { *v = rhs.as_uint32(); },
 
                Value::UInt64(v) => { *v = rhs.as_uint64(); },
 
                Value::SInt8(v) => { *v = rhs.as_sint8(); },
 
                Value::SInt16(v) => { *v = rhs.as_sint16(); },
 
                Value::SInt32(v) => { *v = rhs.as_sint32(); },
 
                Value::SInt64(v) => { *v = rhs.as_sint64(); },
 
                Value::Array(v) => { to_dealloc = Some(*v); *v = rhs.as_array(); },
 
                Value::Enum(v) => { *v = rhs.as_enum(); },
 
                Value::Union(lhs_tag, lhs_heap_pos) => {
 
                    to_dealloc = Some(*lhs_heap_pos);
 
                    let (rhs_tag, rhs_heap_pos) = rhs.as_union();
 
                    *lhs_tag = rhs_tag;
 
                    *lhs_heap_pos = rhs_heap_pos;
 
                }
 
                Value::Struct(v) => { to_dealloc = Some(*v); *v = rhs.as_struct(); },
 
                _ => unreachable!("apply_assignment_operator {:?} on lhs {:?} and rhs {:?}", op, lhs, rhs),
 
            }
 
        },
 
        AO::Concatenated => {
 
            let lhs_heap_pos = lhs.get_heap_pos().unwrap() as usize;
 
            let rhs_heap_pos = rhs.get_heap_pos().unwrap() as usize;
 

	
 
            // To prevent borrowing crap, swap out heap region with a temp empty array
 
            let mut total = Vec::new();
 
            std::mem::swap(&mut total, &mut store.heap_regions[lhs_heap_pos].values);
 

	
 
            // Push everything onto the swapped vector
 
            let rhs_len = store.heap_regions[rhs_heap_pos].values.len();
 
            total.reserve(rhs_len);
 
            for value_idx in 0..rhs_len {
 
                total.push(store.clone_value(store.heap_regions[rhs_heap_pos].values[value_idx].clone()));
 
            }
 

	
 
            // Swap back in place
 
            std::mem::swap(&mut total, &mut store.heap_regions[lhs_heap_pos].values);
 

	
 
            // We took ownership of the RHS, but we copied it into the LHS, so
 
            // different form assignment we need to drop the RHS heap pos.
 
            to_dealloc = Some(rhs_heap_pos as u32);
 
        },
 
        AO::Multiplied =>   { apply_int_op!(lhs, *=,  op, rhs) },
 
        AO::Divided =>      { apply_int_op!(lhs, /=,  op, rhs) },
 
        AO::Remained =>     { apply_int_op!(lhs, %=,  op, rhs) },
 
        AO::Added =>        { apply_int_op!(lhs, +=,  op, rhs) },
 
        AO::Subtracted =>   { apply_int_op!(lhs, -=,  op, rhs) },
 
        AO::ShiftedLeft =>  { apply_int_op!(lhs, <<=, op, rhs) },
 
        AO::ShiftedRight => { apply_int_op!(lhs, >>=, op, rhs) },
 
        AO::BitwiseAnded => { apply_int_op!(lhs, &=,  op, rhs) },
 
        AO::BitwiseXored => { apply_int_op!(lhs, ^=,  op, rhs) },
 
        AO::BitwiseOred =>  { apply_int_op!(lhs, |=,  op, rhs) },
 
    }
 

	
 
    if let Some(heap_pos) = to_dealloc {
 
        store.drop_heap_pos(heap_pos);
 
    }
 
}
 

	
 
pub(crate) fn apply_binary_operator(store: &mut Store, lhs: &Value, op: BinaryOperator, rhs: &Value) -> Value {
 
    use BinaryOperator as BO;
 

	
 
    macro_rules! apply_int_op_and_return_self {
 
        ($lhs:ident, $operator_tokens:tt, $operator:ident, $rhs:ident) => {
 
            return match $lhs {
 
                Value::UInt8(v)  => { Value::UInt8( *v $operator_tokens $rhs.as_uint8() ) },
 
                Value::UInt16(v) => { Value::UInt16(*v $operator_tokens $rhs.as_uint16()) },
 
                Value::UInt32(v) => { Value::UInt32(*v $operator_tokens $rhs.as_uint32()) },
 
                Value::UInt64(v) => { Value::UInt64(*v $operator_tokens $rhs.as_uint64()) },
 
                Value::SInt8(v)  => { Value::SInt8( *v $operator_tokens $rhs.as_sint8() ) },
 
                Value::SInt16(v) => { Value::SInt16(*v $operator_tokens $rhs.as_sint16()) },
 
                Value::SInt32(v) => { Value::SInt32(*v $operator_tokens $rhs.as_sint32()) },
 
                Value::SInt64(v) => { Value::SInt64(*v $operator_tokens $rhs.as_sint64()) },
 
                _ => unreachable!("apply_binary_operator {:?} on lhs {:?} and rhs {:?}", $operator, $lhs, $rhs)
 
            };
 
        }
 
    }
 

	
 
    macro_rules! apply_int_op_and_return_bool {
 
        ($lhs:ident, $operator_tokens:tt, $operator:ident, $rhs:ident) => {
 
            return match $lhs {
 
                Value::UInt8(v)  => { Value::Bool(*v $operator_tokens $rhs.as_uint8() ) },
 
                Value::UInt16(v) => { Value::Bool(*v $operator_tokens $rhs.as_uint16()) },
 
                Value::UInt32(v) => { Value::Bool(*v $operator_tokens $rhs.as_uint32()) },
 
                Value::UInt64(v) => { Value::Bool(*v $operator_tokens $rhs.as_uint64()) },
 
                Value::SInt8(v)  => { Value::Bool(*v $operator_tokens $rhs.as_sint8() ) },
 
                Value::SInt16(v) => { Value::Bool(*v $operator_tokens $rhs.as_sint16()) },
 
                Value::SInt32(v) => { Value::Bool(*v $operator_tokens $rhs.as_sint32()) },
 
                Value::SInt64(v) => { Value::Bool(*v $operator_tokens $rhs.as_sint64()) },
 
                _ => unreachable!("apply_binary_operator {:?} on lhs {:?} and rhs {:?}", $operator, $lhs, $rhs)
 
            };
 
        }
 
    }
 

	
 
    // We need to handle concatenate in a special way because it needs the store
 
    // mutably.
 
    if op == BO::Concatenate {
 
        let target_heap_pos = store.alloc_heap();
 
        let lhs_heap_pos;
 
        let rhs_heap_pos;
 

	
 
        let lhs = store.maybe_read_ref(lhs);
 
        let rhs = store.maybe_read_ref(rhs);
 

	
 
        let value_kind;
 

	
 
        match lhs {
 
            Value::Message(lhs_pos) => {
 
                lhs_heap_pos = *lhs_pos;
 
                rhs_heap_pos = rhs.as_message();
 
                value_kind = ValueKind::Message;
 
            },
 
            Value::String(lhs_pos) => {
 
                lhs_heap_pos = *lhs_pos;
 
                rhs_heap_pos = rhs.as_string();
 
                value_kind = ValueKind::String;
 
            },
 
            Value::Array(lhs_pos) => {
 
                lhs_heap_pos = *lhs_pos;
 
                rhs_heap_pos = rhs.as_array();
 
                value_kind = ValueKind::Array;
 
            },
 
            _ => unreachable!("apply_binary_operator {:?} on lhs {:?} and rhs {:?}", op, lhs, rhs)
 
        }
 

	
 
        let lhs_heap_pos = lhs_heap_pos as usize;
 
        let rhs_heap_pos = rhs_heap_pos as usize;
 

	
 
        // TODO: I hate this, but fine...
 
        let mut concatenated = Vec::new();
 
        let lhs_len = store.heap_regions[lhs_heap_pos].values.len();
 
        let rhs_len = store.heap_regions[rhs_heap_pos].values.len();
 
        concatenated.reserve(lhs_len + rhs_len);
 
        for idx in 0..lhs_len {
 
            concatenated.push(store.clone_value(store.heap_regions[lhs_heap_pos].values[idx].clone()));
 
        }
 
        for idx in 0..rhs_len {
 
            concatenated.push(store.clone_value(store.heap_regions[rhs_heap_pos].values[idx].clone()));
 
        }
 

	
 
        store.heap_regions[target_heap_pos as usize].values = concatenated;
 

	
 
        return match value_kind{
 
            ValueKind::Message => Value::Message(target_heap_pos),
 
            ValueKind::String => Value::String(target_heap_pos),
 
            ValueKind::Array => Value::Array(target_heap_pos),
 
        };
 
    }
 

	
 
    // If any of the values are references, retrieve the thing they're referring
 
    // to.
 
    let lhs = store.maybe_read_ref(lhs);
 
    let rhs = store.maybe_read_ref(rhs);
 

	
 
    match op {
 
        BO::Concatenate => unreachable!(),
 
        BO::LogicalOr => {
 
            return Value::Bool(lhs.as_bool() || rhs.as_bool());
 
        },
 
        BO::LogicalAnd => {
 
            return Value::Bool(lhs.as_bool() && rhs.as_bool());
 
        },
 
        BO::BitwiseOr        => { apply_int_op_and_return_self!(lhs, |,  op, rhs); },
 
        BO::BitwiseXor       => { apply_int_op_and_return_self!(lhs, ^,  op, rhs); },
 
        BO::BitwiseAnd       => { apply_int_op_and_return_self!(lhs, &,  op, rhs); },
 
        BO::Equality         => { Value::Bool(apply_equality_operator(store, lhs, rhs)) },
 
        BO::Inequality       => { Value::Bool(apply_inequality_operator(store, lhs, rhs)) },
 
        BO::LessThan         => { apply_int_op_and_return_bool!(lhs, <,  op, rhs); },
 
        BO::GreaterThan      => { apply_int_op_and_return_bool!(lhs, >,  op, rhs); },
 
        BO::LessThanEqual    => { apply_int_op_and_return_bool!(lhs, <=, op, rhs); },
 
        BO::GreaterThanEqual => { apply_int_op_and_return_bool!(lhs, >=, op, rhs); },
 
        BO::ShiftLeft        => { apply_int_op_and_return_self!(lhs, <<, op, rhs); },
 
        BO::ShiftRight       => { apply_int_op_and_return_self!(lhs, >>, op, rhs); },
 
        BO::Add              => { apply_int_op_and_return_self!(lhs, +,  op, rhs); },
 
        BO::Subtract         => { apply_int_op_and_return_self!(lhs, -,  op, rhs); },
 
        BO::Multiply         => { apply_int_op_and_return_self!(lhs, *,  op, rhs); },
 
        BO::Divide           => { apply_int_op_and_return_self!(lhs, /,  op, rhs); },
 
        BO::Remainder        => { apply_int_op_and_return_self!(lhs, %,  op, rhs); }
 
    }
 
}
 

	
 
pub(crate) fn apply_unary_operator(store: &mut Store, op: UnaryOperator, value: &Value) -> Value {
 
    use UnaryOperator as UO;
 

	
 
    macro_rules! apply_int_expr_and_return {
 
        ($value:ident, $apply:tt, $op:ident) => {
 
            return match $value {
 
                Value::UInt8(v)  => Value::UInt8($apply *v),
 
                Value::UInt16(v) => Value::UInt16($apply *v),
 
                Value::UInt32(v) => Value::UInt32($apply *v),
 
                Value::UInt64(v) => Value::UInt64($apply *v),
 
                Value::SInt8(v)  => Value::SInt8($apply *v),
 
                Value::SInt16(v) => Value::SInt16($apply *v),
 
                Value::SInt32(v) => Value::SInt32($apply *v),
 
                Value::SInt64(v) => Value::SInt64($apply *v),
 
                _ => unreachable!("apply_unary_operator {:?} on value {:?}", $op, $value),
 
            };
 
        }
 
    }
 

	
 
    // If the value is a reference, retrieve the thing it is referring to
 
    let value = store.maybe_read_ref(value);
 

	
 
    match op {
 
        UO::Positive => {
 
            debug_assert!(value.is_integer());
 
            return value.clone();
 
        },
 
        UO::Negative => {
 
            // TODO: Error on negating unsigned integers
 
            return match value {
 
                Value::SInt8(v) => Value::SInt8(-*v),
 
                Value::SInt16(v) => Value::SInt16(-*v),
 
                Value::SInt32(v) => Value::SInt32(-*v),
 
                Value::SInt64(v) => Value::SInt64(-*v),
 
                _ => unreachable!("apply_unary_operator {:?} on value {:?}", op, value),
 
            }
 
        },
 
        UO::BitwiseNot => { apply_int_expr_and_return!(value, !, op)},
 
        UO::LogicalNot => { return Value::Bool(!value.as_bool()); },
 
    }
 
}
 

	
 
pub(crate) fn apply_casting(store: &mut Store, output_type: &ConcreteType, subject: &Value) -> Result<Value, String> {
 
    // To simplify the casting logic: if the output type is not a simple
 
    // integer/boolean/character, then the type checker made sure that the two
 
    // types must be equal, hence we can do a simple clone.
 
    use ConcreteTypePart as CTP;
 
    let part = &output_type.parts[0];
 
    match part {
 
        CTP::Bool | CTP::Character |
 
        CTP::UInt8 | CTP::UInt16 | CTP::UInt32 | CTP::UInt64 |
 
        CTP::SInt8 | CTP::SInt16 | CTP::SInt32 | CTP::SInt64 => {
 
            // Do the checking of these below
 
            debug_assert_eq!(output_type.parts.len(), 1);
 
        },
 
        _ => {
 
            return Ok(store.clone_value(subject.clone()));
 
        },
 
    }
 

	
 
    // Note: character is not included, needs per-type checking
 
    macro_rules! unchecked_cast {
 
        ($input: expr, $output_part: expr) => {
 
            return Ok(match $output_part {
 
                CTP::UInt8 => Value::UInt8($input as u8),
 
                CTP::UInt16 => Value::UInt16($input as u16),
 
                CTP::UInt32 => Value::UInt32($input as u32),
 
                CTP::UInt64 => Value::UInt64($input as u64),
 
                CTP::SInt8 => Value::SInt8($input as i8),
 
                CTP::SInt16 => Value::SInt16($input as i16),
 
                CTP::SInt32 => Value::SInt32($input as i32),
 
                CTP::SInt64 => Value::SInt64($input as i64),
 
                _ => unreachable!()
 
            })
 
        }
 
    }
 

	
 
    macro_rules! from_unsigned_cast {
 
        ($input:expr, $input_type:ty, $output_part:expr) => {
 
            {
 
                let target_type_name = match $output_part {
 
                    CTP::Bool => return Ok(Value::Bool($input != 0)),
 
                    CTP::Character => if $input <= u8::MAX as $input_type {
 
                        return Ok(Value::Char(($input as u8) as char))
 
                    } else {
 
                        KW_TYPE_CHAR_STR
 
                    },
 
                    CTP::UInt8 => if $input <= u8::MAX as $input_type {
 
                        return Ok(Value::UInt8($input as u8))
 
                    } else {
 
                        KW_TYPE_UINT8_STR
 
                    },
 
                    CTP::UInt16 => if $input <= u16::MAX as $input_type {
 
                        return Ok(Value::UInt16($input as u16))
 
                    } else {
 
                        KW_TYPE_UINT16_STR
 
                    },
 
                    CTP::UInt32 => if $input <= u32::MAX as $input_type {
 
                        return Ok(Value::UInt32($input as u32))
 
                    } else {
 
                        KW_TYPE_UINT32_STR
 
                    },
 
                    CTP::UInt64 => return Ok(Value::UInt64($input as u64)), // any unsigned int to u64 is fine
 
                    CTP::SInt8 => if $input <= i8::MAX as $input_type {
 
                        return Ok(Value::SInt8($input as i8))
 
                    } else {
 
                        KW_TYPE_SINT8_STR
 
                    },
 
                    CTP::SInt16 => if $input <= i16::MAX as $input_type {
 
                        return Ok(Value::SInt16($input as i16))
 
                    } else {
 
                        KW_TYPE_SINT16_STR
 
                    },
 
                    CTP::SInt32 => if $input <= i32::MAX as $input_type {
 
                        return Ok(Value::SInt32($input as i32))
 
                    } else {
 
                        KW_TYPE_SINT32_STR
 
                    },
 
                    CTP::SInt64 => if $input <= i64::MAX as $input_type {
 
                        return Ok(Value::SInt64($input as i64))
 
                    } else {
 
                        KW_TYPE_SINT64_STR
 
                    },
 
                    _ => unreachable!(),
 
                };
 

	
 
                return Err(format!("value is '{}' which doesn't fit in a type '{}'", $input, target_type_name));
 
            }
 
        }
 
    }
 

	
 
    macro_rules! from_signed_cast {
 
        // Programmer note: for signed checking we cannot do
 
        //  output_type::MAX as input_type,
 
        //
 
        // because if the output type's width is larger than the input type,
 
        // then the cast results in a negative number. So we mask with the
 
        // maximum possible value the input type can become. As in:
 
        //  (output_type::MAX as input_type) & input_type::MAX
 
        //
 
        // This way:
 
        // 1. output width is larger than input width: fine in all cases, we
 
        //  simply compare against the max input value, which is always true.
 
        // 2. output width is equal to input width: by masking we "remove the
 
        //  signed bit from the unsigned number" and again compare against the
 
        //  maximum input value.
 
        // 3. output width is smaller than the input width: masking does nothing
 
        //  because the signed bit is never set, and we simply compare against
 
        //  the maximum possible output value.
 
        //
 
        // A similar kind of mechanism for the minimum value, but here we do
 
        // a binary OR. We do a:
 
        //  (output_type::MIN as input_type) & input_type::MIN
 
        //
 
        // This way:
 
        // 1. output width is larger than input width: initial cast truncates to
 
        //  0, then we OR with the actual minimum value, so we attain the
 
        //  minimum value of the input type.
 
        // 2. output width is equal to input width: we OR the minimum value with
 
        //  itself.
 
        // 3. output width is smaller than input width: the cast produces the
 
        //  min value of the output type, the subsequent OR does nothing, as it
 
        //  essentially just sets the signed bit (which must already be set,
 
        //  since we're dealing with a signed minimum value)
 
        //
 
        // After all of this expanding, we simply hope the compiler does a best
 
        // effort constant expression evaluation, and presto!
 
        ($input:expr, $input_type:ty, $output_type:expr) => {
 
            {
 
                let target_type_name = match $output_type {
 
                    CTP::Bool => return Ok(Value::Bool($input != 0)),
 
                    CTP::Character => if $input >= 0 && $input <= (u8::max as $input_type & <$input_type>::MAX) {
 
                        return Ok(Value::Char(($input as u8) as char))
 
                    } else {
 
                        KW_TYPE_CHAR_STR
 
                    },
 
                    CTP::UInt8 => if $input >= 0 && $input <= ((u8::MAX as $input_type) & <$input_type>::MAX) {
 
                        return Ok(Value::UInt8($input as u8));
 
                    } else {
 
                        KW_TYPE_UINT8_STR
 
                    },
 
                    CTP::UInt16 => if $input >= 0 && $input <= ((u16::MAX as $input_type) & <$input_type>::MAX) {
 
                        return Ok(Value::UInt16($input as u16));
 
                    } else {
 
                        KW_TYPE_UINT16_STR
 
                    },
 
                    CTP::UInt32 => if $input >= 0 && $input <= ((u32::MAX as $input_type) & <$input_type>::MAX) {
 
                        return Ok(Value::UInt32($input as u32));
 
                    } else {
 
                        KW_TYPE_UINT32_STR
 
                    },
 
                    CTP::UInt64 => if $input >= 0 && $input <= ((u64::MAX as $input_type) & <$input_type>::MAX) {
 
                        return Ok(Value::UInt64($input as u64));
 
                    } else {
 
                        KW_TYPE_UINT64_STR
 
                    },
 
                    CTP::SInt8 => if $input >= ((i8::MIN as $input_type) | <$input_type>::MIN) && $input <= ((i8::MAX as $input_type) & <$input_type>::MAX) {
 
                        return Ok(Value::SInt8($input as i8));
 
                    } else {
 
                        KW_TYPE_SINT8_STR
 
                    },
 
                    CTP::SInt16 => if $input >= ((i16::MIN as $input_type | <$input_type>::MIN)) && $input <= ((i16::MAX as $input_type) & <$input_type>::MAX) {
 
                        return Ok(Value::SInt16($input as i16));
 
                    } else {
 
                        KW_TYPE_SINT16_STR
 
                    },
 
                    CTP::SInt32 => if $input >= ((i32::MIN as $input_type | <$input_type>::MIN)) && $input <= ((i32::MAX as $input_type) & <$input_type>::MAX) {
 
                        return Ok(Value::SInt32($input as i32));
 
                    } else {
 
                        KW_TYPE_SINT32_STR
 
                    },
 
                    CTP::SInt64 => return Ok(Value::SInt64($input as i64)),
 
                    _ => unreachable!(),
 
                };
 

	
 
                return Err(format!("value is '{}' which doesn't fit in a type '{}'", $input, target_type_name));
 
            }
 
        }
 
    }
 

	
 
    // If here, then the types might still be equal, but at least we're dealing
 
    // with a simple integer/boolean/character input and output type.
 
    let subject = store.maybe_read_ref(subject);
 
    match subject {
 
        Value::Bool(val) => {
 
            match part {
 
                CTP::Bool => return Ok(Value::Bool(*val)),
 
                CTP::Character => return Ok(Value::Char(1 as char)),
 
                _ => unchecked_cast!(*val, part),
 
            }
 
        },
 
        Value::Char(val) => {
 
            match part {
 
                CTP::Bool => return Ok(Value::Bool(*val != 0 as char)),
 
                CTP::Character => return Ok(Value::Char(*val)),
 
                _ => unchecked_cast!(*val, part),
 
            }
 
        },
 
        Value::UInt8(val) => from_unsigned_cast!(*val, u8, part),
 
        Value::UInt16(val) => from_unsigned_cast!(*val, u16, part),
 
        Value::UInt32(val) => from_unsigned_cast!(*val, u32, part),
 
        Value::UInt64(val) => from_unsigned_cast!(*val, u64, part),
 
        Value::SInt8(val) => from_signed_cast!(*val, i8, part),
 
        Value::SInt16(val) => from_signed_cast!(*val, i16, part),
 
        Value::SInt32(val) => from_signed_cast!(*val, i32, part),
 
        Value::SInt64(val) => from_signed_cast!(*val, i64, part),
 
        _ => unreachable!("mismatch between 'cast' type checking and 'cast' evaluation"),
 
    }
 
}
 

	
 
/// Recursively checks for equality.
 
pub(crate) fn apply_equality_operator(store: &Store, lhs: &Value, rhs: &Value) -> bool {
 
    let lhs = store.maybe_read_ref(lhs);
 
    let rhs = store.maybe_read_ref(rhs);
 

	
 
    fn eval_equality_heap(store: &Store, lhs_pos: HeapPos, rhs_pos: HeapPos) -> bool {
 
        let lhs_vals = &store.heap_regions[lhs_pos as usize].values;
 
        let rhs_vals = &store.heap_regions[rhs_pos as usize].values;
 
        let lhs_len = lhs_vals.len();
 
        if lhs_len != rhs_vals.len() {
 
            return false;
 
        }
 

	
 
        for idx in 0..lhs_len {
 
            let lhs_val = &lhs_vals[idx];
 
            let rhs_val = &rhs_vals[idx];
 
            if !apply_equality_operator(store, lhs_val, rhs_val) {
 
                return false;
 
            }
 
        }
 

	
 
        return true;
 
    }
 

	
 
    match lhs {
 
        Value::Input(v) => *v == rhs.as_input(),
 
        Value::Output(v) => *v == rhs.as_output(),
 
        Value::Message(lhs_pos) => eval_equality_heap(store, *lhs_pos, rhs.as_message()),
 
        Value::Null => todo!("remove null"),
 
        Value::Bool(v) => *v == rhs.as_bool(),
 
        Value::Char(v) => *v == rhs.as_char(),
 
        Value::String(lhs_pos) => eval_equality_heap(store, *lhs_pos, rhs.as_string()),
 
        Value::UInt8(v) => *v == rhs.as_uint8(),
 
        Value::UInt16(v) => *v == rhs.as_uint16(),
 
        Value::UInt32(v) => *v == rhs.as_uint32(),
 
        Value::UInt64(v) => *v == rhs.as_uint64(),
 
        Value::SInt8(v) => *v == rhs.as_sint8(),
 
        Value::SInt16(v) => *v == rhs.as_sint16(),
 
        Value::SInt32(v) => *v == rhs.as_sint32(),
 
        Value::SInt64(v) => *v == rhs.as_sint64(),
 
        Value::Array(lhs_pos) => eval_equality_heap(store, *lhs_pos, rhs.as_array()),
 
        Value::Enum(v) => *v == rhs.as_enum(),
 
        Value::Union(lhs_tag, lhs_pos) => {
 
            let (rhs_tag, rhs_pos) = rhs.as_union();
 
            if *lhs_tag != rhs_tag {
 
                return false;
 
            }
 
            eval_equality_heap(store, *lhs_pos, rhs_pos)
 
        },
 
        Value::Struct(lhs_pos) => eval_equality_heap(store, *lhs_pos, rhs.as_struct()),
 
        _ => unreachable!("apply_equality_operator to lhs {:?}", lhs),
 
    }
 
}
 

	
 
/// Recursively checks for inequality
 
pub(crate) fn apply_inequality_operator(store: &Store, lhs: &Value, rhs: &Value) -> bool {
 
    let lhs = store.maybe_read_ref(lhs);
 
    let rhs = store.maybe_read_ref(rhs);
 

	
 
    fn eval_inequality_heap(store: &Store, lhs_pos: HeapPos, rhs_pos: HeapPos) -> bool {
 
        let lhs_vals = &store.heap_regions[lhs_pos as usize].values;
 
        let rhs_vals = &store.heap_regions[rhs_pos as usize].values;
 
        let lhs_len = lhs_vals.len();
 
        if lhs_len != rhs_vals.len() {
 
            return true;
 
        }
 

	
 
        for idx in 0..lhs_len {
 
            let lhs_val = &lhs_vals[idx];
 
            let rhs_val = &rhs_vals[idx];
 
            if apply_inequality_operator(store, lhs_val, rhs_val) {
 
                return true;
 
            }
 
        }
 

	
 
        return false;
 
    }
 

	
 
    match lhs {
 
        Value::Input(v) => *v != rhs.as_input(),
 
        Value::Output(v) => *v != rhs.as_output(),
 
        Value::Message(lhs_pos) => eval_inequality_heap(store, *lhs_pos, rhs.as_message()),
src/protocol/mod.rs
Show inline comments
 
mod arena;
 
pub(crate) mod eval;
 
pub(crate) mod input_source;
 
mod parser;
 
#[cfg(test)] mod tests;
 

	
 
pub(crate) mod ast;
 
pub(crate) mod ast_printer;
 

	
 
use std::sync::Mutex;
 

	
 
use crate::collections::{StringPool, StringRef};
 
use crate::common::*;
 
use crate::protocol::ast::*;
 
use crate::protocol::eval::*;
 
use crate::protocol::input_source::*;
 
use crate::protocol::parser::*;
 
use crate::protocol::type_table::*;
 

	
 
/// A protocol description module
 
pub struct Module {
 
    pub(crate) source: InputSource,
 
    pub(crate) root_id: RootId,
 
    pub(crate) name: Option<StringRef<'static>>,
 
}
 
/// Description of a protocol object, used to configure new connectors.
 
#[repr(C)]
 
pub struct ProtocolDescription {
 
    pub(crate) modules: Vec<Module>,
 
    pub(crate) heap: Heap,
 
    pub(crate) types: TypeTable,
 
    pub(crate) pool: Mutex<StringPool>,
 
}
 
#[derive(Debug, Clone)]
 
pub(crate) struct ComponentState {
 
    pub(crate) prompt: Prompt,
 
}
 

	
 
#[allow(dead_code)]
 
pub(crate) enum EvalContext<'a> {
 
    Nonsync(&'a mut NonsyncProtoContext<'a>),
 
    Sync(&'a mut SyncProtoContext<'a>),
 
    None,
 
}
 
//////////////////////////////////////////////
 

	
 
#[derive(Debug)]
 
pub enum ComponentCreationError {
 
    ModuleDoesntExist,
 
    DefinitionDoesntExist,
 
    DefinitionNotComponent,
 
    InvalidNumArguments,
 
    InvalidArgumentType(usize),
 
    UnownedPort,
 
    InSync,
 
}
 

	
 
impl std::fmt::Debug for ProtocolDescription {
 
    fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
 
        write!(f, "(An opaque protocol description)")
 
    }
 
}
 
impl ProtocolDescription {
 
    // TODO: Allow for multi-file compilation
 
    pub fn parse(buffer: &[u8]) -> Result<Self, String> {
 
        // TODO: @fixme, keep code compilable, but needs support for multiple
 
        //  input files.
 
        let source = InputSource::new(String::new(), Vec::from(buffer));
 
        let mut parser = Parser::new();
 
        parser.feed(source).expect("failed to feed source");
 
        
 
        if let Err(err) = parser.parse() {
 
            println!("ERROR:\n{}", err);
 
            return Err(format!("{}", err))
 
        }
 

	
 
        debug_assert_eq!(parser.modules.len(), 1, "only supporting one module here for now");
 
        let modules: Vec<Module> = parser.modules.into_iter()
 
            .map(|module| Module{
 
                source: module.source,
 
                root_id: module.root_id,
 
                name: module.name.map(|(_, name)| name)
 
            })
 
            .collect();
 

	
 
        return Ok(ProtocolDescription {
 
            modules,
 
            heap: parser.heap,
 
            types: parser.type_table,
 
            pool: Mutex::new(parser.string_pool),
 
        });
 
    }
 

	
 
    #[deprecated]
 
    pub(crate) fn component_polarities(
 
        &self,
 
        module_name: &[u8],
 
        identifier: &[u8],
 
    ) -> Result<Vec<Polarity>, AddComponentError> {
 
        use AddComponentError::*;
 

	
 
        let module_root = self.lookup_module_root(module_name);
 
        if module_root.is_none() {
 
            return Err(AddComponentError::NoSuchModule);
 
        }
 
        let module_root = module_root.unwrap();
 

	
 
        let root = &self.heap[module_root];
 
        let def = root.get_definition_ident(&self.heap, identifier);
 
        if def.is_none() {
 
            return Err(NoSuchComponent);
 
        }
 

	
 
        let def = &self.heap[def.unwrap()];
 
        if !def.is_component() {
 
            return Err(NoSuchComponent);
 
        }
 

	
 
        for &param in def.parameters().iter() {
 
            let param = &self.heap[param];
 
            let first_element = &param.parser_type.elements[0];
 

	
 
            match first_element.variant {
 
                ParserTypeVariant::Input | ParserTypeVariant::Output => continue,
 
                _ => {
 
                    return Err(NonPortTypeParameters);
 
                }
 
            }
 
        }
 

	
 
        let mut result = Vec::new();
 
        for &param in def.parameters().iter() {
 
            let param = &self.heap[param];
 
            let first_element = &param.parser_type.elements[0];
 

	
 
            if first_element.variant == ParserTypeVariant::Input {
 
                result.push(Polarity::Getter)
 
            } else if first_element.variant == ParserTypeVariant::Output {
 
                result.push(Polarity::Putter)
 
            } else {
 
                unreachable!()
 
            }
 
        }
 
        Ok(result)
 
    }
 

	
 
    // expects port polarities to be correct
 
    #[deprecated]
 
    pub(crate) fn new_component(&self, module_name: &[u8], identifier: &[u8], ports: &[PortId]) -> ComponentState {
 
        let mut args = Vec::new();
 
        for (&x, y) in ports.iter().zip(self.component_polarities(module_name, identifier).unwrap()) {
 
            match y {
 
                Polarity::Getter => args.push(Value::Input(x)),
 
                Polarity::Putter => args.push(Value::Output(x)),
 
            }
 
        }
 

	
 
        let module_root = self.lookup_module_root(module_name).unwrap();
 
        let root = &self.heap[module_root];
 
        let def = root.get_definition_ident(&self.heap, identifier).unwrap();
 
        // TODO: Check for polymorph
 

	
 
        ComponentState { prompt: Prompt::new(&self.types, &self.heap, def, 0, ValueGroup::new_stack(args)) }
 
    }
 

	
 
    // TODO: Ofcourse, rename this at some point, perhaps even remove it in its
 
    //  entirety. Find some way to interface with the parameter's types.
 
    pub(crate) fn new_component_v2(
 
        &self, module_name: &[u8], identifier: &[u8], arguments: ValueGroup
 
    ) -> Result<Prompt, ComponentCreationError> {
 
        // Find the module in which the definition can be found
 
        let module_root = self.lookup_module_root(module_name);
 
        if module_root.is_none() {
 
            return Err(ComponentCreationError::ModuleDoesntExist);
 
        }
 
        let module_root = module_root.unwrap();
 

	
 
        let root = &self.heap[module_root];
 
        let definition_id = root.get_definition_ident(&self.heap, identifier);
 
        if definition_id.is_none() {
 
            return Err(ComponentCreationError::DefinitionDoesntExist);
 
        }
 
        let definition_id = definition_id.unwrap();
 

	
 
        let definition = &self.heap[definition_id];
 
        if !definition.is_component() {
 
            return Err(ComponentCreationError::DefinitionNotComponent);
 
        }
 

	
 
        // Make sure that the types of the provided value group matches that of
 
        // the expected types.
 
        let definition = definition.as_component();
 
        if !definition.poly_vars.is_empty() {
 
            return Err(ComponentCreationError::DefinitionNotComponent);
 
        }
 

	
 
        // - check number of arguments
 
        let expr_data = self.types.get_procedure_expression_data(&definition_id, 0);
 
        if expr_data.arg_types.len() != arguments.values.len() {
 
            return Err(ComponentCreationError::InvalidNumArguments);
 
        }
 

	
 
        // - for each argument try to make sure the types match
 
        for arg_idx in 0..arguments.values.len() {
 
            let expected_type = &expr_data.arg_types[arg_idx];
 
            let provided_value = &arguments.values[arg_idx];
 
            if !self.verify_same_type(expected_type, 0, &arguments, provided_value) {
 
                return Err(ComponentCreationError::InvalidArgumentType(arg_idx));
 
            }
 
        }
 

	
 
        // By now we're sure that all of the arguments are correct. So create
 
        // the connector.
 
        return Ok(Prompt::new(&self.types, &self.heap, definition_id, 0, arguments));
 
    }
 

	
 
    fn lookup_module_root(&self, module_name: &[u8]) -> Option<RootId> {
 
        for module in self.modules.iter() {
 
            match &module.name {
 
                Some(name) => if name.as_bytes() == module_name {
 
                    return Some(module.root_id);
 
                },
 
                None => if module_name.is_empty() {
 
                    return Some(module.root_id);
 
                }
 
            }
 
        }
 

	
 
        return None;
 
    }
 

	
 
    fn verify_same_type(&self, expected: &ConcreteType, expected_idx: usize, arguments: &ValueGroup, argument: &Value) -> bool {
 
        use ConcreteTypePart as CTP;
 

	
 
        match &expected.parts[expected_idx] {
 
            CTP::Void | CTP::Message | CTP::Slice | CTP::Function(_, _) | CTP::Component(_, _) => unreachable!(),
 
            CTP::Bool => if let Value::Bool(_) = argument { true } else { false },
 
            CTP::UInt8 => if let Value::UInt8(_) = argument { true } else { false },
 
            CTP::UInt16 => if let Value::UInt16(_) = argument { true } else { false },
 
            CTP::UInt32 => if let Value::UInt32(_) = argument { true } else { false },
 
            CTP::UInt64 => if let Value::UInt64(_) = argument { true } else { false },
 
            CTP::SInt8 => if let Value::SInt8(_) = argument { true } else { false },
 
            CTP::SInt16 => if let Value::SInt16(_) = argument { true } else { false },
 
            CTP::SInt32 => if let Value::SInt32(_) = argument { true } else { false },
 
            CTP::SInt64 => if let Value::SInt64(_) = argument { true } else { false },
 
            CTP::Character => if let Value::Char(_) = argument { true } else { false },
 
            CTP::String => {
 
                // Match outer string type and embedded character types
 
                if let Value::String(heap_pos) = argument {
 
                    for element in &arguments.regions[*heap_pos as usize] {
 
                        if let Value::Char(_) = element {} else {
 
                            return false;
 
                        }
 
                    }
 
                } else {
 
                    return false;
 
                }
 

	
 
                return true;
 
            },
 
            CTP::Array => {
 
                if let Value::Array(heap_pos) = argument {
 
                    let heap_pos = *heap_pos;
 
                    for element in &arguments.regions[heap_pos as usize] {
 
                        if !self.verify_same_type(expected, expected_idx + 1, arguments, element) {
 
                            return false;
 
                        }
 
                    }
 
                    return true;
 
                } else {
 
                    return false;
 
                }
 
            },
 
            CTP::Input => if let Value::Input(_) = argument { true } else { false },
 
            CTP::Output => if let Value::Output(_) = argument { true } else { false },
 
            CTP::Instance(definition_id, _num_embedded) => {
 
                let definition = self.types.get_base_definition(definition_id).unwrap();
 
                match &definition.definition {
 
                    DefinedTypeVariant::Enum(definition) => {
 
                        if let Value::Enum(variant_value) = argument {
 
                            let is_valid = definition.variants.iter()
 
                                .any(|v| v.value == *variant_value);
 
                            return is_valid;
 
                        }
 
                    },
 
                    _ => todo!("implement full type checking on user-supplied arguments"),
 
                }
 

	
 
                return false;
 
            },
 
        }
 
    }
 
}
 

	
 
// TODO: @temp Should just become a concrete thing that is passed in
 
pub trait RunContext {
 
    fn performed_put(&mut self, port: PortId) -> bool;
 
    fn performed_get(&mut self, port: PortId) -> Option<ValueGroup>; // None if still waiting on message
 
    fn fires(&mut self, port: PortId) -> Option<Value>; // None if not yet branched
 
    fn performed_fork(&mut self) -> Option<bool>; // None if not yet forked
 
    fn created_channel(&mut self) -> Option<(Value, Value)>; // None if not yet prepared
 
}
 

	
 
#[derive(Debug)]
 
pub enum RunResult {
 
    // Can only occur outside sync blocks
 
    ComponentTerminated, // component has exited its procedure
 
    ComponentAtSyncStart,
 
    NewComponent(DefinitionId, i32, ValueGroup), // should also be possible inside sync
 
    NewChannel, // should also be possible inside sync
 
    // Can only occur inside sync blocks
 
    BranchInconsistent, // branch has inconsistent behaviour
 
    BranchMissingPortState(PortId), // branch doesn't know about port firing
 
    BranchGet(PortId), // branch hasn't received message on input port yet
 
    BranchAtSyncEnd,
 
    BranchFork,
 
    BranchPut(PortId, ValueGroup),
 
}
 

	
 
impl ComponentState {
 
    pub(crate) fn run(&mut self, ctx: &mut impl RunContext, pd: &ProtocolDescription) -> RunResult {
 
        use EvalContinuation as EC;
 
        use RunResult as RR;
 

	
 
        loop {
 
            let step_result = self.prompt.step(&pd.types, &pd.heap, &pd.modules, ctx);
 
            match step_result {
 
                Err(reason) => {
 
                    // TODO: @temp
 
                    println!("Evaluation error:\n{}", reason);
 
                    todo!("proper error handling/bubbling up");
 
                },
 
                Ok(continuation) => match continuation {
 
                    // TODO: Probably want to remove this translation
 
                    EC::Stepping => continue,
 
                    EC::BranchInconsistent => return RR::BranchInconsistent,
 
                    EC::ComponentTerminated => return RR::ComponentTerminated,
 
                    EC::SyncBlockStart => return RR::ComponentAtSyncStart,
 
                    EC::SyncBlockEnd => return RR::BranchAtSyncEnd,
 
                    EC::NewComponent(definition_id, monomorph_idx, args) =>
 
                        return RR::NewComponent(definition_id, monomorph_idx, args),
 
                    EC::NewChannel =>
 
                        return RR::NewChannel,
 
                    EC::NewFork =>
 
                        return RR::BranchFork,
 
                    EC::BlockFires(port_id) => return RR::BranchMissingPortState(port_id),
 
                    EC::BlockGet(port_id) => return RR::BranchGet(port_id),
 
                    EC::Put(port_id, value_group) => {
 
                        return RR::BranchPut(port_id, value_group);
 
                    },
 
                }
 
            }
 
        }
 
    }
 
}
 

	
 
// TODO: @remove the old stuff
 
impl ComponentState {
 
    pub(crate) fn nonsync_run<'a: 'b, 'b>(
 
        &'a mut self,
 
        context: &'b mut NonsyncProtoContext<'b>,
 
        pd: &'a ProtocolDescription,
 
    ) -> NonsyncBlocker {
 
        let mut context = EvalContext::Nonsync(context);
 
        loop {
 
            let result = self.prompt.step(&pd.types, &pd.heap, &pd.modules, &mut context);
 
            match result {
 
                Err(err) => {
 
                    println!("Evaluation error:\n{}", err);
 
                    panic!("proper error handling when component fails");
 
                },
 
                Ok(cont) => match cont {
 
                    EvalContinuation::Stepping => continue,
 
                    EvalContinuation::BranchInconsistent => return NonsyncBlocker::Inconsistent,
 
                    EvalContinuation::ComponentTerminated => return NonsyncBlocker::ComponentExit,
 
                    EvalContinuation::SyncBlockStart => return NonsyncBlocker::SyncBlockStart,
 
                    // Not possible to end sync block if never entered one
 
                    EvalContinuation::SyncBlockEnd => unreachable!(),
 
                    EvalContinuation::NewComponent(definition_id, monomorph_idx, args) => {
 
                        // Look up definition (TODO for now, assume it is a definition)
 
                        // Look up definition
 
                        let mut moved_ports = HashSet::new();
 
                        for arg in args.values.iter() {
 
                            match arg {
 
                                Value::Output(port) => {
 
                                    moved_ports.insert(*port);
 
                                }
 
                                Value::Input(port) => {
 
                                    moved_ports.insert(*port);
 
                                }
 
                                _ => {}
 
                            }
 
                        }
 
                        for region in args.regions.iter() {
 
                            for arg in region {
 
                                match arg {
 
                                    Value::Output(port) => { moved_ports.insert(*port); },
 
                                    Value::Input(port) => { moved_ports.insert(*port); },
 
                                    _ => {},
 
                                }
 
                            }
 
                        }
 
                        let init_state = ComponentState { prompt: Prompt::new(&pd.types, &pd.heap, definition_id, monomorph_idx, args) };
 
                        context.new_component(moved_ports, init_state);
 
                        // Continue stepping
 
                        continue;
 
                    },
 
                    EvalContinuation::NewChannel => {
 
                        // Because of the way we emulate the old context for now, we can safely
 
                        // assume that this will never happen. The old context thingamajig always
 
                        // creates a channel, it never bubbles a "need to create a channel" message
 
                        // to the runtime
 
                        unreachable!();
 
                    },
 
                    EvalContinuation::NewFork => unreachable!(),
 
                    // Outside synchronous blocks, no fires/get/put happens
 
                    EvalContinuation::BlockFires(_) => unreachable!(),
 
                    EvalContinuation::BlockGet(_) => unreachable!(),
 
                    EvalContinuation::Put(_, _) => unreachable!(),
 
                },
 
            }
 
        }
 
    }
 

	
 
    pub(crate) fn sync_run<'a: 'b, 'b>(
 
        &'a mut self,
 
        context: &'b mut SyncProtoContext<'b>,
 
        pd: &'a ProtocolDescription,
 
    ) -> SyncBlocker {
 
        let mut context = EvalContext::Sync(context);
 
        loop {
 
            let result = self.prompt.step(&pd.types, &pd.heap, &pd.modules, &mut context);
 
            match result {
 
                Err(err) => {
 
                    println!("Evaluation error:\n{}", err);
 
                    panic!("proper error handling when component fails");
 
                },
 
                Ok(cont) => match cont {
 
                    EvalContinuation::Stepping => continue,
 
                    EvalContinuation::BranchInconsistent => return SyncBlocker::Inconsistent,
 
                    // First need to exit synchronous block before definition may end
 
                    EvalContinuation::ComponentTerminated => unreachable!(),
 
                    // No nested synchronous blocks
 
                    EvalContinuation::SyncBlockStart => unreachable!(),
 
                    EvalContinuation::SyncBlockEnd => return SyncBlocker::SyncBlockEnd,
 
                    // Not possible to create component in sync block
 
                    EvalContinuation::NewComponent(_, _, _) => unreachable!(),
 
                    EvalContinuation::NewChannel => unreachable!(),
 
                    EvalContinuation::NewFork => unreachable!(),
 
                    EvalContinuation::BlockFires(port) => {
 
                        return SyncBlocker::CouldntCheckFiring(port);
 
                    },
 
                    EvalContinuation::BlockGet(port) => {
 
                        return SyncBlocker::CouldntReadMsg(port);
 
                    },
 
                    EvalContinuation::Put(port, message) => {
 
                        let payload;
 

	
 
                        // Extract bytes from `put`
 
                        match &message.values[0] {
 
                            Value::Null => {
 
                                return SyncBlocker::Inconsistent;
 
                            },
 
                            Value::Message(heap_pos) => {
 
                                // Create a copy of the payload
 
                                let values = &message.regions[*heap_pos as usize];
 
                                let mut bytes = Vec::with_capacity(values.len());
 
                                for value in values {
 
                                    bytes.push(value.as_uint8());
 
                                }
 
                                payload = Payload(Arc::new(bytes));
 
                            }
 
                            _ => unreachable!(),
 
                        }
 
                        return SyncBlocker::PutMsg(port, payload);
 
                    }
 
                },
 
            }
 
        }
 
    }
 
}
 

	
 
impl RunContext for EvalContext<'_> {
 
    fn performed_put(&mut self, port: PortId) -> bool {
 
        match self {
 
            EvalContext::None => unreachable!(),
 
            EvalContext::Nonsync(_) => unreachable!(),
 
            EvalContext::Sync(ctx) => {
 
                ctx.did_put_or_get(port)
 
            }
 
        }
 
    }
 

	
 
    fn performed_get(&mut self, port: PortId) -> Option<ValueGroup> {
 
        match self {
 
            EvalContext::None => unreachable!(),
 
            EvalContext::Nonsync(_) => unreachable!(),
 
            EvalContext::Sync(ctx) => {
 
                let payload = ctx.read_msg(port);
 
                if payload.is_none() {
 
                    return None;
 
                }
 

	
 
                let payload = payload.unwrap();
 
                let mut transformed = Vec::with_capacity(payload.len());
 
                for byte in payload.0.iter() {
 
                    transformed.push(Value::UInt8(*byte));
 
                }
 

	
 
                let value_group = ValueGroup{
 
                    values: vec![Value::Message(0)],
 
                    regions: vec![transformed],
 
                };
 

	
 
                return Some(value_group);
 
            }
 
        }
 
    }
 

	
 
    fn fires(&mut self, port: PortId) -> Option<Value> {
 
        match self {
 
            EvalContext::None => unreachable!(),
 
            EvalContext::Nonsync(_) => unreachable!(),
 
            EvalContext::Sync(context) => {
 
                match context.is_firing(port) {
 
                    Some(did_fire) => Some(Value::Bool(did_fire)),
 
                    None => None,
 
                }
 
            }
 
        }
 
    }
 

	
 
    fn created_channel(&mut self) -> Option<(Value, Value)> {
 
        match self {
 
            EvalContext::None => unreachable!(),
 
            EvalContext::Nonsync(context) => {
 
                let [from, to] = context.new_port_pair();
 
                let from = Value::Output(from);
 
                let to = Value::Input(to);
 
                return Some((from, to));
 
            },
 
            EvalContext::Sync(_) => unreachable!(),
 
        }
 
    }
 

	
 
    fn performed_fork(&mut self) -> Option<bool> {
 
        // Never actually used in the old runtime
 
        return None;
 
    }
 
}
 

	
 
// TODO: @remove once old runtime has disappeared
 
impl EvalContext<'_> {
 
    // fn random(&mut self) -> LongValue {
 
    //     match self {
 
    //         // EvalContext::None => unreachable!(),
 
    //         EvalContext::Nonsync(_context) => todo!(),
 
    //         EvalContext::Sync(_) => unreachable!(),
 
    //     }
 
    // }
 
    fn new_component(&mut self, moved_ports: HashSet<PortId>, init_state: ComponentState) -> () {
 
        match self {
 
            EvalContext::None => unreachable!(),
 
            EvalContext::Nonsync(context) => {
 
                context.new_component(moved_ports, init_state)
 
            }
 
            EvalContext::Sync(_) => unreachable!(),
 
        }
 
    }
 
    fn new_channel(&mut self) -> [Value; 2] {
 
        match self {
 
            EvalContext::None => unreachable!(),
 
            EvalContext::Nonsync(context) => {
 
                let [from, to] = context.new_port_pair();
 
                let from = Value::Output(from);
 
                let to = Value::Input(to);
 
                return [from, to];
 
            }
 
            EvalContext::Sync(_) => unreachable!(),
 
        }
 
    }
 
    fn fires(&mut self, port: Value) -> Option<Value> {
 
        match self {
 
            EvalContext::None => unreachable!(),
 
            EvalContext::Nonsync(_) => unreachable!(),
 
            EvalContext::Sync(context) => match port {
 
                Value::Output(port) => context.is_firing(port).map(Value::Bool),
 
                Value::Input(port) => context.is_firing(port).map(Value::Bool),
 
                _ => unreachable!(),
 
            },
 
        }
 
    }
 
    fn get(&mut self, port: Value, store: &mut Store) -> Option<Value> {
 
        match self {
 
            EvalContext::None => unreachable!(),
 
            EvalContext::Nonsync(_) => unreachable!(),
 
            EvalContext::Sync(context) => match port {
 
                Value::Input(port) => {
 
                    let payload = context.read_msg(port);
 
                    if payload.is_none() { return None; }
 

	
 
                    let heap_pos = store.alloc_heap();
 
                    let heap_pos_usize = heap_pos as usize;
 
                    let payload = payload.unwrap();
 
                    store.heap_regions[heap_pos_usize].values.reserve(payload.0.len());
 
                    for value in payload.0.iter() {
 
                        store.heap_regions[heap_pos_usize].values.push(Value::UInt8(*value));
 
                    }
 

	
 
                    return Some(Value::Message(heap_pos));
 
                }
 
                _ => unreachable!(),
 
            },
 
        }
 
    }
 
    fn did_put(&mut self, port: Value) -> bool {
 
        match self {
 
            EvalContext::None => unreachable!("did_put in None context"),
 
            EvalContext::Nonsync(_) => unreachable!("did_put in nonsync context"),
 
            EvalContext::Sync(context) => match port {
 
                Value::Output(port) => {
 
                    context.did_put_or_get(port)
 
                },
 
                _ => unreachable!("did_put on non-output port value")
 
            }
 
        }
 
    }
 
}
src/protocol/parser/mod.rs
Show inline comments
 
pub(crate) mod symbol_table;
 
pub(crate) mod type_table;
 
pub(crate) mod tokens;
 
pub(crate) mod token_parsing;
 
pub(crate) mod pass_tokenizer;
 
pub(crate) mod pass_symbols;
 
pub(crate) mod pass_imports;
 
pub(crate) mod pass_definitions;
 
pub(crate) mod pass_validation_linking;
 
pub(crate) mod pass_typing;
 
mod visitor;
 

	
 
use tokens::*;
 
use crate::collections::*;
 
use visitor::Visitor;
 
use pass_tokenizer::PassTokenizer;
 
use pass_symbols::PassSymbols;
 
use pass_imports::PassImport;
 
use pass_definitions::PassDefinitions;
 
use pass_validation_linking::PassValidationLinking;
 
use pass_typing::{PassTyping, ResolveQueue};
 
use symbol_table::*;
 
use type_table::TypeTable;
 

	
 
use crate::protocol::ast::*;
 
use crate::protocol::input_source::*;
 

	
 
use crate::protocol::ast_printer::ASTWriter;
 

	
 
#[derive(Debug, PartialEq, Eq, PartialOrd, Ord)]
 
pub enum ModuleCompilationPhase {
 
    Tokenized,              // source is tokenized
 
    SymbolsScanned,         // all definitions are linked to their type class
 
    ImportsResolved,        // all imports are added to the symbol table
 
    DefinitionsParsed,      // produced the AST for the entire module
 
    TypesAddedToTable,      // added all definitions to the type table
 
    ValidatedAndLinked,     // AST is traversed and has linked the required AST nodes
 
    // When we continue with the compiler:
 
    // Typed,                  // Type inference and checking has been performed
 
}
 

	
 
pub struct Module {
 
    // Buffers
 
    pub source: InputSource,
 
    pub tokens: TokenBuffer,
 
    // Identifiers
 
    pub root_id: RootId,
 
    pub name: Option<(PragmaId, StringRef<'static>)>,
 
    pub version: Option<(PragmaId, i64)>,
 
    pub phase: ModuleCompilationPhase,
 
}
 

	
 
// TODO: This is kind of wrong. Because when we're producing bytecode we would
 
//       like the bytecode itself to not have the notion of the size of a pointer
 
//       type. But until I figure out what we do want I'll just set everything
 
//       to a 64-bit architecture.
 
pub struct TargetArch {
 
    pub array_size_alignment: (usize, usize),
 
    pub slice_size_alignment: (usize, usize),
 
    pub string_size_alignment: (usize, usize),
 
    pub port_size_alignment: (usize, usize),
 
    pub pointer_size_alignment: (usize, usize),
 
}
 

	
 
pub struct PassCtx<'a> {
 
    heap: &'a mut Heap,
 
    symbols: &'a mut SymbolTable,
 
    pool: &'a mut StringPool,
 
    arch: &'a TargetArch,
 
}
 

	
 
pub struct Parser {
 
    // Storage of all information created/gathered during compilation.
 
    pub(crate) heap: Heap,
 
    pub(crate) string_pool: StringPool, // Do not deallocate, holds all strings
 
    pub(crate) modules: Vec<Module>,
 
    pub(crate) symbol_table: SymbolTable,
 
    pub(crate) type_table: TypeTable,
 
    // Compiler passes, used as little state machine that keep their memory
 
    // around.
 
    pass_tokenizer: PassTokenizer,
 
    pass_symbols: PassSymbols,
 
    pass_import: PassImport,
 
    pass_definitions: PassDefinitions,
 
    pass_validation: PassValidationLinking,
 
    pass_typing: PassTyping,
 
    // Compiler options
 
    pub write_ast_to: Option<String>,
 
    pub(crate) arch: TargetArch,
 
}
 

	
 
impl Parser {
 
    pub fn new() -> Self {
 
        let mut parser = Parser{
 
            heap: Heap::new(),
 
            string_pool: StringPool::new(),
 
            modules: Vec::new(),
 
            symbol_table: SymbolTable::new(),
 
            type_table: TypeTable::new(),
 
            pass_tokenizer: PassTokenizer::new(),
 
            pass_symbols: PassSymbols::new(),
 
            pass_import: PassImport::new(),
 
            pass_definitions: PassDefinitions::new(),
 
            pass_validation: PassValidationLinking::new(),
 
            pass_typing: PassTyping::new(),
 
            write_ast_to: None,
 
            arch: TargetArch {
 
                array_size_alignment: (3*8, 8), // pointer, length, capacity
 
                slice_size_alignment: (2*8, 8), // pointer, length
 
                string_size_alignment: (3*8, 8), // pointer, length, capacity
 
                port_size_alignment: (3*4, 4), // two u32s: connector + port ID
 
                pointer_size_alignment: (8, 8),
 
            }
 
        };
 

	
 
        parser.symbol_table.insert_scope(None, SymbolScope::Global);
 

	
 
        fn quick_type(variants: &[ParserTypeVariant]) -> ParserType {
 
            let mut t = ParserType{ elements: Vec::with_capacity(variants.len()), full_span: InputSpan::new() };
 
            for variant in variants {
 
                t.elements.push(ParserTypeElement{ element_span: InputSpan::new(), variant: variant.clone() });
 
            }
 
            t
 
        }
 

	
 
        use ParserTypeVariant as PTV;
 
        insert_builtin_function(&mut parser, "get", &["T"], |id| (
 
            vec![
 
                ("input", quick_type(&[PTV::Input, PTV::PolymorphicArgument(id.upcast(), 0)]))
 
            ],
 
            quick_type(&[PTV::PolymorphicArgument(id.upcast(), 0)])
 
        ));
 
        insert_builtin_function(&mut parser, "put", &["T"], |id| (
 
            vec![
 
                ("output", quick_type(&[PTV::Output, PTV::PolymorphicArgument(id.upcast(), 0)])),
 
                ("value", quick_type(&[PTV::PolymorphicArgument(id.upcast(), 0)])),
 
            ],
 
            quick_type(&[PTV::Void])
 
        ));
 
        insert_builtin_function(&mut parser, "fires", &["T"], |id| (
 
            vec![
 
                ("port", quick_type(&[PTV::InputOrOutput, PTV::PolymorphicArgument(id.upcast(), 0)]))
 
            ],
 
            quick_type(&[PTV::Bool])
 
        ));
 
        insert_builtin_function(&mut parser, "create", &["T"], |id| (
 
            vec![
 
                ("length", quick_type(&[PTV::IntegerLike]))
 
            ],
 
            quick_type(&[PTV::ArrayLike, PTV::PolymorphicArgument(id.upcast(), 0)])
 
        ));
 
        insert_builtin_function(&mut parser, "length", &["T"], |id| (
 
            vec![
 
                ("array", quick_type(&[PTV::ArrayLike, PTV::PolymorphicArgument(id.upcast(), 0)]))
 
            ],
 
            quick_type(&[PTV::UInt32]) // TODO: @PtrInt
 
        ));
 
        insert_builtin_function(&mut parser, "assert", &[], |_id| (
 
            vec![
 
                ("condition", quick_type(&[PTV::Bool])),
 
            ],
 
            quick_type(&[PTV::Void])
 
        ));
 
        insert_builtin_function(&mut parser, "print", &[], |_id| (
 
            vec![
 
                ("message", quick_type(&[PTV::String])),
 
            ],
 
            quick_type(&[PTV::Void])
 
        ));
 

	
 
        parser
 
    }
 

	
 
    pub fn feed(&mut self, mut source: InputSource) -> Result<(), ParseError> {
 
        // TODO: @Optimize
 
        let mut token_buffer = TokenBuffer::new();
 
        self.pass_tokenizer.tokenize(&mut source, &mut token_buffer)?;
 

	
 
        let module = Module{
 
            source,
 
            tokens: token_buffer,
 
            root_id: RootId::new_invalid(),
 
            name: None,
 
            version: None,
 
            phase: ModuleCompilationPhase::Tokenized,
 
        };
 
        self.modules.push(module);
 

	
 
        Ok(())
 
    }
 

	
 
    pub fn parse(&mut self) -> Result<(), ParseError> {
 
        let mut pass_ctx = PassCtx{
 
            heap: &mut self.heap,
 
            symbols: &mut self.symbol_table,
 
            pool: &mut self.string_pool,
 
            arch: &self.arch,
 
        };
 

	
 
        // Advance all modules to the phase where all symbols are scanned
 
        for module_idx in 0..self.modules.len() {
 
            self.pass_symbols.parse(&mut self.modules, module_idx, &mut pass_ctx)?;
 
        }
 

	
 
        // With all symbols scanned, perform further compilation until we can
 
        // add all base types to the type table.
 
        for module_idx in 0..self.modules.len() {
 
            self.pass_import.parse(&mut self.modules, module_idx, &mut pass_ctx)?;
 
            self.pass_definitions.parse(&mut self.modules, module_idx, &mut pass_ctx)?;
 
        }
 

	
 
        // Add every known type to the type table
 
        self.type_table.build_base_types(&mut self.modules, &mut pass_ctx)?;
 

	
 
        // Continue compilation with the remaining phases now that the types
 
        // are all in the type table
 
        for module_idx in 0..self.modules.len() {
 
            let mut ctx = visitor::Ctx{
 
                heap: &mut self.heap,
 
                modules: &mut self.modules,
 
                module_idx,
 
                symbols: &mut self.symbol_table,
 
                types: &mut self.type_table,
 
                arch: &self.arch,
 
            };
 
            self.pass_validation.visit_module(&mut ctx)?;
 
        }
 

	
 
        // Perform typechecking on all modules
 
        let mut queue = ResolveQueue::new();
 
        for module_idx in 0..self.modules.len() {
 
            let mut ctx = visitor::Ctx{
 
                heap: &mut self.heap,
 
                modules: &mut self.modules,
 
                module_idx,
 
                symbols: &mut self.symbol_table,
 
                types: &mut self.type_table,
 
                arch: &self.arch,
 
            };
 
            PassTyping::queue_module_definitions(&mut ctx, &mut queue);
 
        };
 
        while !queue.is_empty() {
 
            let top = queue.pop().unwrap();
 
            let mut ctx = visitor::Ctx{
 
                heap: &mut self.heap,
 
                modules: &mut self.modules,
 
                module_idx: top.root_id.index as usize,
 
                symbols: &mut self.symbol_table,
 
                types: &mut self.type_table,
 
                arch: &self.arch,
 
            };
 
            self.pass_typing.handle_module_definition(&mut ctx, &mut queue, top)?;
 
        }
 

	
 
        // Write out desired information
 
        if let Some(filename) = &self.write_ast_to {
 
            let mut writer = ASTWriter::new();
 
            let mut file = std::fs::File::create(std::path::Path::new(filename)).unwrap();
 
            writer.write_ast(&mut file, &self.heap);
 
        }
 

	
 
        Ok(())
 
    }
 
}
 

	
 
// Note: args and return type need to be a function because we need to know the function ID.
 
fn insert_builtin_function<T: Fn(FunctionDefinitionId) -> (Vec<(&'static str, ParserType)>, ParserType)> (
 
    p: &mut Parser, func_name: &str, polymorphic: &[&str], arg_and_return_fn: T) {
 

	
 
    let mut poly_vars = Vec::with_capacity(polymorphic.len());
 
    for poly_var in polymorphic {
 
        poly_vars.push(Identifier{ span: InputSpan::new(), value: p.string_pool.intern(poly_var.as_bytes()) });
 
    }
 

	
 
    let func_ident_ref = p.string_pool.intern(func_name.as_bytes());
 
    let func_id = p.heap.alloc_function_definition(|this| FunctionDefinition{
 
        this,
 
        defined_in: RootId::new_invalid(),
 
        builtin: true,
 
        span: InputSpan::new(),
 
        identifier: Identifier{ span: InputSpan::new(), value: func_ident_ref.clone() },
 
        poly_vars,
 
        return_types: Vec::new(),
 
        parameters: Vec::new(),
 
        body: BlockStatementId::new_invalid(),
 
        num_expressions_in_body: -1,
 
    });
 

	
 
    let (args, ret) = arg_and_return_fn(func_id);
 

	
 
    let mut parameters = Vec::with_capacity(args.len());
 
    for (arg_name, arg_type) in args {
 
        let identifier = Identifier{ span: InputSpan::new(), value: p.string_pool.intern(arg_name.as_bytes()) };
 
        let param_id = p.heap.alloc_variable(|this| Variable{
 
            this,
 
            kind: VariableKind::Parameter,
 
            parser_type: arg_type.clone(),
 
            identifier,
 
            relative_pos_in_block: 0,
 
            unique_id_in_scope: 0
 
        });
 
        parameters.push(param_id);
 
    }
 

	
 
    let func = &mut p.heap[func_id];
 
    func.parameters = parameters;
 
    func.return_types.push(ret);
 

	
 
    p.symbol_table.insert_symbol(SymbolScope::Global, Symbol{
 
        name: func_ident_ref,
 
        variant: SymbolVariant::Definition(SymbolDefinition{
 
            defined_in_module: RootId::new_invalid(),
 
            defined_in_scope: SymbolScope::Global,
 
            definition_span: InputSpan::new(),
 
            identifier_span: InputSpan::new(),
 
            imported_at: None,
 
            class: DefinitionClass::Function,
 
            definition_id: func_id.upcast(),
 
        })
 
    }).unwrap();
 
}
 
\ No newline at end of file
src/protocol/parser/pass_definitions.rs
Show inline comments
 
use crate::protocol::ast::*;
 
use super::symbol_table::*;
 
use super::{Module, ModuleCompilationPhase, PassCtx};
 
use super::tokens::*;
 
use super::token_parsing::*;
 
use crate::protocol::input_source::{InputSource as InputSource, InputPosition as InputPosition, InputSpan, ParseError};
 
use crate::collections::*;
 

	
 
/// Parses all the tokenized definitions into actual AST nodes.
 
pub(crate) struct PassDefinitions {
 
    // State associated with the definition currently being processed
 
    cur_definition: DefinitionId,
 
    // Temporary buffers of various kinds
 
    buffer: String,
 
    struct_fields: ScopedBuffer<StructFieldDefinition>,
 
    enum_variants: ScopedBuffer<EnumVariantDefinition>,
 
    union_variants: ScopedBuffer<UnionVariantDefinition>,
 
    variables: ScopedBuffer<VariableId>,
 
    expressions: ScopedBuffer<ExpressionId>,
 
    statements: ScopedBuffer<StatementId>,
 
    parser_types: ScopedBuffer<ParserType>,
 
}
 

	
 
impl PassDefinitions {
 
    pub(crate) fn new() -> Self {
 
        Self{
 
            cur_definition: DefinitionId::new_invalid(),
 
            buffer: String::with_capacity(128),
 
            struct_fields: ScopedBuffer::new_reserved(128),
 
            enum_variants: ScopedBuffer::new_reserved(128),
 
            union_variants: ScopedBuffer::new_reserved(128),
 
            variables: ScopedBuffer::new_reserved(128),
 
            expressions: ScopedBuffer::new_reserved(128),
 
            statements: ScopedBuffer::new_reserved(128),
 
            parser_types: ScopedBuffer::new_reserved(128),
 
        }
 
    }
 

	
 
    pub(crate) fn parse(&mut self, modules: &mut [Module], module_idx: usize, ctx: &mut PassCtx) -> Result<(), ParseError> {
 
        let module = &modules[module_idx];
 
        let module_range = &module.tokens.ranges[0];
 
        debug_assert_eq!(module.phase, ModuleCompilationPhase::ImportsResolved);
 
        debug_assert_eq!(module_range.range_kind, TokenRangeKind::Module);
 

	
 
        // Although we only need to parse the definitions, we want to go through
 
        // code ranges as well such that we can throw errors if we get
 
        // unexpected tokens at the module level of the source.
 
        let mut range_idx = module_range.first_child_idx;
 
        loop {
 
            let range_idx_usize = range_idx as usize;
 
            let cur_range = &module.tokens.ranges[range_idx_usize];
 

	
 
            match cur_range.range_kind {
 
                TokenRangeKind::Module => unreachable!(), // should not be reachable
 
                TokenRangeKind::Pragma | TokenRangeKind::Import => {
 
                    // Already fully parsed, fall through and go to next range
 
                },
 
                TokenRangeKind::Definition | TokenRangeKind::Code => {
 
                    // Visit range even if it is a "code" range to provide
 
                    // proper error messages.
 
                    self.visit_range(modules, module_idx, ctx, range_idx_usize)?;
 
                },
 
            }
 

	
 
            if cur_range.next_sibling_idx == NO_SIBLING {
 
                break;
 
            } else {
 
                range_idx = cur_range.next_sibling_idx;
 
            }
 
        }
 

	
 
        modules[module_idx].phase = ModuleCompilationPhase::DefinitionsParsed;
 

	
 
        Ok(())
 
    }
 

	
 
    fn visit_range(
 
        &mut self, modules: &[Module], module_idx: usize, ctx: &mut PassCtx, range_idx: usize
 
    ) -> Result<(), ParseError> {
 
        let module = &modules[module_idx];
 
        let cur_range = &module.tokens.ranges[range_idx];
 
        debug_assert!(cur_range.range_kind == TokenRangeKind::Definition || cur_range.range_kind == TokenRangeKind::Code);
 

	
 
        // Detect which definition we're parsing
 
        let mut iter = module.tokens.iter_range(cur_range);
 
        loop {
 
            let next = iter.next();
 
            if next.is_none() {
 
                return Ok(())
 
            }
 

	
 
            // Token was not None, so peek_ident returns None if not an ident
 
            let ident = peek_ident(&module.source, &mut iter);
 
            match ident {
 
                Some(KW_STRUCT) => self.visit_struct_definition(module, &mut iter, ctx)?,
 
                Some(KW_ENUM) => self.visit_enum_definition(module, &mut iter, ctx)?,
 
                Some(KW_UNION) => self.visit_union_definition(module, &mut iter, ctx)?,
 
                Some(KW_FUNCTION) => self.visit_function_definition(module, &mut iter, ctx)?,
 
                Some(KW_PRIMITIVE) | Some(KW_COMPOSITE) => self.visit_component_definition(module, &mut iter, ctx)?,
 
                _ => return Err(ParseError::new_error_str_at_pos(
 
                    &module.source, iter.last_valid_pos(),
 
                    "unexpected symbol, expected a keyword marking the start of a definition"
 
                )),
 
            }
 
        }
 
    }
 

	
 
    fn visit_struct_definition(
 
        &mut self, module: &Module, iter: &mut TokenIter, ctx: &mut PassCtx
 
    ) -> Result<(), ParseError> {
 
        consume_exact_ident(&module.source, iter, KW_STRUCT)?;
 
        let (ident_text, _) = consume_ident(&module.source, iter)?;
 

	
 
        // Retrieve preallocated DefinitionId
 
        let module_scope = SymbolScope::Module(module.root_id);
 
        let definition_id = ctx.symbols.get_symbol_by_name_defined_in_scope(module_scope, ident_text)
 
            .unwrap().variant.as_definition().definition_id;
 
        self.cur_definition = definition_id;
 

	
 
        // Parse struct definition
 
        consume_polymorphic_vars_spilled(&module.source, iter, ctx)?;
 

	
 
        let mut fields_section = self.struct_fields.start_section();
 
        consume_comma_separated(
 
            TokenKind::OpenCurly, TokenKind::CloseCurly, &module.source, iter, ctx,
 
            |source, iter, ctx| {
 
                let poly_vars = ctx.heap[definition_id].poly_vars();
 

	
 
                let start_pos = iter.last_valid_pos();
 
                let parser_type = consume_parser_type(
 
                    source, iter, &ctx.symbols, &ctx.heap, poly_vars, module_scope,
 
                    definition_id, false, 0
 
                )?;
 
                let field = consume_ident_interned(source, iter, ctx)?;
 
                Ok(StructFieldDefinition{
 
                    span: InputSpan::from_positions(start_pos, field.span.end),
 
                    field, parser_type
 
                })
 
            },
 
            &mut fields_section, "a struct field", "a list of struct fields", None
 
        )?;
 

	
 
        // Transfer to preallocated definition
 
        let struct_def = ctx.heap[definition_id].as_struct_mut();
 
        struct_def.fields = fields_section.into_vec();
 

	
 
        Ok(())
 
    }
 

	
 
    fn visit_enum_definition(
 
        &mut self, module: &Module, iter: &mut TokenIter, ctx: &mut PassCtx
 
    ) -> Result<(), ParseError> {
 
        consume_exact_ident(&module.source, iter, KW_ENUM)?;
 
        let (ident_text, _) = consume_ident(&module.source, iter)?;
 

	
 
        // Retrieve preallocated DefinitionId
 
        let module_scope = SymbolScope::Module(module.root_id);
 
        let definition_id = ctx.symbols.get_symbol_by_name_defined_in_scope(module_scope, ident_text)
 
            .unwrap().variant.as_definition().definition_id;
 
        self.cur_definition = definition_id;
 

	
 
        // Parse enum definition
 
        consume_polymorphic_vars_spilled(&module.source, iter, ctx)?;
 

	
 
        let mut enum_section = self.enum_variants.start_section();
 
        consume_comma_separated(
 
            TokenKind::OpenCurly, TokenKind::CloseCurly, &module.source, iter, ctx,
 
            |source, iter, ctx| {
 
                let identifier = consume_ident_interned(source, iter, ctx)?;
 
                let value = if iter.next() == Some(TokenKind::Equal) {
 
                    iter.consume();
 
                    let (variant_number, _) = consume_integer_literal(source, iter, &mut self.buffer)?;
 
                    EnumVariantValue::Integer(variant_number as i64) // TODO: @int
 
                } else {
 
                    EnumVariantValue::None
 
                };
 
                Ok(EnumVariantDefinition{ identifier, value })
 
            },
 
            &mut enum_section, "an enum variant", "a list of enum variants", None
 
        )?;
 

	
 
        // Transfer to definition
 
        let enum_def = ctx.heap[definition_id].as_enum_mut();
 
        enum_def.variants = enum_section.into_vec();
 

	
 
        Ok(())
 
    }
 

	
 
    fn visit_union_definition(
 
        &mut self, module: &Module, iter: &mut TokenIter, ctx: &mut PassCtx
 
    ) -> Result<(), ParseError> {
 
        consume_exact_ident(&module.source, iter, KW_UNION)?;
 
        let (ident_text, _) = consume_ident(&module.source, iter)?;
 

	
 
        // Retrieve preallocated DefinitionId
 
        let module_scope = SymbolScope::Module(module.root_id);
 
        let definition_id = ctx.symbols.get_symbol_by_name_defined_in_scope(module_scope, ident_text)
 
            .unwrap().variant.as_definition().definition_id;
 
        self.cur_definition = definition_id;
 

	
 
        // Parse union definition
 
        consume_polymorphic_vars_spilled(&module.source, iter, ctx)?;
 

	
 
        let mut variants_section = self.union_variants.start_section();
 
        consume_comma_separated(
 
            TokenKind::OpenCurly, TokenKind::CloseCurly, &module.source, iter, ctx,
 
            |source, iter, ctx| {
 
                let identifier = consume_ident_interned(source, iter, ctx)?;
 
                let mut close_pos = identifier.span.end;
 

	
 
                let mut types_section = self.parser_types.start_section();
 

	
 
                let has_embedded = maybe_consume_comma_separated(
 
                    TokenKind::OpenParen, TokenKind::CloseParen, source, iter, ctx,
 
                    |source, iter, ctx| {
 
                        let poly_vars = ctx.heap[definition_id].poly_vars();
 
                        consume_parser_type(
 
                            source, iter, &ctx.symbols, &ctx.heap, poly_vars,
 
                            module_scope, definition_id, false, 0
 
                        )
 
                    },
 
                    &mut types_section, "an embedded type", Some(&mut close_pos)
 
                )?;
 
                let value = if has_embedded {
 
                    types_section.into_vec()
 
                } else {
 
                    types_section.forget();
 
                    Vec::new()
 
                };
 

	
 
                Ok(UnionVariantDefinition{
 
                    span: InputSpan::from_positions(identifier.span.begin, close_pos),
 
                    identifier,
 
                    value
 
                })
 
            },
 
            &mut variants_section, "a union variant", "a list of union variants", None
 
        )?;
 

	
 
        // Transfer to AST
 
        let union_def = ctx.heap[definition_id].as_union_mut();
 
        union_def.variants = variants_section.into_vec();
 

	
 
        Ok(())
 
    }
 

	
 
    fn visit_function_definition(
 
        &mut self, module: &Module, iter: &mut TokenIter, ctx: &mut PassCtx
 
    ) -> Result<(), ParseError> {
 
        // Retrieve function name
 
        consume_exact_ident(&module.source, iter, KW_FUNCTION)?;
 
        let (ident_text, _) = consume_ident(&module.source, iter)?;
 

	
 
        // Retrieve preallocated DefinitionId
 
        let module_scope = SymbolScope::Module(module.root_id);
 
        let definition_id = ctx.symbols.get_symbol_by_name_defined_in_scope(module_scope, ident_text)
 
            .unwrap().variant.as_definition().definition_id;
 
        self.cur_definition = definition_id;
 

	
 
        consume_polymorphic_vars_spilled(&module.source, iter, ctx)?;
 

	
 
        // Parse function's argument list
 
        let mut parameter_section = self.variables.start_section();
 
        consume_parameter_list(
 
            &module.source, iter, ctx, &mut parameter_section, module_scope, definition_id
 
        )?;
 
        let parameters = parameter_section.into_vec();
 

	
 
        // Consume return types
 
        consume_token(&module.source, iter, TokenKind::ArrowRight)?;
 
        let mut return_types = self.parser_types.start_section();
 
        let mut open_curly_pos = iter.last_valid_pos(); // bogus value
 
        consume_comma_separated_until(
 
            TokenKind::OpenCurly, &module.source, iter, ctx,
 
            |source, iter, ctx| {
 
                let poly_vars = ctx.heap[definition_id].poly_vars();
 
                consume_parser_type(source, iter, &ctx.symbols, &ctx.heap, poly_vars, module_scope, definition_id, false, 0)
 
            },
 
            &mut return_types, "a return type", Some(&mut open_curly_pos)
 
        )?;
 
        let return_types = return_types.into_vec();
 

	
 
        // TODO: @ReturnValues
 
        match return_types.len() {
 
            0 => return Err(ParseError::new_error_str_at_pos(&module.source, open_curly_pos, "expected a return type")),
 
            1 => {},
 
            _ => return Err(ParseError::new_error_str_at_pos(&module.source, open_curly_pos, "multiple return types are not (yet) allowed")),
 
        }
 

	
 
        // Consume block
 
        let body = self.consume_block_statement_without_leading_curly(module, iter, ctx, open_curly_pos)?;
 

	
 
        // Assign everything in the preallocated AST node
 
        let function = ctx.heap[definition_id].as_function_mut();
 
        function.return_types = return_types;
 
        function.parameters = parameters;
 
        function.body = body;
 

	
 
        Ok(())
 
    }
 

	
 
    fn visit_component_definition(
 
        &mut self, module: &Module, iter: &mut TokenIter, ctx: &mut PassCtx
 
    ) -> Result<(), ParseError> {
 
        // Consume component variant and name
 
        let (_variant_text, _) = consume_any_ident(&module.source, iter)?;
 
        debug_assert!(_variant_text == KW_PRIMITIVE || _variant_text == KW_COMPOSITE);
 
        let (ident_text, _) = consume_ident(&module.source, iter)?;
 

	
 
        // Retrieve preallocated definition
 
        let module_scope = SymbolScope::Module(module.root_id);
 
        let definition_id = ctx.symbols.get_symbol_by_name_defined_in_scope(module_scope, ident_text)
 
            .unwrap().variant.as_definition().definition_id;
 
        self.cur_definition = definition_id;
 

	
 
        consume_polymorphic_vars_spilled(&module.source, iter, ctx)?;
 

	
 
        // Parse component's argument list
 
        let mut parameter_section = self.variables.start_section();
 
        consume_parameter_list(
 
            &module.source, iter, ctx, &mut parameter_section, module_scope, definition_id
 
        )?;
 
        let parameters = parameter_section.into_vec();
 

	
 
        // Consume block
 
        let body = self.consume_block_statement(module, iter, ctx)?;
 

	
 
        // Assign everything in the AST node
 
        let component = ctx.heap[definition_id].as_component_mut();
 
        component.parameters = parameters;
 
        component.body = body;
 

	
 
        Ok(())
 
    }
 

	
 
    /// Consumes a block statement. If the resulting statement is not a block
 
    /// (e.g. for a shorthand "if (expr) single_statement") then it will be
 
    /// wrapped in one
 
    fn consume_block_or_wrapped_statement(
 
        &mut self, module: &Module, iter: &mut TokenIter, ctx: &mut PassCtx
 
    ) -> Result<BlockStatementId, ParseError> {
 
        if Some(TokenKind::OpenCurly) == iter.next() {
 
            // This is a block statement
 
            self.consume_block_statement(module, iter, ctx)
 
        } else {
 
            // Not a block statement, so wrap it in one
 
            let mut statements = self.statements.start_section();
 
            let wrap_begin_pos = iter.last_valid_pos();
 
            self.consume_statement(module, iter, ctx, &mut statements)?;
 
            let wrap_end_pos = iter.last_valid_pos();
 

	
 
            let statements = statements.into_vec();
 

	
 
            let id = ctx.heap.alloc_block_statement(|this| BlockStatement{
 
                this,
 
                is_implicit: true,
 
                span: InputSpan::from_positions(wrap_begin_pos, wrap_end_pos),
 
                statements,
 
                end_block: EndBlockStatementId::new_invalid(),
 
                scope_node: ScopeNode::new_invalid(),
 
                first_unique_id_in_scope: -1,
 
                next_unique_id_in_scope: -1,
 
                relative_pos_in_parent: 0,
 
                locals: Vec::new(),
 
                labels: Vec::new(),
 
                next: StatementId::new_invalid(),
 
            });
 

	
 
            let end_block = ctx.heap.alloc_end_block_statement(|this| EndBlockStatement{
 
                this, start_block: id, next: StatementId::new_invalid()
 
            });
 

	
 
            let block_stmt = &mut ctx.heap[id];
 
            block_stmt.end_block = end_block;
 

	
 
            Ok(id)
 
        }
 
    }
 

	
 
    /// Consumes a statement and returns a boolean indicating whether it was a
 
    /// block or not.
 
    fn consume_statement(
 
        &mut self, module: &Module, iter: &mut TokenIter, ctx: &mut PassCtx, section: &mut ScopedSection<StatementId>
 
    ) -> Result<(), ParseError> {
 
        let next = iter.next().expect("consume_statement has a next token");
 

	
 
        if next == TokenKind::OpenCurly {
 
            let id = self.consume_block_statement(module, iter, ctx)?;
 
            section.push(id.upcast());
 
        } else if next == TokenKind::Ident {
 
            let ident = peek_ident(&module.source, iter).unwrap();
 
            if ident == KW_STMT_IF {
 
                // Consume if statement and place end-if statement directly
 
                // after it.
 
                let id = self.consume_if_statement(module, iter, ctx)?;
 
                section.push(id.upcast());
 

	
 
                let end_if = ctx.heap.alloc_end_if_statement(|this| EndIfStatement{
 
                    this, start_if: id, next: StatementId::new_invalid()
 
                });
 
                section.push(end_if.upcast());
 

	
 
                let if_stmt = &mut ctx.heap[id];
 
                if_stmt.end_if = end_if;
 
            } else if ident == KW_STMT_WHILE {
 
                let id = self.consume_while_statement(module, iter, ctx)?;
 
                section.push(id.upcast());
 

	
 
                let end_while = ctx.heap.alloc_end_while_statement(|this| EndWhileStatement{
 
                    this, start_while: id, next: StatementId::new_invalid()
 
                });
 
                section.push(end_while.upcast());
 

	
 
                let while_stmt = &mut ctx.heap[id];
 
                while_stmt.end_while = end_while;
 
            } else if ident == KW_STMT_BREAK {
 
                let id = self.consume_break_statement(module, iter, ctx)?;
 
                section.push(id.upcast());
 
            } else if ident == KW_STMT_CONTINUE {
 
                let id = self.consume_continue_statement(module, iter, ctx)?;
 
                section.push(id.upcast());
 
            } else if ident == KW_STMT_SYNC {
 
                let id = self.consume_synchronous_statement(module, iter, ctx)?;
 
                section.push(id.upcast());
 

	
 
                let end_sync = ctx.heap.alloc_end_synchronous_statement(|this| EndSynchronousStatement {
 
                    this,
 
                    start_sync: id,
 
                    next: StatementId::new_invalid()
 
                });
 
                section.push(end_sync.upcast());
 

	
 
                let sync_stmt = &mut ctx.heap[id];
 
                sync_stmt.end_sync = end_sync;
 
            } else if ident == KW_STMT_FORK {
 
                let id = self.consume_fork_statement(module, iter, ctx)?;
 
                section.push(id.upcast());
 

	
 
                let end_fork = ctx.heap.alloc_end_fork_statement(|this| EndForkStatement{
 
                    this,
 
                    start_fork: id,
 
                    next: StatementId::new_invalid(),
 
                });
 
                section.push(end_fork.upcast());
 

	
 
                let fork_stmt = &mut ctx.heap[id];
 
                fork_stmt.end_fork = end_fork;
 
            } else if ident == KW_STMT_RETURN {
 
                let id = self.consume_return_statement(module, iter, ctx)?;
 
                section.push(id.upcast());
 
            } else if ident == KW_STMT_GOTO {
 
                let id = self.consume_goto_statement(module, iter, ctx)?;
 
                section.push(id.upcast());
 
            } else if ident == KW_STMT_NEW {
 
                let id = self.consume_new_statement(module, iter, ctx)?;
 
                section.push(id.upcast());
 
            } else if ident == KW_STMT_CHANNEL {
 
                let id = self.consume_channel_statement(module, iter, ctx)?;
 
                section.push(id.upcast().upcast());
 
            } else if iter.peek() == Some(TokenKind::Colon) {
 
                self.consume_labeled_statement(module, iter, ctx, section)?;
 
            } else {
 
                // Two fallback possibilities: the first one is a memory
 
                // declaration, the other one is to parse it as a regular
 
                // expression. This is a bit ugly
 
                if let Some((memory_stmt_id, assignment_stmt_id)) = self.maybe_consume_memory_statement(module, iter, ctx)? {
 
                    section.push(memory_stmt_id.upcast().upcast());
 
                    section.push(assignment_stmt_id.upcast());
 
                } else {
 
                    let id = self.consume_expression_statement(module, iter, ctx)?;
 
                    section.push(id.upcast());
 
                }
 
            }
 
        } else {
 
            let id = self.consume_expression_statement(module, iter, ctx)?;
 
            section.push(id.upcast());
 
        }
 

	
 
        return Ok(());
 
    }
 

	
 
    fn consume_block_statement(
 
        &mut self, module: &Module, iter: &mut TokenIter, ctx: &mut PassCtx
 
    ) -> Result<BlockStatementId, ParseError> {
 
        let open_span = consume_token(&module.source, iter, TokenKind::OpenCurly)?;
 
        self.consume_block_statement_without_leading_curly(module, iter, ctx, open_span.begin)
 
    }
 

	
 
    fn consume_block_statement_without_leading_curly(
 
        &mut self, module: &Module, iter: &mut TokenIter, ctx: &mut PassCtx, open_curly_pos: InputPosition
 
    ) -> Result<BlockStatementId, ParseError> {
 
        let mut stmt_section = self.statements.start_section();
 
        let mut next = iter.next();
 
        while next != Some(TokenKind::CloseCurly) {
 
            if next.is_none() {
 
                return Err(ParseError::new_error_str_at_pos(
 
                    &module.source, iter.last_valid_pos(), "expected a statement or '}'"
 
                ));
 
            }
 
            self.consume_statement(module, iter, ctx, &mut stmt_section)?;
 
            next = iter.next();
 
        }
 

	
 
        let statements = stmt_section.into_vec();
 
        let mut block_span = consume_token(&module.source, iter, TokenKind::CloseCurly)?;
 
        block_span.begin = open_curly_pos;
 

	
 
        let id = ctx.heap.alloc_block_statement(|this| BlockStatement{
 
            this,
 
            is_implicit: false,
 
            span: block_span,
 
            statements,
 
            end_block: EndBlockStatementId::new_invalid(),
 
            scope_node: ScopeNode::new_invalid(),
 
            first_unique_id_in_scope: -1,
 
            next_unique_id_in_scope: -1,
 
            relative_pos_in_parent: 0,
 
            locals: Vec::new(),
 
            labels: Vec::new(),
 
            next: StatementId::new_invalid(),
 
        });
 

	
 
        let end_block = ctx.heap.alloc_end_block_statement(|this| EndBlockStatement{
 
            this, start_block: id, next: StatementId::new_invalid()
 
        });
 

	
 
        let block_stmt = &mut ctx.heap[id];
 
        block_stmt.end_block = end_block;
 

	
 
        Ok(id)
 
    }
 

	
 
    fn consume_if_statement(
 
        &mut self, module: &Module, iter: &mut TokenIter, ctx: &mut PassCtx
 
    ) -> Result<IfStatementId, ParseError> {
 
        let if_span = consume_exact_ident(&module.source, iter, KW_STMT_IF)?;
 
        consume_token(&module.source, iter, TokenKind::OpenParen)?;
 
        let test = self.consume_expression(module, iter, ctx)?;
 
        consume_token(&module.source, iter, TokenKind::CloseParen)?;
 
        let true_body = self.consume_block_or_wrapped_statement(module, iter, ctx)?;
 

	
 
        let false_body = if has_ident(&module.source, iter, KW_STMT_ELSE) {
 
            iter.consume();
 
            let false_body = self.consume_block_or_wrapped_statement(module, iter, ctx)?;
 
            Some(false_body)
 
        } else {
 
            None
 
        };
 

	
 
        Ok(ctx.heap.alloc_if_statement(|this| IfStatement{
 
            this,
 
            span: if_span,
 
            test,
 
            true_body,
 
            false_body,
 
            end_if: EndIfStatementId::new_invalid(),
 
        }))
 
    }
 

	
 
    fn consume_while_statement(
 
        &mut self, module: &Module, iter: &mut TokenIter, ctx: &mut PassCtx
 
    ) -> Result<WhileStatementId, ParseError> {
 
        let while_span = consume_exact_ident(&module.source, iter, KW_STMT_WHILE)?;
 
        consume_token(&module.source, iter, TokenKind::OpenParen)?;
 
        let test = self.consume_expression(module, iter, ctx)?;
 
        consume_token(&module.source, iter, TokenKind::CloseParen)?;
 
        let body = self.consume_block_or_wrapped_statement(module, iter, ctx)?;
 

	
 
        Ok(ctx.heap.alloc_while_statement(|this| WhileStatement{
 
            this,
 
            span: while_span,
 
            test,
 
            body,
 
            end_while: EndWhileStatementId::new_invalid(),
 
            in_sync: SynchronousStatementId::new_invalid(),
 
        }))
 
    }
 

	
 
    fn consume_break_statement(
 
        &mut self, module: &Module, iter: &mut TokenIter, ctx: &mut PassCtx
 
    ) -> Result<BreakStatementId, ParseError> {
 
        let break_span = consume_exact_ident(&module.source, iter, KW_STMT_BREAK)?;
 
        let label = if Some(TokenKind::Ident) == iter.next() {
 
            let label = consume_ident_interned(&module.source, iter, ctx)?;
 
            Some(label)
 
        } else {
 
            None
 
        };
 
        consume_token(&module.source, iter, TokenKind::SemiColon)?;
 
        Ok(ctx.heap.alloc_break_statement(|this| BreakStatement{
 
            this,
 
            span: break_span,
 
            label,
 
            target: None,
 
        }))
 
    }
 

	
 
    fn consume_continue_statement(
 
        &mut self, module: &Module, iter: &mut TokenIter, ctx: &mut PassCtx
 
    ) -> Result<ContinueStatementId, ParseError> {
 
        let continue_span = consume_exact_ident(&module.source, iter, KW_STMT_CONTINUE)?;
 
        let label=  if Some(TokenKind::Ident) == iter.next() {
 
            let label = consume_ident_interned(&module.source, iter, ctx)?;
 
            Some(label)
 
        } else {
 
            None
 
        };
 
        consume_token(&module.source, iter, TokenKind::SemiColon)?;
 
        Ok(ctx.heap.alloc_continue_statement(|this| ContinueStatement{
 
            this,
 
            span: continue_span,
 
            label,
 
            target: None
 
        }))
 
    }
 

	
 
    fn consume_synchronous_statement(
 
        &mut self, module: &Module, iter: &mut TokenIter, ctx: &mut PassCtx
 
    ) -> Result<SynchronousStatementId, ParseError> {
 
        let synchronous_span = consume_exact_ident(&module.source, iter, KW_STMT_SYNC)?;
 
        let body = self.consume_block_or_wrapped_statement(module, iter, ctx)?;
 

	
 
        Ok(ctx.heap.alloc_synchronous_statement(|this| SynchronousStatement{
 
            this,
 
            span: synchronous_span,
 
            body,
 
            end_sync: EndSynchronousStatementId::new_invalid(),
 
        }))
 
    }
 

	
 
    fn consume_fork_statement(
 
        &mut self, module: &Module, iter: &mut TokenIter, ctx: &mut PassCtx
 
    ) -> Result<ForkStatementId, ParseError> {
 
        let fork_span = consume_exact_ident(&module.source, iter, KW_STMT_FORK)?;
 
        let left_body = self.consume_block_or_wrapped_statement(module, iter, ctx)?;
 

	
 
        let right_body = if has_ident(&module.source, iter, KW_STMT_OR) {
 
            iter.consume();
 
            let right_body = self.consume_block_or_wrapped_statement(module, iter, ctx)?;
 
            Some(right_body)
 
        } else {
 
            None
 
        };
 

	
 
        Ok(ctx.heap.alloc_fork_statement(|this| ForkStatement{
 
            this,
 
            span: fork_span,
 
            left_body,
 
            right_body,
 
            end_fork: EndForkStatementId::new_invalid(),
 
        }))
 
    }
 

	
 
    fn consume_return_statement(
 
        &mut self, module: &Module, iter: &mut TokenIter, ctx: &mut PassCtx
 
    ) -> Result<ReturnStatementId, ParseError> {
 
        let return_span = consume_exact_ident(&module.source, iter, KW_STMT_RETURN)?;
 
        let mut scoped_section = self.expressions.start_section();
 

	
 
        consume_comma_separated_until(
 
            TokenKind::SemiColon, &module.source, iter, ctx,
 
            |_source, iter, ctx| self.consume_expression(module, iter, ctx),
 
            &mut scoped_section, "an expression", None
 
        )?;
 
        let expressions = scoped_section.into_vec();
 

	
 
        if expressions.is_empty() {
src/protocol/parser/pass_tokenizer.rs
Show inline comments
 
use crate::protocol::input_source::{
 
    InputSource as InputSource,
 
    ParseError,
 
    InputPosition as InputPosition,
 
};
 

	
 
use super::tokens::*;
 
use super::token_parsing::*;
 

	
 
/// Tokenizer is a reusable parser to tokenize multiple source files using the
 
/// same allocated buffers. In a well-formed program, we produce a consistent
 
/// tree of token ranges such that we may identify tokens that represent a
 
/// defintion or an import before producing the entire AST.
 
///
 
/// If the program is not well-formed then the tree may be inconsistent, but we
 
/// will detect this once we transform the tokens into the AST. To ensure a
 
/// consistent AST-producing phase we will require the import to have balanced
 
/// curly braces
 
pub(crate) struct PassTokenizer {
 
    // Stack of input positions of opening curly braces, used to detect
 
    // unmatched opening braces, unmatched closing braces are detected
 
    // immediately.
 
    curly_stack: Vec<InputPosition>,
 
    // Points to an element in the `TokenBuffer.ranges` variable.
 
    stack_idx: usize,
 
}
 

	
 
impl PassTokenizer {
 
    pub(crate) fn new() -> Self {
 
        Self{
 
            curly_stack: Vec::with_capacity(32),
 
            stack_idx: 0
 
        }
 
    }
 

	
 
    pub(crate) fn tokenize(&mut self, source: &mut InputSource, target: &mut TokenBuffer) -> Result<(), ParseError> {
 
        // Assert source and buffer are at start
 
        debug_assert_eq!(source.pos().offset, 0);
 
        debug_assert!(target.tokens.is_empty());
 
        debug_assert!(target.ranges.is_empty());
 

	
 
        // Set up for tokenization by pushing the first range onto the stack.
 
        // This range may get transformed into the appropriate range kind later,
 
        // see `push_range` and `pop_range`.
 
        self.stack_idx = 0;
 
        target.ranges.push(TokenRange{
 
            parent_idx: NO_RELATION,
 
            range_kind: TokenRangeKind::Module,
 
            curly_depth: 0,
 
            start: 0,
 
            end: 0,
 
            num_child_ranges: 0,
 
            first_child_idx: NO_RELATION,
 
            last_child_idx: NO_RELATION,
 
            next_sibling_idx: NO_RELATION,
 
        });
 

	
 
        // Main tokenization loop
 
        while let Some(c) = source.next() {
 
            let token_index = target.tokens.len() as u32;
 

	
 
            if is_char_literal_start(c) {
 
                self.consume_char_literal(source, target)?;
 
            } else if is_string_literal_start(c) {
 
                self.consume_string_literal(source, target)?;
 
            } else if is_identifier_start(c) {
 
                let ident = self.consume_identifier(source, target)?;
 

	
 
                if demarks_definition(ident) {
 
                    self.push_range(target, TokenRangeKind::Definition, token_index);
 
                } else if demarks_import(ident) {
 
                    self.push_range(target, TokenRangeKind::Import, token_index);
 
                }
 
            } else if is_integer_literal_start(c) {
 
                self.consume_number(source, target)?;
 
            } else if is_pragma_start_or_pound(c) {
 
                let was_pragma = self.consume_pragma_or_pound(c, source, target)?;
 
                if was_pragma {
 
                    self.push_range(target, TokenRangeKind::Pragma, token_index);
 
                }
 
            } else if self.is_line_comment_start(c, source) {
 
                self.consume_line_comment(source, target)?;
 
            } else if self.is_block_comment_start(c, source) {
 
                self.consume_block_comment(source, target)?;
 
            } else if is_whitespace(c) {
 
                let contained_newline = self.consume_whitespace(source);
 
                if contained_newline {
 
                    let range = &target.ranges[self.stack_idx];
 
                    if range.range_kind == TokenRangeKind::Pragma {
 
                        self.pop_range(target, target.tokens.len() as u32);
 
                    }
 
                }
 
            } else {
 
                let was_punctuation = self.maybe_parse_punctuation(c, source, target)?;
 
                if let Some((token, token_pos)) = was_punctuation {
 
                    if token == TokenKind::OpenCurly {
 
                        self.curly_stack.push(token_pos);
 
                    } else if token == TokenKind::CloseCurly {
 
                        // Check if this marks the end of a range we're
 
                        // currently processing
 
                        if self.curly_stack.is_empty() {
 
                            return Err(ParseError::new_error_str_at_pos(
 
                                source, token_pos, "unmatched closing curly brace '}'"
 
                            ));
 
                        }
 

	
 
                        self.curly_stack.pop();
 

	
 
                        let range = &target.ranges[self.stack_idx];
 
                        if range.range_kind == TokenRangeKind::Definition && range.curly_depth == self.curly_stack.len() as u32 {
 
                            self.pop_range(target, target.tokens.len() as u32);
 
                        }
 

	
 
                        // Exit early if we have more closing curly braces than
 
                        // opening curly braces
 
                    } else if token == TokenKind::SemiColon {
 
                        // Check if this marks the end of an import
 
                        let range = &target.ranges[self.stack_idx];
 
                        if range.range_kind == TokenRangeKind::Import {
 
                            self.pop_range(target, target.tokens.len() as u32);
 
                        }
 
                    }
 
                } else {
 
                    return Err(ParseError::new_error_str_at_pos(
 
                        source, source.pos(), "unexpected character"
 
                    ));
 
                }
 
            }
 
        }
 

	
 
        // End of file, check if our state is correct
 
        if let Some(error) = source.had_error.take() {
 
            return Err(error);
 
        }
 

	
 
        if !self.curly_stack.is_empty() {
 
            // Let's not add a lot of heuristics and just tell the programmer
 
            // that something is wrong
 
            let last_unmatched_open = self.curly_stack.pop().unwrap();
 
            return Err(ParseError::new_error_str_at_pos(
 
                source, last_unmatched_open, "unmatched opening curly brace '{'"
 
            ));
 
        }
 

	
 
        // Ranges that did not depend on curly braces may have missing tokens.
 
        // So close all of the active tokens
 
        while self.stack_idx != 0 {
 
            self.pop_range(target, target.tokens.len() as u32);
 
        }
 

	
 
        // And finally, we may have a token range at the end that doesn't belong
 
        // to a range yet, so insert a "code" range if this is the case.
 
        debug_assert_eq!(self.stack_idx, 0);
 
        let last_registered_idx = target.ranges[0].end;
 
        let last_token_idx = target.tokens.len() as u32;
 
        if last_registered_idx != last_token_idx {
 
            self.add_code_range(target, 0, last_registered_idx, last_token_idx, NO_RELATION);
 
        }
 

	
 
        // TODO: @remove once I'm sure the algorithm works. For now it is better
 
        //  if the debugging is a little more expedient
 
        if cfg!(debug_assertions) {
 
            // For each range make sure its children make sense
 
            for parent_idx in 0..target.ranges.len() {
 
                let cur_range = &target.ranges[parent_idx];
 
                if cur_range.num_child_ranges == 0 {
 
                    assert_eq!(cur_range.first_child_idx, NO_RELATION);
 
                    assert_eq!(cur_range.last_child_idx, NO_RELATION);
 
                } else {
 
                    assert_ne!(cur_range.first_child_idx, NO_RELATION);
 
                    assert_ne!(cur_range.last_child_idx, NO_RELATION);
 

	
 
                    let mut child_counter = 0u32;
 
                    let mut last_valid_child_idx = cur_range.first_child_idx;
 
                    let mut child_idx = cur_range.first_child_idx;
 
                    while child_idx != NO_RELATION {
 
                        let child_range = &target.ranges[child_idx as usize];
 
                        assert_eq!(child_range.parent_idx, parent_idx as i32);
 
                        last_valid_child_idx = child_idx;
 
                        child_idx = child_range.next_sibling_idx;
 
                        child_counter += 1;
 
                    }
 

	
 
                    assert_eq!(cur_range.last_child_idx, last_valid_child_idx);
 
                    assert_eq!(cur_range.num_child_ranges, child_counter);
 
                }
 
            }
 
        }
 

	
 
        Ok(())
 
    }
 

	
 
    fn is_line_comment_start(&self, first_char: u8, source: &InputSource) -> bool {
 
        return first_char == b'/' && Some(b'/') == source.lookahead(1);
 
    }
 

	
 
    fn is_block_comment_start(&self, first_char: u8, source: &InputSource) -> bool {
 
        return first_char == b'/' && Some(b'*') == source.lookahead(1);
 
    }
 

	
 
    fn maybe_parse_punctuation(
 
        &mut self, first_char: u8, source: &mut InputSource, target: &mut TokenBuffer
 
    ) -> Result<Option<(TokenKind, InputPosition)>, ParseError> {
 
        debug_assert!(first_char != b'#', "'#' needs special handling");
 
        debug_assert!(first_char != b'\'', "'\'' needs special handling");
 
        debug_assert!(first_char != b'"', "'\"' needs special handling");
 

	
 
        let pos = source.pos();
 
        let token_kind;
 
        if first_char == b'!' {
 
            source.consume();
 
            if Some(b'=') == source.next() {
 
                source.consume();
 
                token_kind = TokenKind::NotEqual;
 
            } else {
 
                token_kind = TokenKind::Exclamation;
 
            }
 
        } else if first_char == b'%' {
 
            source.consume();
 
            if Some(b'=') == source.next() {
 
                source.consume();
 
                token_kind = TokenKind::PercentEquals;
 
            } else {
 
                token_kind = TokenKind::Percent;
 
            }
 
        } else if first_char == b'&' {
 
            source.consume();
 
            let next = source.next();
 
            if Some(b'&') == next {
 
                source.consume();
 
                token_kind = TokenKind::AndAnd;
 
            } else if Some(b'=') == next {
 
                source.consume();
 
                token_kind = TokenKind::AndEquals;
 
            } else {
 
                token_kind = TokenKind::And;
 
            }
 
        } else if first_char == b'(' {
 
            source.consume();
 
            token_kind = TokenKind::OpenParen;
 
        } else if first_char == b')' {
 
            source.consume();
 
            token_kind = TokenKind::CloseParen;
 
        } else if first_char == b'*' {
 
            source.consume();
 
            if let Some(b'=') = source.next() {
 
                source.consume();
 
                token_kind = TokenKind::StarEquals;
 
            } else {
 
                token_kind = TokenKind::Star;
 
            }
 
        } else if first_char == b'+' {
 
            source.consume();
 
            let next = source.next();
 
            if Some(b'+') == next {
 
                source.consume();
 
                token_kind = TokenKind::PlusPlus;
 
            } else if Some(b'=') == next {
 
                source.consume();
 
                token_kind = TokenKind::PlusEquals;
 
            } else {
 
                token_kind = TokenKind::Plus;
 
            }
 
        } else if first_char == b',' {
 
            source.consume();
 
            token_kind = TokenKind::Comma;
 
        } else if first_char == b'-' {
 
            source.consume();
 
            let next = source.next();
 
            if Some(b'-') == next {
 
                source.consume();
 
                token_kind = TokenKind::MinusMinus;
 
            } else if Some(b'>') == next {
 
                source.consume();
 
                token_kind = TokenKind::ArrowRight;
 
            } else if Some(b'=') == next {
 
                source.consume();
 
                token_kind = TokenKind::MinusEquals;
 
            } else {
 
                token_kind = TokenKind::Minus;
 
            }
 
        } else if first_char == b'.' {
 
            source.consume();
 
            if let Some(b'.') = source.next() {
 
                source.consume();
 
                token_kind = TokenKind::DotDot;
 
            } else {
 
                token_kind = TokenKind::Dot
 
            }
 
        } else if first_char == b'/' {
 
            source.consume();
 
            debug_assert_ne!(Some(b'/'), source.next());
 
            debug_assert_ne!(Some(b'*'), source.next());
 
            if let Some(b'=') = source.next() {
 
                source.consume();
 
                token_kind = TokenKind::SlashEquals;
 
            } else {
 
                token_kind = TokenKind::Slash;
 
            }
 
        } else if first_char == b':' {
 
            source.consume();
 
            if let Some(b':') = source.next() {
 
                source.consume();
 
                token_kind = TokenKind::ColonColon;
 
            } else {
 
                token_kind = TokenKind::Colon;
 
            }
 
        } else if first_char == b';' {
 
            source.consume();
 
            token_kind = TokenKind::SemiColon;
 
        } else if first_char == b'<' {
 
            source.consume();
 
            let next = source.next();
 
            if let Some(b'<') = next {
 
                source.consume();
 
                if let Some(b'=') = source.next() {
 
                    source.consume();
 
                    token_kind = TokenKind::ShiftLeftEquals;
 
                } else {
 
                    token_kind = TokenKind::ShiftLeft;
 
                }
 
            } else if let Some(b'=') = next {
 
                source.consume();
 
                token_kind = TokenKind::LessEquals;
 
            } else {
 
                token_kind = TokenKind::OpenAngle;
 
            }
 
        } else if first_char == b'=' {
 
            source.consume();
 
            if let Some(b'=') = source.next() {
 
                source.consume();
 
                token_kind = TokenKind::EqualEqual;
 
            } else {
 
                token_kind = TokenKind::Equal;
 
            }
 
        } else if first_char == b'>' {
 
            source.consume();
 
            let next = source.next();
 
            if Some(b'>') == next {
 
                source.consume();
 
                if Some(b'=') == source.next() {
 
                    source.consume();
 
                    token_kind = TokenKind::ShiftRightEquals;
 
                } else {
 
                    token_kind = TokenKind::ShiftRight;
 
                }
 
            } else if Some(b'=') == next {
 
                source.consume();
 
                token_kind = TokenKind::GreaterEquals;
 
            } else {
 
                token_kind = TokenKind::CloseAngle;
 
            }
 
        } else if first_char == b'?' {
 
            source.consume();
 
            token_kind = TokenKind::Question;
 
        } else if first_char == b'@' {
 
            source.consume();
 
            if let Some(b'=') = source.next() {
 
                source.consume();
 
                token_kind = TokenKind::AtEquals;
 
            } else {
 
                token_kind = TokenKind::At;
 
            }
 
        } else if first_char == b'[' {
 
            source.consume();
 
            token_kind = TokenKind::OpenSquare;
 
        } else if first_char == b']' {
 
            source.consume();
 
            token_kind = TokenKind::CloseSquare;
 
        } else if first_char == b'^' {
 
            source.consume();
 
            if let Some(b'=') = source.next() {
 
                source.consume();
 
                token_kind = TokenKind::CaretEquals;
 
            } else {
 
                token_kind = TokenKind::Caret;
 
            }
 
        } else if first_char == b'{' {
 
            source.consume();
 
            token_kind = TokenKind::OpenCurly;
 
        } else if first_char == b'|' {
 
            source.consume();
 
            let next = source.next();
 
            if Some(b'|') == next {
 
                source.consume();
 
                token_kind = TokenKind::OrOr;
 
            } else if Some(b'=') == next {
 
                source.consume();
 
                token_kind = TokenKind::OrEquals;
 
            } else {
 
                token_kind = TokenKind::Or;
 
            }
 
        } else if first_char == b'}' {
 
            source.consume();
 
            token_kind = TokenKind::CloseCurly;
 
        } else if first_char == b'~' {
 
            source.consume();
 
            token_kind = TokenKind::Tilde;
 
        } else {
 
            self.check_ascii(source)?;
 
            return Ok(None);
 
        }
 

	
 
        target.tokens.push(Token::new(token_kind, pos));
 
        Ok(Some((token_kind, pos)))
 
    }
 

	
 
    fn consume_char_literal(&mut self, source: &mut InputSource, target: &mut TokenBuffer) -> Result<(), ParseError> {
 
        let begin_pos = source.pos();
 

	
 
        // Consume the leading quote
 
        debug_assert!(source.next().unwrap() == b'\'');
 
        source.consume();
 

	
 
        let mut prev_char = b'\'';
 
        while let Some(c) = source.next() {
 
            if !c.is_ascii() {
 
                return Err(ParseError::new_error_str_at_pos(source, source.pos(), "non-ASCII character in char literal"));
 
            }
 
            source.consume();
 

	
 
            // Make sure ending quote was not escaped
 
            if c == b'\'' && prev_char != b'\\' {
 
                prev_char = c;
 
                break;
 
            }
 

	
 
            prev_char = c;
 
        }
 

	
 
        if prev_char != b'\'' {
 
            // Unterminated character literal, reached end of file.
 
            return Err(ParseError::new_error_str_at_pos(source, begin_pos, "encountered unterminated character literal"));
 
        }
 

	
 
        let end_pos = source.pos();
 

	
 
        target.tokens.push(Token::new(TokenKind::Character, begin_pos));
 
        target.tokens.push(Token::new(TokenKind::SpanEnd, end_pos));
 

	
 
        Ok(())
 
    }
 

	
 
    fn consume_string_literal(&mut self, source: &mut InputSource, target: &mut TokenBuffer) -> Result<(), ParseError> {
 
        let begin_pos = source.pos();
 

	
 
        // Consume the leading double quotes
 
        debug_assert!(source.next().unwrap() == b'"');
 
        source.consume();
 

	
 
        let mut prev_char = b'"';
 
        while let Some(c) = source.next() {
 
            if !c.is_ascii() {
 
                return Err(ParseError::new_error_str_at_pos(source, source.pos(), "non-ASCII character in string literal"));
 
            }
 

	
 
            source.consume();
 
            if c == b'"' && prev_char != b'\\' {
 
                // Unescaped string terminator
 
                prev_char = c;
 
                break;
 
            }
 

	
 
            if prev_char == b'\\' && c == b'\\' {
 
                // Escaped backslash, set prev_char to bogus to not conflict
 
                // with escaped-" and unterminated string literal detection.
 
                prev_char = b'\0';
 
            } else {
 
                prev_char = c;
 
            }
 
        }
 

	
 
        if prev_char != b'"' {
 
            // Unterminated string literal
 
            return Err(ParseError::new_error_str_at_pos(source, begin_pos, "encountered unterminated string literal"));
 
        }
 

	
 
        let end_pos = source.pos();
 
        target.tokens.push(Token::new(TokenKind::String, begin_pos));
 
        target.tokens.push(Token::new(TokenKind::SpanEnd, end_pos));
 

	
 
        Ok(())
 
    }
 

	
 
    fn consume_pragma_or_pound(&mut self, first_char: u8, source: &mut InputSource, target: &mut TokenBuffer) -> Result<bool, ParseError> {
 
        let start_pos = source.pos();
 
        debug_assert_eq!(first_char, b'#');
 
        source.consume();
 

	
 
        let next = source.next();
 
        if next.is_none() || !is_identifier_start(next.unwrap()) {
 
            // Just a pound sign
 
            target.tokens.push(Token::new(TokenKind::Pound, start_pos));
 
            Ok(false)
 
        } else {
 
            // Pound sign followed by identifier
 
            source.consume();
 
            while let Some(c) = source.next() {
 
                if !is_identifier_remaining(c) {
 
                    break;
 
                }
 
                source.consume();
 
            }
 

	
 
            self.check_ascii(source)?;
 

	
 
            let end_pos = source.pos();
 
            target.tokens.push(Token::new(TokenKind::Pragma, start_pos));
 
            target.tokens.push(Token::new(TokenKind::SpanEnd, end_pos));
 
            Ok(true)
 
        }
 
    }
 

	
 
    fn consume_line_comment(&mut self, source: &mut InputSource, target: &mut TokenBuffer) -> Result<(), ParseError> {
 
        let begin_pos = source.pos();
 

	
 
        // Consume the leading "//"
 
        debug_assert!(source.next().unwrap() == b'/' && source.lookahead(1).unwrap() == b'/');
 
        source.consume();
 
        source.consume();
 

	
 
        let mut prev_char = b'/';
 
        let mut cur_char = b'/';
 
        while let Some(c) = source.next() {
 
            prev_char = cur_char;
 
            cur_char = c;
 

	
 
            if c == b'\n' {
 
                // End of line, note that the newline is not consumed
 
                break;
 
            }
 

	
 
            source.consume();
 
        }
 

	
 
        let mut end_pos = source.pos();
 
        debug_assert_eq!(begin_pos.line, end_pos.line);
 

	
 
        // Modify offset to not include the newline characters
 
        if cur_char == b'\n' {
 
            if prev_char == b'\r' {
 
                end_pos.offset -= 2;
 
            } else {
 
                end_pos.offset -= 1;
 
            }
 
            // Consume final newline
 
            source.consume();
 
        } else {
 
            // End of comment was due to EOF
 
            debug_assert!(source.next().is_none())
 
        }
 

	
 
        target.tokens.push(Token::new(TokenKind::LineComment, begin_pos));
 
        target.tokens.push(Token::new(TokenKind::SpanEnd, end_pos));
 

	
 
        Ok(())
 
    }
 

	
 
    fn consume_block_comment(&mut self, source: &mut InputSource, target: &mut TokenBuffer) -> Result<(), ParseError> {
 
        let begin_pos = source.pos();
 

	
 
        // Consume the leading "/*"
 
        debug_assert!(source.next().unwrap() == b'/' && source.lookahead(1).unwrap() == b'*');
 
        source.consume();
 
        source.consume();
 

	
 
        // Explicitly do not put prev_char at "*", because then "/*/" would
 
        // represent a valid and closed block comment
 
        let mut prev_char = b' ';
 
        let mut is_closed = false;
 
        while let Some(c) = source.next() {
 
            source.consume();
 
            if prev_char == b'*' && c == b'/' {
src/protocol/parser/pass_typing.rs
Show inline comments
 
@@ -536,770 +536,768 @@ impl InferenceType {
 
            }
 
            
 
            if part_a.is_marker() { idx_a += 1; continue; }
 
            if part_b.is_marker() { idx_b += 1; continue; }
 

	
 
            if let Some(depth_change) = Self::check_part_for_single_type(
 
                type_parts_a, &mut idx_a, type_parts_b, &mut idx_b
 
            ) {
 
                depth += depth_change;
 
                continue;
 
            }
 
            if let Some(depth_change) = Self::check_part_for_single_type(
 
                type_parts_b, &mut idx_b, type_parts_a, &mut idx_a
 
            ) {
 
                depth += depth_change;
 
                continue;
 
            }
 

	
 
            return false;
 
        }
 

	
 
        true
 
    }
 

	
 
    /// Performs the conversion of the inference type into a concrete type.
 
    /// By calling this function you must make sure that no unspecified types
 
    /// (e.g. Unknown or IntegerLike) exist in the type. Will not clear or check
 
    /// if the supplied `ConcreteType` is empty, will simply append to the parts
 
    /// vector.
 
    fn write_concrete_type(&self, concrete_type: &mut ConcreteType) {
 
        use InferenceTypePart as ITP;
 
        use ConcreteTypePart as CTP;
 

	
 
        // Make sure inference type is specified but concrete type is not yet specified
 
        debug_assert!(!self.parts.is_empty());
 
        concrete_type.parts.reserve(self.parts.len());
 

	
 
        let mut idx = 0;
 
        while idx < self.parts.len() {
 
            let part = &self.parts[idx];
 
            let converted_part = match part {
 
                ITP::Marker(_) => {
 
                    // Markers are removed when writing to the concrete type.
 
                    idx += 1;
 
                    continue;
 
                },
 
                ITP::Unknown | ITP::NumberLike |
 
                ITP::IntegerLike | ITP::ArrayLike | ITP::PortLike => {
 
                    // Should not happen if type inferencing works correctly: we
 
                    // should have returned a programmer-readable error or have
 
                    // inferred all types.
 
                    unreachable!("attempted to convert inference type part {:?} into concrete type", part);
 
                },
 
                ITP::Void => CTP::Void,
 
                ITP::Message => CTP::Message,
 
                ITP::Bool => CTP::Bool,
 
                ITP::UInt8 => CTP::UInt8,
 
                ITP::UInt16 => CTP::UInt16,
 
                ITP::UInt32 => CTP::UInt32,
 
                ITP::UInt64 => CTP::UInt64,
 
                ITP::SInt8 => CTP::SInt8,
 
                ITP::SInt16 => CTP::SInt16,
 
                ITP::SInt32 => CTP::SInt32,
 
                ITP::SInt64 => CTP::SInt64,
 
                ITP::Character => CTP::Character,
 
                ITP::String => {
 
                    // Inferred type has a 'char' subtype to simplify array
 
                    // checking, we remove it here.
 
                    debug_assert_eq!(self.parts[idx + 1], InferenceTypePart::Character);
 
                    idx += 1;
 
                    CTP::String
 
                },
 
                ITP::Array => CTP::Array,
 
                ITP::Slice => CTP::Slice,
 
                ITP::Input => CTP::Input,
 
                ITP::Output => CTP::Output,
 
                ITP::Instance(id, num) => CTP::Instance(*id, *num),
 
            };
 

	
 
            concrete_type.parts.push(converted_part);
 
            idx += 1;
 
        }
 
    }
 

	
 
    /// Writes a human-readable version of the type to a string. This is used
 
    /// to display error messages
 
    fn write_display_name(
 
        buffer: &mut String, heap: &Heap, parts: &[InferenceTypePart], mut idx: usize
 
    ) -> usize {
 
        use InferenceTypePart as ITP;
 

	
 
        match &parts[idx] {
 
            ITP::Marker(_marker_idx) => {
 
                if debug_log_enabled!() {
 
                    buffer.push_str(&format!("{{Marker:{}}}", *_marker_idx));
 
                }
 
                idx = Self::write_display_name(buffer, heap, parts, idx + 1);
 
            },
 
            ITP::Unknown => buffer.push_str("?"),
 
            ITP::NumberLike => buffer.push_str("numberlike"),
 
            ITP::IntegerLike => buffer.push_str("integerlike"),
 
            ITP::ArrayLike => {
 
                idx = Self::write_display_name(buffer, heap, parts, idx + 1);
 
                buffer.push_str("[?]");
 
            },
 
            ITP::PortLike => {
 
                buffer.push_str("portlike<");
 
                idx = Self::write_display_name(buffer, heap, parts, idx + 1);
 
                buffer.push('>');
 
            }
 
            ITP::Void => buffer.push_str("void"),
 
            ITP::Bool => buffer.push_str(KW_TYPE_BOOL_STR),
 
            ITP::UInt8 => buffer.push_str(KW_TYPE_UINT8_STR),
 
            ITP::UInt16 => buffer.push_str(KW_TYPE_UINT16_STR),
 
            ITP::UInt32 => buffer.push_str(KW_TYPE_UINT32_STR),
 
            ITP::UInt64 => buffer.push_str(KW_TYPE_UINT64_STR),
 
            ITP::SInt8 => buffer.push_str(KW_TYPE_SINT8_STR),
 
            ITP::SInt16 => buffer.push_str(KW_TYPE_SINT16_STR),
 
            ITP::SInt32 => buffer.push_str(KW_TYPE_SINT32_STR),
 
            ITP::SInt64 => buffer.push_str(KW_TYPE_SINT64_STR),
 
            ITP::Character => buffer.push_str(KW_TYPE_CHAR_STR),
 
            ITP::String => {
 
                buffer.push_str(KW_TYPE_STRING_STR);
 
                idx += 1; // skip the 'char' subtype
 
            },
 
            ITP::Message => {
 
                buffer.push_str(KW_TYPE_MESSAGE_STR);
 
                buffer.push('<');
 
                idx = Self::write_display_name(buffer, heap, parts, idx + 1);
 
                buffer.push('>');
 
            },
 
            ITP::Array => {
 
                idx = Self::write_display_name(buffer, heap, parts, idx + 1);
 
                buffer.push_str("[]");
 
            },
 
            ITP::Slice => {
 
                idx = Self::write_display_name(buffer, heap, parts, idx + 1);
 
                buffer.push_str("[..]");
 
            },
 
            ITP::Input => {
 
                buffer.push_str(KW_TYPE_IN_PORT_STR);
 
                buffer.push('<');
 
                idx = Self::write_display_name(buffer, heap, parts, idx + 1);
 
                buffer.push('>');
 
            },
 
            ITP::Output => {
 
                buffer.push_str(KW_TYPE_OUT_PORT_STR);
 
                buffer.push('<');
 
                idx = Self::write_display_name(buffer, heap, parts, idx + 1);
 
                buffer.push('>');
 
            },
 
            ITP::Instance(definition_id, num_sub) => {
 
                let definition = &heap[*definition_id];
 
                buffer.push_str(definition.identifier().value.as_str());
 
                if *num_sub > 0 {
 
                    buffer.push('<');
 
                    idx = Self::write_display_name(buffer, heap, parts, idx + 1);
 
                    for _sub_idx in 1..*num_sub {
 
                        buffer.push_str(", ");
 
                        idx = Self::write_display_name(buffer, heap, parts, idx + 1);
 
                    }
 
                    buffer.push('>');
 
                }
 
            },
 
        }
 

	
 
        idx
 
    }
 

	
 
    /// Returns the display name of a (part of) the type tree. Will allocate a
 
    /// string.
 
    fn partial_display_name(heap: &Heap, parts: &[InferenceTypePart]) -> String {
 
        let mut buffer = String::with_capacity(parts.len() * 6);
 
        Self::write_display_name(&mut buffer, heap, parts, 0);
 
        buffer
 
    }
 

	
 
    /// Returns the display name of the full type tree. Will allocate a string.
 
    fn display_name(&self, heap: &Heap) -> String {
 
        Self::partial_display_name(heap, &self.parts)
 
    }
 
}
 

	
 
impl Default for InferenceType {
 
    fn default() -> Self {
 
        Self{
 
            has_marker: false,
 
            is_done: false,
 
            parts: Vec::new(),
 
        }
 
    }
 
}
 

	
 
/// Iterator over the subtrees that follow a marker in an `InferenceType`
 
/// instance. Returns immutable slices over the internal parts
 
struct InferenceTypeMarkerIter<'a> {
 
    parts: &'a [InferenceTypePart],
 
    idx: usize,
 
}
 

	
 
impl<'a> InferenceTypeMarkerIter<'a> {
 
    fn new(parts: &'a [InferenceTypePart]) -> Self {
 
        Self{ parts, idx: 0 }
 
    }
 
}
 

	
 
impl<'a> Iterator for InferenceTypeMarkerIter<'a> {
 
    type Item = (u32, &'a [InferenceTypePart]);
 

	
 
    fn next(&mut self) -> Option<Self::Item> {
 
        // Iterate until we find a marker
 
        while self.idx < self.parts.len() {
 
            if let InferenceTypePart::Marker(marker) = self.parts[self.idx] {
 
                // Found a marker, find the subtree end
 
                let start_idx = self.idx + 1;
 
                let end_idx = InferenceType::find_subtree_end_idx(self.parts, start_idx);
 

	
 
                // Modify internal index, then return items
 
                self.idx = end_idx;
 
                return Some((marker, &self.parts[start_idx..end_idx]));
 
            }
 

	
 
            self.idx += 1;
 
        }
 

	
 
        None
 
    }
 
}
 

	
 
#[derive(Debug, PartialEq, Eq)]
 
enum DualInferenceResult {
 
    Neither,        // neither argument is clarified
 
    First,          // first argument is clarified using the second one
 
    Second,         // second argument is clarified using the first one
 
    Both,           // both arguments are clarified
 
    Incompatible,   // types are incompatible: programmer error
 
}
 

	
 
impl DualInferenceResult {
 
    fn modified_lhs(&self) -> bool {
 
        match self {
 
            DualInferenceResult::First | DualInferenceResult::Both => true,
 
            _ => false
 
        }
 
    }
 
    fn modified_rhs(&self) -> bool {
 
        match self {
 
            DualInferenceResult::Second | DualInferenceResult::Both => true,
 
            _ => false
 
        }
 
    }
 
}
 

	
 
#[derive(Debug, PartialEq, Eq)]
 
enum SingleInferenceResult {
 
    Unmodified,
 
    Modified,
 
    Incompatible
 
}
 

	
 
enum DefinitionType{
 
    Component(ComponentDefinitionId),
 
    Function(FunctionDefinitionId),
 
}
 

	
 
impl DefinitionType {
 
    fn definition_id(&self) -> DefinitionId {
 
        match self {
 
            DefinitionType::Component(v) => v.upcast(),
 
            DefinitionType::Function(v) => v.upcast(),
 
        }
 
    }
 
}
 

	
 
pub(crate) struct ResolveQueueElement {
 
    // Note that using the `definition_id` and the `monomorph_idx` one may
 
    // query the type table for the full procedure type, thereby retrieving
 
    // the polymorphic arguments to the procedure.
 
    pub(crate) root_id: RootId,
 
    pub(crate) definition_id: DefinitionId,
 
    pub(crate) reserved_monomorph_idx: i32,
 
}
 

	
 
pub(crate) type ResolveQueue = Vec<ResolveQueueElement>;
 

	
 
#[derive(Clone)]
 
struct InferenceExpression {
 
    expr_type: InferenceType,       // result type from expression
 
    expr_id: ExpressionId,          // expression that is evaluated
 
    field_or_monomorph_idx: i32,    // index of field, of index of monomorph array in type table
 
    extra_data_idx: i32,     // index of extra data needed for inference
 
}
 

	
 
impl Default for InferenceExpression {
 
    fn default() -> Self {
 
        Self{
 
            expr_type: InferenceType::default(),
 
            expr_id: ExpressionId::new_invalid(),
 
            field_or_monomorph_idx: -1,
 
            extra_data_idx: -1,
 
        }
 
    }
 
}
 

	
 
/// This particular visitor will recurse depth-first into the AST and ensures
 
/// that all expressions have the appropriate types.
 
pub(crate) struct PassTyping {
 
    // Current definition we're typechecking.
 
    reserved_idx: i32,
 
    definition_type: DefinitionType,
 
    poly_vars: Vec<ConcreteType>,
 

	
 
    // Buffers for iteration over substatements and subexpressions
 
    stmt_buffer: Vec<StatementId>,
 
    expr_buffer: Vec<ExpressionId>,
 

	
 
    // Mapping from parser type to inferred type. We attempt to continue to
 
    // specify these types until we're stuck or we've fully determined the type.
 
    var_types: HashMap<VariableId, VarData>,            // types of variables
 
    expr_types: Vec<InferenceExpression>,                     // will be transferred to type table at end
 
    extra_data: Vec<ExtraData>,       // data for polymorph inference
 
    // Keeping track of which expressions need to be reinferred because the
 
    // expressions they're linked to made progression on an associated type
 
    expr_queued: DequeSet<i32>,
 
}
 

	
 
// TODO: @Rename, this is used for a lot of type inferencing. It seems like
 
//  there is a different underlying architecture waiting to surface.
 
struct ExtraData {
 
    expr_id: ExpressionId, // the expression with which this data is associated
 
    definition_id: DefinitionId, // the definition, only used for user feedback
 
    /// Progression of polymorphic variables (if any)
 
    poly_vars: Vec<InferenceType>,
 
    /// Progression of types of call arguments or struct members
 
    embedded: Vec<InferenceType>,
 
    returned: InferenceType,
 
}
 

	
 
impl Default for ExtraData {
 
    fn default() -> Self {
 
        Self{
 
            expr_id: ExpressionId::new_invalid(),
 
            definition_id: DefinitionId::new_invalid(),
 
            poly_vars: Vec::new(),
 
            embedded: Vec::new(),
 
            returned: InferenceType::default(),
 
        }
 
    }
 
}
 

	
 
struct VarData {
 
    /// Type of the variable
 
    var_type: InferenceType,
 
    /// VariableExpressions that use the variable
 
    used_at: Vec<ExpressionId>,
 
    /// For channel statements we link to the other variable such that when one
 
    /// channel's interior type is resolved, we can also resolve the other one.
 
    linked_var: Option<VariableId>,
 
}
 

	
 
impl VarData {
 
    fn new_channel(var_type: InferenceType, other_port: VariableId) -> Self {
 
        Self{ var_type, used_at: Vec::new(), linked_var: Some(other_port) }
 
    }
 
    fn new_local(var_type: InferenceType) -> Self {
 
        Self{ var_type, used_at: Vec::new(), linked_var: None }
 
    }
 
}
 

	
 
impl PassTyping {
 
    pub(crate) fn new() -> Self {
 
        PassTyping {
 
            reserved_idx: -1,
 
            definition_type: DefinitionType::Function(FunctionDefinitionId::new_invalid()),
 
            poly_vars: Vec::new(),
 
            stmt_buffer: Vec::with_capacity(STMT_BUFFER_INIT_CAPACITY),
 
            expr_buffer: Vec::with_capacity(EXPR_BUFFER_INIT_CAPACITY),
 
            var_types: HashMap::new(),
 
            expr_types: Vec::new(),
 
            extra_data: Vec::new(),
 
            expr_queued: DequeSet::new(),
 
        }
 
    }
 

	
 
    // TODO: @cleanup Unsure about this, maybe a pattern will arise after
 
    //  a while.
 
    pub(crate) fn queue_module_definitions(ctx: &mut Ctx, queue: &mut ResolveQueue) {
 
        debug_assert_eq!(ctx.module().phase, ModuleCompilationPhase::ValidatedAndLinked);
 
        let root_id = ctx.module().root_id;
 
        let root = &ctx.heap.protocol_descriptions[root_id];
 
        for definition_id in &root.definitions {
 
            let definition = &ctx.heap[*definition_id];
 

	
 
            let first_concrete_part = match definition {
 
                Definition::Function(definition) => {
 
                    if definition.poly_vars.is_empty() {
 
                        Some(ConcreteTypePart::Function(*definition_id, 0))
 
                    } else {
 
                        None
 
                    }
 
                }
 
                Definition::Component(definition) => {
 
                    if definition.poly_vars.is_empty() {
 
                        Some(ConcreteTypePart::Component(*definition_id, 0))
 
                    } else {
 
                        None
 
                    }
 
                },
 
                Definition::Enum(_) | Definition::Struct(_) | Definition::Union(_) => None,
 
            };
 

	
 
            if let Some(first_concrete_part) = first_concrete_part {
 
                let concrete_type = ConcreteType{ parts: vec![first_concrete_part] };
 
                let reserved_idx = ctx.types.reserve_procedure_monomorph_index(definition_id, concrete_type);
 
                queue.push(ResolveQueueElement{
 
                    root_id,
 
                    definition_id: *definition_id,
 
                    reserved_monomorph_idx: reserved_idx,
 
                })
 
            }
 
        }
 
    }
 

	
 
    pub(crate) fn handle_module_definition(
 
        &mut self, ctx: &mut Ctx, queue: &mut ResolveQueue, element: ResolveQueueElement
 
    ) -> VisitorResult {
 
        self.reset();
 
        debug_assert_eq!(ctx.module().root_id, element.root_id);
 
        debug_assert!(self.poly_vars.is_empty());
 

	
 
        // Prepare for visiting the definition
 
        self.reserved_idx = element.reserved_monomorph_idx;
 

	
 
        let proc_base = ctx.types.get_base_definition(&element.definition_id).unwrap();
 
        if proc_base.is_polymorph {
 
            let proc_monos = proc_base.definition.procedure_monomorphs();
 
            let proc_mono = &(*proc_monos)[element.reserved_monomorph_idx as usize];
 

	
 
            for poly_arg in proc_mono.concrete_type.embedded_iter(0) {
 
                self.poly_vars.push(ConcreteType{ parts: Vec::from(poly_arg) });
 
            }
 
        }
 

	
 
        // Visit the definition, setting up the type resolving process, then
 
        // (attempt to) resolve all types
 
        self.visit_definition(ctx, element.definition_id)?;
 
        self.resolve_types(ctx, queue)?;
 
        Ok(())
 
    }
 

	
 
    fn reset(&mut self) {
 
        self.reserved_idx = -1;
 
        self.definition_type = DefinitionType::Function(FunctionDefinitionId::new_invalid());
 
        self.poly_vars.clear();
 
        self.stmt_buffer.clear();
 
        self.expr_buffer.clear();
 
        self.var_types.clear();
 
        self.expr_types.clear();
 
        self.extra_data.clear();
 
        self.expr_queued.clear();
 
    }
 
}
 

	
 
impl Visitor for PassTyping {
 
    // Definitions
 

	
 
    fn visit_component_definition(&mut self, ctx: &mut Ctx, id: ComponentDefinitionId) -> VisitorResult {
 
        self.definition_type = DefinitionType::Component(id);
 

	
 
        let comp_def = &ctx.heap[id];
 
        debug_assert_eq!(comp_def.poly_vars.len(), self.poly_vars.len(), "component polyvars do not match imposed polyvars");
 

	
 
        debug_log!("{}", "-".repeat(50));
 
        debug_log!("Visiting component '{}': {}", comp_def.identifier.value.as_str(), id.0.index);
 
        debug_log!("{}", "-".repeat(50));
 

	
 
        // Reserve data for expression types
 
        debug_assert!(self.expr_types.is_empty());
 
        self.expr_types.resize(comp_def.num_expressions_in_body as usize, Default::default());
 

	
 
        // Visit parameters
 
        for param_id in comp_def.parameters.clone() {
 
            let param = &ctx.heap[param_id];
 
            let var_type = self.determine_inference_type_from_parser_type_elements(&param.parser_type.elements, true);
 
            debug_assert!(var_type.is_done, "expected component arguments to be concrete types");
 
            self.var_types.insert(param_id, VarData::new_local(var_type));
 
        }
 

	
 
        // Visit the body and all of its expressions
 
        let body_stmt_id = ctx.heap[id].body;
 
        self.visit_block_stmt(ctx, body_stmt_id)
 
    }
 

	
 
    fn visit_function_definition(&mut self, ctx: &mut Ctx, id: FunctionDefinitionId) -> VisitorResult {
 
        self.definition_type = DefinitionType::Function(id);
 

	
 
        let func_def = &ctx.heap[id];
 
        debug_assert_eq!(func_def.poly_vars.len(), self.poly_vars.len(), "function polyvars do not match imposed polyvars");
 

	
 
        debug_log!("{}", "-".repeat(50));
 
        debug_log!("Visiting function '{}': {}", func_def.identifier.value.as_str(), id.0.index);
 
        if debug_log_enabled!() {
 
            debug_log!("Polymorphic variables:");
 
            for (_idx, poly_var) in self.poly_vars.iter().enumerate() {
 
                let mut infer_type_parts = Vec::new();
 
                Self::determine_inference_type_from_concrete_type(
 
                    &mut infer_type_parts, &poly_var.parts
 
                );
 
                let _infer_type = InferenceType::new(false, true, infer_type_parts);
 
                debug_log!(" - [{:03}] {:?}", _idx, _infer_type.display_name(&ctx.heap));
 
            }
 
        }
 
        debug_log!("{}", "-".repeat(50));
 

	
 
        // Reserve data for expression types
 
        debug_assert!(self.expr_types.is_empty());
 
        self.expr_types.resize(func_def.num_expressions_in_body as usize, Default::default());
 

	
 
        // Visit parameters
 
        for param_id in func_def.parameters.clone() {
 
            let param = &ctx.heap[param_id];
 
            let var_type = self.determine_inference_type_from_parser_type_elements(&param.parser_type.elements, true);
 
            debug_assert!(var_type.is_done, "expected function arguments to be concrete types");
 
            self.var_types.insert(param_id, VarData::new_local(var_type));
 
        }
 

	
 
        // Visit all of the expressions within the body
 
        let body_stmt_id = ctx.heap[id].body;
 
        self.visit_block_stmt(ctx, body_stmt_id)
 
    }
 

	
 
    // Statements
 

	
 
    fn visit_block_stmt(&mut self, ctx: &mut Ctx, id: BlockStatementId) -> VisitorResult {
 
        // Transfer statements for traversal
 
        let block = &ctx.heap[id];
 

	
 
        for stmt_id in block.statements.clone() {
 
            self.visit_stmt(ctx, stmt_id)?;
 
        }
 

	
 
        Ok(())
 
    }
 

	
 
    fn visit_local_memory_stmt(&mut self, ctx: &mut Ctx, id: MemoryStatementId) -> VisitorResult {
 
        let memory_stmt = &ctx.heap[id];
 

	
 
        let local = &ctx.heap[memory_stmt.variable];
 
        let var_type = self.determine_inference_type_from_parser_type_elements(&local.parser_type.elements, true);
 
        self.var_types.insert(memory_stmt.variable, VarData::new_local(var_type));
 

	
 
        Ok(())
 
    }
 

	
 
    fn visit_local_channel_stmt(&mut self, ctx: &mut Ctx, id: ChannelStatementId) -> VisitorResult {
 
        let channel_stmt = &ctx.heap[id];
 

	
 
        let from_local = &ctx.heap[channel_stmt.from];
 
        let from_var_type = self.determine_inference_type_from_parser_type_elements(&from_local.parser_type.elements, true);
 
        self.var_types.insert(from_local.this, VarData::new_channel(from_var_type, channel_stmt.to));
 

	
 
        let to_local = &ctx.heap[channel_stmt.to];
 
        let to_var_type = self.determine_inference_type_from_parser_type_elements(&to_local.parser_type.elements, true);
 
        self.var_types.insert(to_local.this, VarData::new_channel(to_var_type, channel_stmt.from));
 

	
 
        Ok(())
 
    }
 

	
 
    fn visit_labeled_stmt(&mut self, ctx: &mut Ctx, id: LabeledStatementId) -> VisitorResult {
 
        let labeled_stmt = &ctx.heap[id];
 
        let substmt_id = labeled_stmt.body;
 
        self.visit_stmt(ctx, substmt_id)
 
    }
 

	
 
    fn visit_if_stmt(&mut self, ctx: &mut Ctx, id: IfStatementId) -> VisitorResult {
 
        let if_stmt = &ctx.heap[id];
 

	
 
        let true_body_id = if_stmt.true_body;
 
        let false_body_id = if_stmt.false_body;
 
        let test_expr_id = if_stmt.test;
 

	
 
        self.visit_expr(ctx, test_expr_id)?;
 
        self.visit_block_stmt(ctx, true_body_id)?;
 
        if let Some(false_body_id) = false_body_id {
 
            self.visit_block_stmt(ctx, false_body_id)?;
 
        }
 

	
 
        Ok(())
 
    }
 

	
 
    fn visit_while_stmt(&mut self, ctx: &mut Ctx, id: WhileStatementId) -> VisitorResult {
 
        let while_stmt = &ctx.heap[id];
 

	
 
        let body_id = while_stmt.body;
 
        let test_expr_id = while_stmt.test;
 

	
 
        self.visit_expr(ctx, test_expr_id)?;
 
        self.visit_block_stmt(ctx, body_id)?;
 

	
 
        Ok(())
 
    }
 

	
 
    fn visit_synchronous_stmt(&mut self, ctx: &mut Ctx, id: SynchronousStatementId) -> VisitorResult {
 
        let sync_stmt = &ctx.heap[id];
 
        let body_id = sync_stmt.body;
 

	
 
        self.visit_block_stmt(ctx, body_id)
 
    }
 

	
 
    fn visit_fork_stmt(&mut self, ctx: &mut Ctx, id: ForkStatementId) -> VisitorResult {
 
        let fork_stmt = &ctx.heap[id];
 
        let left_body_id = fork_stmt.left_body;
 
        let right_body_id = fork_stmt.right_body;
 

	
 
        self.visit_block_stmt(ctx, left_body_id)?;
 
        if let Some(right_body_id) = right_body_id {
 
            self.visit_block_stmt(ctx, right_body_id)?;
 
        }
 

	
 
        Ok(())
 
    }
 

	
 
    fn visit_return_stmt(&mut self, ctx: &mut Ctx, id: ReturnStatementId) -> VisitorResult {
 
        let return_stmt = &ctx.heap[id];
 
        debug_assert_eq!(return_stmt.expressions.len(), 1);
 
        let expr_id = return_stmt.expressions[0];
 

	
 
        self.visit_expr(ctx, expr_id)
 
    }
 

	
 
    fn visit_new_stmt(&mut self, ctx: &mut Ctx, id: NewStatementId) -> VisitorResult {
 
        let new_stmt = &ctx.heap[id];
 
        let call_expr_id = new_stmt.expression;
 

	
 
        self.visit_call_expr(ctx, call_expr_id)
 
    }
 

	
 
    fn visit_expr_stmt(&mut self, ctx: &mut Ctx, id: ExpressionStatementId) -> VisitorResult {
 
        let expr_stmt = &ctx.heap[id];
 
        let subexpr_id = expr_stmt.expression;
 

	
 
        self.visit_expr(ctx, subexpr_id)
 
    }
 

	
 
    // Expressions
 

	
 
    fn visit_assignment_expr(&mut self, ctx: &mut Ctx, id: AssignmentExpressionId) -> VisitorResult {
 
        let upcast_id = id.upcast();
 
        self.insert_initial_expr_inference_type(ctx, upcast_id)?;
 

	
 
        let assign_expr = &ctx.heap[id];
 
        let left_expr_id = assign_expr.left;
 
        let right_expr_id = assign_expr.right;
 

	
 
        self.visit_expr(ctx, left_expr_id)?;
 
        self.visit_expr(ctx, right_expr_id)?;
 

	
 
        self.progress_assignment_expr(ctx, id)
 
    }
 

	
 
    fn visit_binding_expr(&mut self, ctx: &mut Ctx, id: BindingExpressionId) -> VisitorResult {
 
        let upcast_id = id.upcast();
 
        self.insert_initial_expr_inference_type(ctx, upcast_id)?;
 

	
 
        let binding_expr = &ctx.heap[id];
 
        let bound_to_id = binding_expr.bound_to;
 
        let bound_from_id = binding_expr.bound_from;
 

	
 
        self.visit_expr(ctx, bound_to_id)?;
 
        self.visit_expr(ctx, bound_from_id)?;
 

	
 
        self.progress_binding_expr(ctx, id)
 
    }
 

	
 
    fn visit_conditional_expr(&mut self, ctx: &mut Ctx, id: ConditionalExpressionId) -> VisitorResult {
 
        let upcast_id = id.upcast();
 
        self.insert_initial_expr_inference_type(ctx, upcast_id)?;
 

	
 
        let conditional_expr = &ctx.heap[id];
 
        let test_expr_id = conditional_expr.test;
 
        let true_expr_id = conditional_expr.true_expression;
 
        let false_expr_id = conditional_expr.false_expression;
 

	
 
        self.visit_expr(ctx, test_expr_id)?;
 
        self.visit_expr(ctx, true_expr_id)?;
 
        self.visit_expr(ctx, false_expr_id)?;
 

	
 
        self.progress_conditional_expr(ctx, id)
 
    }
 

	
 
    fn visit_binary_expr(&mut self, ctx: &mut Ctx, id: BinaryExpressionId) -> VisitorResult {
 
        let upcast_id = id.upcast();
 
        self.insert_initial_expr_inference_type(ctx, upcast_id)?;
 

	
 
        let binary_expr = &ctx.heap[id];
 
        let lhs_expr_id = binary_expr.left;
 
        let rhs_expr_id = binary_expr.right;
 

	
 
        self.visit_expr(ctx, lhs_expr_id)?;
 
        self.visit_expr(ctx, rhs_expr_id)?;
 

	
 
        self.progress_binary_expr(ctx, id)
 
    }
 

	
 
    fn visit_unary_expr(&mut self, ctx: &mut Ctx, id: UnaryExpressionId) -> VisitorResult {
 
        let upcast_id = id.upcast();
 
        self.insert_initial_expr_inference_type(ctx, upcast_id)?;
 

	
 
        let unary_expr = &ctx.heap[id];
 
        let arg_expr_id = unary_expr.expression;
 

	
 
        self.visit_expr(ctx, arg_expr_id)?;
 

	
 
        self.progress_unary_expr(ctx, id)
 
    }
 

	
 
    fn visit_indexing_expr(&mut self, ctx: &mut Ctx, id: IndexingExpressionId) -> VisitorResult {
 
        let upcast_id = id.upcast();
 
        self.insert_initial_expr_inference_type(ctx, upcast_id)?;
 

	
 
        let indexing_expr = &ctx.heap[id];
 
        let subject_expr_id = indexing_expr.subject;
 
        let index_expr_id = indexing_expr.index;
 

	
 
        self.visit_expr(ctx, subject_expr_id)?;
 
        self.visit_expr(ctx, index_expr_id)?;
 

	
 
        self.progress_indexing_expr(ctx, id)
 
    }
 

	
 
    fn visit_slicing_expr(&mut self, ctx: &mut Ctx, id: SlicingExpressionId) -> VisitorResult {
 
        let upcast_id = id.upcast();
 
        self.insert_initial_expr_inference_type(ctx, upcast_id)?;
 

	
 
        let slicing_expr = &ctx.heap[id];
 
        let subject_expr_id = slicing_expr.subject;
 
        let from_expr_id = slicing_expr.from_index;
 
        let to_expr_id = slicing_expr.to_index;
 

	
 
        self.visit_expr(ctx, subject_expr_id)?;
 
        self.visit_expr(ctx, from_expr_id)?;
 
        self.visit_expr(ctx, to_expr_id)?;
 

	
 
        self.progress_slicing_expr(ctx, id)
 
    }
 

	
 
    fn visit_select_expr(&mut self, ctx: &mut Ctx, id: SelectExpressionId) -> VisitorResult {
 
        let upcast_id = id.upcast();
 
        self.insert_initial_expr_inference_type(ctx, upcast_id)?;
 

	
 
        let select_expr = &ctx.heap[id];
 
        let subject_expr_id = select_expr.subject;
 

	
 
        self.visit_expr(ctx, subject_expr_id)?;
 

	
 
        self.progress_select_expr(ctx, id)
 
    }
 

	
 
    fn visit_literal_expr(&mut self, ctx: &mut Ctx, id: LiteralExpressionId) -> VisitorResult {
 
        let upcast_id = id.upcast();
 
        self.insert_initial_expr_inference_type(ctx, upcast_id)?;
 

	
 
        let literal_expr = &ctx.heap[id];
 
        match &literal_expr.value {
 
            Literal::Null | Literal::False | Literal::True |
 
            Literal::Integer(_) | Literal::Character(_) | Literal::String(_) => {
 
                // No subexpressions
 
            },
 
            Literal::Struct(literal) => {
 
                // TODO: @performance
src/protocol/parser/token_parsing.rs
Show inline comments
 
use crate::collections::ScopedSection;
 
use crate::protocol::ast::*;
 
use crate::protocol::input_source::{
 
    InputSource as InputSource,
 
    InputPosition as InputPosition,
 
    InputSpan,
 
    ParseError,
 
};
 
use super::tokens::*;
 
use super::symbol_table::*;
 
use super::{Module, PassCtx};
 

	
 
// Keywords
 
pub(crate) const KW_LET:       &'static [u8] = b"let";
 
pub(crate) const KW_AS:        &'static [u8] = b"as";
 
pub(crate) const KW_STRUCT:    &'static [u8] = b"struct";
 
pub(crate) const KW_ENUM:      &'static [u8] = b"enum";
 
pub(crate) const KW_UNION:     &'static [u8] = b"union";
 
pub(crate) const KW_FUNCTION:  &'static [u8] = b"func";
 
pub(crate) const KW_PRIMITIVE: &'static [u8] = b"primitive";
 
pub(crate) const KW_COMPOSITE: &'static [u8] = b"composite";
 
pub(crate) const KW_IMPORT:    &'static [u8] = b"import";
 

	
 
// Keywords - literals
 
pub(crate) const KW_LIT_TRUE:  &'static [u8] = b"true";
 
pub(crate) const KW_LIT_FALSE: &'static [u8] = b"false";
 
pub(crate) const KW_LIT_NULL:  &'static [u8] = b"null";
 

	
 
// Keywords - function(like)s
 
pub(crate) const KW_CAST:        &'static [u8] = b"cast";
 
pub(crate) const KW_FUNC_GET:    &'static [u8] = b"get";
 
pub(crate) const KW_FUNC_PUT:    &'static [u8] = b"put";
 
pub(crate) const KW_FUNC_FIRES:  &'static [u8] = b"fires";
 
pub(crate) const KW_FUNC_CREATE: &'static [u8] = b"create";
 
pub(crate) const KW_FUNC_LENGTH: &'static [u8] = b"length";
 
pub(crate) const KW_FUNC_ASSERT: &'static [u8] = b"assert";
 
pub(crate) const KW_FUNC_PRINT:  &'static [u8] = b"print";
 

	
 
// Keywords - statements
 
pub(crate) const KW_STMT_CHANNEL:  &'static [u8] = b"channel";
 
pub(crate) const KW_STMT_IF:       &'static [u8] = b"if";
 
pub(crate) const KW_STMT_ELSE:     &'static [u8] = b"else";
 
pub(crate) const KW_STMT_WHILE:    &'static [u8] = b"while";
 
pub(crate) const KW_STMT_BREAK:    &'static [u8] = b"break";
 
pub(crate) const KW_STMT_CONTINUE: &'static [u8] = b"continue";
 
pub(crate) const KW_STMT_GOTO:     &'static [u8] = b"goto";
 
pub(crate) const KW_STMT_RETURN:   &'static [u8] = b"return";
 
pub(crate) const KW_STMT_SYNC:     &'static [u8] = b"sync";
 
pub(crate) const KW_STMT_FORK:     &'static [u8] = b"fork";
 
pub(crate) const KW_STMT_OR:       &'static [u8] = b"or";
 
pub(crate) const KW_STMT_NEW:      &'static [u8] = b"new";
 

	
 
// Keywords - types
 
// Since types are needed for returning diagnostic information to the user, the
 
// string variants are put here as well.
 
pub(crate) const KW_TYPE_IN_PORT_STR:  &'static str = "in";
 
pub(crate) const KW_TYPE_OUT_PORT_STR: &'static str = "out";
 
pub(crate) const KW_TYPE_MESSAGE_STR:  &'static str = "msg";
 
pub(crate) const KW_TYPE_BOOL_STR:     &'static str = "bool";
 
pub(crate) const KW_TYPE_UINT8_STR:    &'static str = "u8";
 
pub(crate) const KW_TYPE_UINT16_STR:   &'static str = "u16";
 
pub(crate) const KW_TYPE_UINT32_STR:   &'static str = "u32";
 
pub(crate) const KW_TYPE_UINT64_STR:   &'static str = "u64";
 
pub(crate) const KW_TYPE_SINT8_STR:    &'static str = "s8";
 
pub(crate) const KW_TYPE_SINT16_STR:   &'static str = "s16";
 
pub(crate) const KW_TYPE_SINT32_STR:   &'static str = "s32";
 
pub(crate) const KW_TYPE_SINT64_STR:   &'static str = "s64";
 
pub(crate) const KW_TYPE_CHAR_STR:     &'static str = "char";
 
pub(crate) const KW_TYPE_STRING_STR:   &'static str = "string";
 
pub(crate) const KW_TYPE_INFERRED_STR: &'static str = "auto";
 

	
 
pub(crate) const KW_TYPE_IN_PORT:  &'static [u8] = KW_TYPE_IN_PORT_STR.as_bytes();
 
pub(crate) const KW_TYPE_OUT_PORT: &'static [u8] = KW_TYPE_OUT_PORT_STR.as_bytes();
 
pub(crate) const KW_TYPE_MESSAGE:  &'static [u8] = KW_TYPE_MESSAGE_STR.as_bytes();
 
pub(crate) const KW_TYPE_BOOL:     &'static [u8] = KW_TYPE_BOOL_STR.as_bytes();
 
pub(crate) const KW_TYPE_UINT8:    &'static [u8] = KW_TYPE_UINT8_STR.as_bytes();
 
pub(crate) const KW_TYPE_UINT16:   &'static [u8] = KW_TYPE_UINT16_STR.as_bytes();
 
pub(crate) const KW_TYPE_UINT32:   &'static [u8] = KW_TYPE_UINT32_STR.as_bytes();
 
pub(crate) const KW_TYPE_UINT64:   &'static [u8] = KW_TYPE_UINT64_STR.as_bytes();
 
pub(crate) const KW_TYPE_SINT8:    &'static [u8] = KW_TYPE_SINT8_STR.as_bytes();
 
pub(crate) const KW_TYPE_SINT16:   &'static [u8] = KW_TYPE_SINT16_STR.as_bytes();
 
pub(crate) const KW_TYPE_SINT32:   &'static [u8] = KW_TYPE_SINT32_STR.as_bytes();
 
pub(crate) const KW_TYPE_SINT64:   &'static [u8] = KW_TYPE_SINT64_STR.as_bytes();
 
pub(crate) const KW_TYPE_CHAR:     &'static [u8] = KW_TYPE_CHAR_STR.as_bytes();
 
pub(crate) const KW_TYPE_STRING:   &'static [u8] = KW_TYPE_STRING_STR.as_bytes();
 
pub(crate) const KW_TYPE_INFERRED: &'static [u8] = KW_TYPE_INFERRED_STR.as_bytes();
 

	
 
/// A special trait for when consuming comma-separated things such that we can
 
/// push them onto a `Vec` and onto a `ScopedSection`. As we monomorph for
 
/// very specific comma-separated cases I don't expect polymorph bloat.
 
/// Also, I really don't like this solution.
 
pub(crate) trait Extendable {
 
    type Value;
 

	
 
    fn push(&mut self, v: Self::Value);
 
}
 

	
 
impl<T> Extendable for Vec<T> {
 
    type Value = T;
 

	
 
    #[inline]
 
    fn push(&mut self, v: Self::Value) {
 
        (self as &mut Vec<T>).push(v);
 
    }
 
}
 

	
 
impl<T: Sized> Extendable for ScopedSection<T> {
 
    type Value = T;
 

	
 
    #[inline]
 
    fn push(&mut self, v: Self::Value) {
 
        (self as &mut ScopedSection<T>).push(v);
 
    }
 
}
 

	
 
/// Consumes a domain-name identifier: identifiers separated by a dot. For
 
/// simplification of later parsing and span identification the domain-name may
 
/// contain whitespace, but must reside on the same line.
 
pub(crate) fn consume_domain_ident<'a>(
 
    source: &'a InputSource, iter: &mut TokenIter
 
) -> Result<(&'a [u8], InputSpan), ParseError> {
 
    let (_, mut span) = consume_ident(source, iter)?;
 
    while let Some(TokenKind::Dot) = iter.next() {
 
        iter.consume();
 
        let (_, new_span) = consume_ident(source, iter)?;
 
        span.end = new_span.end;
 
    }
 

	
 
    // Not strictly necessary, but probably a reasonable restriction: this
 
    // simplifies parsing of module naming and imports.
 
    if span.begin.line != span.end.line {
 
        return Err(ParseError::new_error_str_at_span(source, span, "module names may not span multiple lines"));
 
    }
 

	
 
    // If module name consists of a single identifier, then it may not match any
 
    // of the reserved keywords
 
    let section = source.section_at_pos(span.begin, span.end);
 
    if is_reserved_keyword(section) {
 
        return Err(ParseError::new_error_str_at_span(source, span, "encountered reserved keyword"));
 
    }
 

	
 
    Ok((source.section_at_pos(span.begin, span.end), span))
 
}
 

	
 
/// Consumes a specific expected token. Be careful to only call this with tokens
 
/// that do not have a variable length.
 
pub(crate) fn consume_token(source: &InputSource, iter: &mut TokenIter, expected: TokenKind) -> Result<InputSpan, ParseError> {
 
    if Some(expected) != iter.next() {
 
        return Err(ParseError::new_error_at_pos(
 
            source, iter.last_valid_pos(),
 
            format!("expected '{}'", expected.token_chars())
 
        ));
 
    }
 
    let span = iter.next_span();
 
    iter.consume();
 
    Ok(span)
 
}
 

	
 
/// Consumes a comma separated list until the closing delimiter is encountered
 
pub(crate) fn consume_comma_separated_until<T, F, E>(
 
    close_delim: TokenKind, source: &InputSource, iter: &mut TokenIter, ctx: &mut PassCtx,
 
    mut consumer_fn: F, target: &mut E, item_name_and_article: &'static str,
 
    close_pos: Option<&mut InputPosition>
 
) -> Result<(), ParseError>
 
    where F: FnMut(&InputSource, &mut TokenIter, &mut PassCtx) -> Result<T, ParseError>,
 
          E: Extendable<Value=T>
 
{
 
    let mut had_comma = true;
 
    let mut next;
 
    loop {
 
        next = iter.next();
 
        if Some(close_delim) == next {
 
            if let Some(close_pos) = close_pos {
 
                // If requested return the position of the closing delimiter
 
                let (_, new_close_pos) = iter.next_positions();
 
                *close_pos = new_close_pos;
 
            }
 
            iter.consume();
 
            break;
 
        } else if !had_comma || next.is_none() {
 
            return Err(ParseError::new_error_at_pos(
 
                source, iter.last_valid_pos(),
 
                format!("expected a '{}', or {}", close_delim.token_chars(), item_name_and_article)
 
            ));
 
        }
 

	
 
        let new_item = consumer_fn(source, iter, ctx)?;
 
        target.push(new_item);
 

	
 
        next = iter.next();
 
        had_comma = next == Some(TokenKind::Comma);
 
        if had_comma {
 
            iter.consume();
 
        }
 
    }
 

	
 
    Ok(())
 
}
 

	
 
/// Consumes a comma-separated list of items if the opening delimiting token is
 
/// encountered. If not, then the iterator will remain at its current position.
 
/// Note that the potential cases may be:
 
/// - No opening delimiter encountered, then we return `false`.
 
/// - Both opening and closing delimiter encountered, but no items.
 
/// - Opening and closing delimiter encountered, and items were processed.
 
/// - Found an opening delimiter, but processing an item failed.
 
pub(crate) fn maybe_consume_comma_separated<T, F, E>(
 
    open_delim: TokenKind, close_delim: TokenKind, source: &InputSource, iter: &mut TokenIter, ctx: &mut PassCtx,
 
    consumer_fn: F, target: &mut E, item_name_and_article: &'static str,
 
    close_pos: Option<&mut InputPosition>
 
) -> Result<bool, ParseError>
 
    where F: FnMut(&InputSource, &mut TokenIter, &mut PassCtx) -> Result<T, ParseError>,
 
          E: Extendable<Value=T>
 
{
 
    if Some(open_delim) != iter.next() {
 
        return Ok(false);
 
    }
 

	
 
    // Opening delimiter encountered, so must parse the comma-separated list.
 
    iter.consume();
 
    consume_comma_separated_until(close_delim, source, iter, ctx, consumer_fn, target, item_name_and_article, close_pos)?;
 

	
 
    Ok(true)
 
}
 

	
 
pub(crate) fn maybe_consume_comma_separated_spilled<F: FnMut(&InputSource, &mut TokenIter, &mut PassCtx) -> Result<(), ParseError>>(
 
    open_delim: TokenKind, close_delim: TokenKind, source: &InputSource,
 
    iter: &mut TokenIter, ctx: &mut PassCtx,
 
    mut consumer_fn: F, item_name_and_article: &'static str
 
) -> Result<bool, ParseError> {
 
    let mut next = iter.next();
 
    if Some(open_delim) != next {
 
        return Ok(false);
 
    }
 

	
 
    iter.consume();
 
    let mut had_comma = true;
 
    loop {
 
        next = iter.next();
 
        if Some(close_delim) == next {
 
            iter.consume();
 
            break;
 
        } else if !had_comma {
 
            return Err(ParseError::new_error_at_pos(
 
                source, iter.last_valid_pos(),
 
                format!("expected a '{}', or {}", close_delim.token_chars(), item_name_and_article)
 
            ));
 
        }
 

	
 
        consumer_fn(source, iter, ctx)?;
 
        next = iter.next();
 
        had_comma = next == Some(TokenKind::Comma);
 
        if had_comma {
 
            iter.consume();
 
        }
 
    }
 

	
 
    Ok(true)
 
}
 

	
 
/// Consumes a comma-separated list and expected the opening and closing
 
/// characters to be present. The returned array may still be empty
 
pub(crate) fn consume_comma_separated<T, F, E>(
 
    open_delim: TokenKind, close_delim: TokenKind, source: &InputSource,
 
    iter: &mut TokenIter, ctx: &mut PassCtx,
 
    consumer_fn: F, target: &mut E, item_name_and_article: &'static str,
 
    list_name_and_article: &'static str, close_pos: Option<&mut InputPosition>
 
) -> Result<(), ParseError>
 
    where F: FnMut(&InputSource, &mut TokenIter, &mut PassCtx) -> Result<T, ParseError>,
 
          E: Extendable<Value=T>
 
{
 
    let first_pos = iter.last_valid_pos();
 
    match maybe_consume_comma_separated(
 
        open_delim, close_delim, source, iter, ctx, consumer_fn, target,
 
        item_name_and_article, close_pos
 
    ) {
 
        Ok(true) => Ok(()),
 
        Ok(false) => {
 
            return Err(ParseError::new_error_at_pos(
 
                source, first_pos,
 
                format!("expected {}", list_name_and_article)
 
            ));
 
        },
 
        Err(err) => Err(err)
 
    }
 
}
 

	
 
/// Consumes an integer literal, may be binary, octal, hexadecimal or decimal,
 
/// and may have separating '_'-characters.
 
/// TODO: @Cleanup, @Performance
 
pub(crate) fn consume_integer_literal(source: &InputSource, iter: &mut TokenIter, buffer: &mut String) -> Result<(u64, InputSpan), ParseError> {
 
    if Some(TokenKind::Integer) != iter.next() {
 
        return Err(ParseError::new_error_str_at_pos(source, iter.last_valid_pos(), "expected an integer literal"));
 
    }
 
    let integer_span = iter.next_span();
 
    iter.consume();
 

	
 
    let integer_text = source.section_at_span(integer_span);
 

	
 
    // Determine radix and offset from prefix
 
    let (radix, input_offset, radix_name) =
 
        if integer_text.starts_with(b"0b") || integer_text.starts_with(b"0B") {
 
            // Binary number
 
            (2, 2, "binary")
 
        } else if integer_text.starts_with(b"0o") || integer_text.starts_with(b"0O") {
 
            // Octal number
 
            (8, 2, "octal")
 
        } else if integer_text.starts_with(b"0x") || integer_text.starts_with(b"0X") {
 
            // Hexadecimal number
 
            (16, 2, "hexadecimal")
 
        } else {
 
            (10, 0, "decimal")
 
        };
 

	
 
    // Take out any of the separating '_' characters
 
    buffer.clear();
 
    for char_idx in input_offset..integer_text.len() {
 
        let char = integer_text[char_idx];
 
        if char == b'_' {
 
            continue;
 
        }
 

	
 
        if !((char >= b'0' && char <= b'9') || (char >= b'A' && char <= b'F') || (char >= b'a' || char <= b'f')) {
 
            return Err(ParseError::new_error_at_span(
 
                source, integer_span,
 
                format!("incorrectly formatted {} number", radix_name)
 
            ));
 
        }
 
        buffer.push(char::from(char));
 
    }
 

	
 
    // Use the cleaned up string to convert to integer
 
    match u64::from_str_radix(&buffer, radix) {
 
        Ok(number) => Ok((number, integer_span)),
 
        Err(_) => Err(ParseError::new_error_at_span(
 
            source, integer_span,
 
            format!("incorrectly formatted {} number", radix_name)
 
        )),
 
    }
 
}
 

	
 
/// Consumes a character literal. We currently support a limited number of
 
/// backslash-escaped characters
 
pub(crate) fn consume_character_literal(
 
    source: &InputSource, iter: &mut TokenIter
 
) -> Result<(char, InputSpan), ParseError> {
 
    if Some(TokenKind::Character) != iter.next() {
 
        return Err(ParseError::new_error_str_at_pos(source, iter.last_valid_pos(), "expected a character literal"));
 
    }
 
    let span = iter.next_span();
 
    iter.consume();
 

	
 
    let char_text = source.section_at_span(span);
 
    if !char_text.is_ascii() {
 
        return Err(ParseError::new_error_str_at_span(
 
            source, span, "expected an ASCII character literal"
 
        ));
 
    }
 

	
 
    match char_text.len() {
 
        0 => return Err(ParseError::new_error_str_at_span(source, span, "too little characters in character literal")),
 
        1 => {
 
            // We already know the text is ascii, so just throw an error if we have the escape
 
            // character.
 
            if char_text[0] == b'\\' {
 
                return Err(ParseError::new_error_str_at_span(source, span, "escape character without subsequent character"));
 
            }
 
            return Ok((char_text[0] as char, span));
 
        },
 
        2 => {
 
            if char_text[0] == b'\\' {
 
                let result = parse_escaped_character(source, span, char_text[1])?;
 
                return Ok((result, span))
 
            }
 
        },
 
        _ => {}
 
    }
 

	
 
    return Err(ParseError::new_error_str_at_span(source, span, "too many characters in character literal"))
 
}
 

	
 
/// Consumes a string literal. We currently support a limited number of
 
/// backslash-escaped characters. Note that the result is stored in the
 
/// buffer.
 
pub(crate) fn consume_string_literal(
 
    source: &InputSource, iter: &mut TokenIter, buffer: &mut String
 
) -> Result<InputSpan, ParseError> {
 
    if Some(TokenKind::String) != iter.next() {
 
        return Err(ParseError::new_error_str_at_pos(source, iter.last_valid_pos(), "expected a string literal"));
 
    }
 

	
 
    buffer.clear();
 
    let span = iter.next_span();
 
    iter.consume();
 

	
 
    let text = source.section_at_span(span);
 
    if !text.is_ascii() {
 
        return Err(ParseError::new_error_str_at_span(source, span, "expected an ASCII string literal"));
 
    }
 

	
 
    debug_assert_eq!(text[0], b'"'); // here as kind of a reminder: the span includes the bounding quotation marks
 
    debug_assert_eq!(text[text.len() - 1], b'"');
 

	
 
    buffer.reserve(text.len() - 2);
 

	
 
    let mut was_escape = false;
 
    for idx in 1..text.len() - 1 {
 
        let cur = text[idx];
 
        let is_escape = cur == b'\\';
 
        if was_escape {
 
            let to_push = parse_escaped_character(source, span, cur)?;
 
            buffer.push(to_push);
 
        } else {
 
            buffer.push(cur as char);
 
        }
 

	
 
        if was_escape && is_escape {
 
            was_escape = false;
 
        } else {
 
            was_escape = is_escape;
 
        }
 
    }
 

	
 
    debug_assert!(!was_escape); // because otherwise we couldn't have ended the string literal
 

	
 
    Ok(span)
 
}
 

	
 
fn parse_escaped_character(source: &InputSource, literal_span: InputSpan, v: u8) -> Result<char, ParseError> {
 
    let result = match v {
 
        b'r' => '\r',
 
        b'n' => '\n',
 
        b't' => '\t',
 
        b'0' => '\0',
 
        b'\\' => '\\',
 
        b'\'' => '\'',
 
        b'"' => '"',
 
        v => {
 
            let msg = if v.is_ascii_graphic() {
 
                format!("unsupported escape character '{}'", v as char)
 
            } else {
 
                format!("unsupported escape character with (unsigned) byte value {}", v)
 
            };
 
            return Err(ParseError::new_error_at_span(source, literal_span, msg))
 
        },
 
    };
 
    Ok(result)
 
}
 

	
 
pub(crate) fn consume_pragma<'a>(source: &'a InputSource, iter: &mut TokenIter) -> Result<(&'a [u8], InputPosition, InputPosition), ParseError> {
 
    if Some(TokenKind::Pragma) != iter.next() {
 
        return Err(ParseError::new_error_str_at_pos(source, iter.last_valid_pos(), "expected a pragma"));
 
    }
 
    let (pragma_start, pragma_end) = iter.next_positions();
 
    iter.consume();
 
    Ok((source.section_at_pos(pragma_start, pragma_end), pragma_start, pragma_end))
 
}
 

	
 
pub(crate) fn has_ident(source: &InputSource, iter: &mut TokenIter, expected: &[u8]) -> bool {
 
    peek_ident(source, iter).map_or(false, |section| section == expected)
 
}
 

	
 
pub(crate) fn peek_ident<'a>(source: &'a InputSource, iter: &mut TokenIter) -> Option<&'a [u8]> {
 
    if Some(TokenKind::Ident) == iter.next() {
 
        let (start, end) = iter.next_positions();
 
        return Some(source.section_at_pos(start, end))
 
    }
 

	
 
    None
 
}
 

	
 
/// Consumes any identifier and returns it together with its span. Does not
 
/// check if the identifier is a reserved keyword.
 
pub(crate) fn consume_any_ident<'a>(
 
    source: &'a InputSource, iter: &mut TokenIter
 
) -> Result<(&'a [u8], InputSpan), ParseError> {
 
    if Some(TokenKind::Ident) != iter.next() {
 
        return Err(ParseError::new_error_str_at_pos(source, iter.last_valid_pos(), "expected an identifier"));
 
    }
 
    let (ident_start, ident_end) = iter.next_positions();
 
    iter.consume();
 
    Ok((source.section_at_pos(ident_start, ident_end), InputSpan::from_positions(ident_start, ident_end)))
 
}
 

	
 
/// Consumes a specific identifier. May or may not be a reserved keyword.
 
pub(crate) fn consume_exact_ident(source: &InputSource, iter: &mut TokenIter, expected: &[u8]) -> Result<InputSpan, ParseError> {
 
    let (ident, pos) = consume_any_ident(source, iter)?;
 
    if ident != expected {
 
        debug_assert!(expected.is_ascii());
 
        return Err(ParseError::new_error_at_pos(
 
            source, iter.last_valid_pos(),
 
            format!("expected the text '{}'", &String::from_utf8_lossy(expected))
 
        ));
 
    }
 
    Ok(pos)
 
}
 

	
 
/// Consumes an identifier that is not a reserved keyword and returns it
 
/// together with its span.
 
pub(crate) fn consume_ident<'a>(
 
    source: &'a InputSource, iter: &mut TokenIter
 
) -> Result<(&'a [u8], InputSpan), ParseError> {
 
    let (ident, span) = consume_any_ident(source, iter)?;
 
    if is_reserved_keyword(ident) {
 
        return Err(ParseError::new_error_str_at_span(source, span, "encountered reserved keyword"));
 
    }
 

	
 
    Ok((ident, span))
 
}
 

	
 
/// Consumes an identifier and immediately intern it into the `StringPool`
 
pub(crate) fn consume_ident_interned(
 
    source: &InputSource, iter: &mut TokenIter, ctx: &mut PassCtx
 
) -> Result<Identifier, ParseError> {
 
    let (value, span) = consume_ident(source, iter)?;
 
    let value = ctx.pool.intern(value);
 
    Ok(Identifier{ span, value })
 
}
 

	
 
fn is_reserved_definition_keyword(text: &[u8]) -> bool {
 
    match text {
 
        KW_STRUCT | KW_ENUM | KW_UNION | KW_FUNCTION | KW_PRIMITIVE | KW_COMPOSITE => true,
 
        _ => false,
 
    }
 
}
 

	
 
fn is_reserved_statement_keyword(text: &[u8]) -> bool {
 
    match text {
 
        KW_IMPORT | KW_AS |
 
        KW_STMT_CHANNEL | KW_STMT_IF | KW_STMT_WHILE |
 
        KW_STMT_BREAK | KW_STMT_CONTINUE | KW_STMT_GOTO | KW_STMT_RETURN |
 
        KW_STMT_SYNC | KW_STMT_FORK | KW_STMT_NEW => true,
 
        _ => false,
 
    }
 
}
 

	
 
fn is_reserved_expression_keyword(text: &[u8]) -> bool {
 
    match text {
 
        KW_LET | KW_CAST |
 
        KW_LIT_TRUE | KW_LIT_FALSE | KW_LIT_NULL |
 
        KW_FUNC_GET | KW_FUNC_PUT | KW_FUNC_FIRES | KW_FUNC_CREATE | KW_FUNC_ASSERT | KW_FUNC_LENGTH | KW_FUNC_PRINT => true,
 
        _ => false,
 
    }
 
}
 

	
 
fn is_reserved_type_keyword(text: &[u8]) -> bool {
 
    match text {
 
        KW_TYPE_IN_PORT | KW_TYPE_OUT_PORT | KW_TYPE_MESSAGE | KW_TYPE_BOOL |
 
        KW_TYPE_UINT8 | KW_TYPE_UINT16 | KW_TYPE_UINT32 | KW_TYPE_UINT64 |
 
        KW_TYPE_SINT8 | KW_TYPE_SINT16 | KW_TYPE_SINT32 | KW_TYPE_SINT64 |
 
        KW_TYPE_CHAR | KW_TYPE_STRING |
 
        KW_TYPE_INFERRED => true,
 
        _ => false,
 
    }
 
}
 

	
 
fn is_reserved_keyword(text: &[u8]) -> bool {
 
    return
 
        is_reserved_definition_keyword(text) ||
 
        is_reserved_statement_keyword(text) ||
 
        is_reserved_expression_keyword(text) ||
 
        is_reserved_type_keyword(text);
 
}
 

	
 
pub(crate) fn seek_module(modules: &[Module], root_id: RootId) -> Option<&Module> {
 
    for module in modules {
 
        if module.root_id == root_id {
 
            return Some(module)
 
        }
 
    }
 

	
 
    return None
 
}
 

	
 
/// Constructs a human-readable message indicating why there is a conflict of
 
/// symbols.
 
// Note: passing the `module_idx` is not strictly necessary, but will prevent
 
// programmer mistakes during development: we get a conflict because we're
 
// currently parsing a particular module.
 
pub(crate) fn construct_symbol_conflict_error(
 
    modules: &[Module], module_idx: usize, ctx: &PassCtx, new_symbol: &Symbol, old_symbol: &Symbol
 
) -> ParseError {
 
    let module = &modules[module_idx];
 
    let get_symbol_span_and_msg = |symbol: &Symbol| -> (String, Option<InputSpan>) {
 
        match &symbol.variant {
 
            SymbolVariant::Module(module) => {
 
                let import = &ctx.heap[module.introduced_at];
 
                return (
 
                    format!("the module aliased as '{}' imported here", symbol.name.as_str()),
 
                    Some(import.as_module().span)
 
                );
 
            },
 
            SymbolVariant::Definition(definition) => {
 
                if definition.defined_in_module.is_invalid() {
 
                    // Must be a builtin thing
 
                    return (format!("the builtin '{}'", symbol.name.as_str()), None)
 
                } else {
 
                    if let Some(import_id) = definition.imported_at {
 
                        let import = &ctx.heap[import_id];
 
                        return (
 
                            format!("the type '{}' imported here", symbol.name.as_str()),
 
                            Some(import.as_symbols().span)
 
                        );
 
                    } else {
 
                        // This is a defined symbol. So this must mean that the
 
                        // error was caused by it being defined.
 
                        debug_assert_eq!(definition.defined_in_module, module.root_id);
 

	
 
                        return (
 
                            format!("the type '{}' defined here", symbol.name.as_str()),
 
                            Some(definition.identifier_span)
 
                        )
 
                    }
 
                }
 
            }
 
        }
 
    };
 

	
 
    let (new_symbol_msg, new_symbol_span) = get_symbol_span_and_msg(new_symbol);
 
    let (old_symbol_msg, old_symbol_span) = get_symbol_span_and_msg(old_symbol);
 
    let new_symbol_span = new_symbol_span.unwrap(); // because new symbols cannot be builtin
 

	
 
    match old_symbol_span {
 
        Some(old_symbol_span) => ParseError::new_error_at_span(
 
            &module.source, new_symbol_span, format!("symbol is defined twice: {}", new_symbol_msg)
 
        ).with_info_at_span(
 
            &module.source, old_symbol_span, format!("it conflicts with {}", old_symbol_msg)
 
        ),
 
        None => ParseError::new_error_at_span(
 
            &module.source, new_symbol_span,
 
            format!("symbol is defined twice: {} conflicts with {}", new_symbol_msg, old_symbol_msg)
 
        )
 
    }
 
}
 
\ No newline at end of file
src/protocol/parser/tokens.rs
Show inline comments
 
use crate::protocol::input_source::{
 
    InputPosition as InputPosition,
 
    InputSpan
 
};
 

	
 
/// Represents a particular kind of token. Some tokens represent
 
/// variable-character tokens. Such a token is always followed by a
 
/// `TokenKind::SpanEnd` token.
 
#[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord)]
 
pub enum TokenKind {
 
    // Variable-character tokens, followed by a SpanEnd token
 
    Ident,          // regular identifier
 
    Pragma,         // identifier with prefixed `#`, range includes `#`
 
    Integer,        // integer literal
 
    String,         // string literal, range includes `"`
 
    Character,      // character literal, range includes `'`
 
    LineComment,    // line comment, range includes leading `//`, but not newline
 
    BlockComment,   // block comment, range includes leading `/*` and trailing `*/`
 
    // Punctuation (single character)
 
    Exclamation,    // !
 
    Question,       // ?
 
    Pound,          // #
 
    OpenAngle,      // <
 
    OpenCurly,      // {
 
    OpenParen,      // (
 
    OpenSquare,     // [
 
    CloseAngle,     // >
 
    CloseCurly,     // }
 
    CloseParen,     // )
 
    CloseSquare,    // ]
 
    Colon,          // :
 
    Comma,          // ,
 
    Dot,            // .
 
    SemiColon,      // ;
 
    // Operator-like (single character)
 
    At,             // @
 
    Plus,           // +
 
    Minus,          // -
 
    Star,           // *
 
    Slash,          // /
 
    Percent,        // %
 
    Caret,          // ^
 
    And,            // &
 
    Or,             // |
 
    Tilde,          // ~
 
    Equal,          // =
 
    // Punctuation (two characters)
 
    ColonColon,     // ::
 
    DotDot,         // ..
 
    ArrowRight,     // ->
 
    // Operator-like (two characters)
 
    AtEquals,       // @=
 
    PlusPlus,       // ++
 
    PlusEquals,     // +=
 
    MinusMinus,     // --
 
    MinusEquals,    // -=
 
    StarEquals,     // *=
 
    SlashEquals,    // /=
 
    PercentEquals,  // %=
 
    CaretEquals,    // ^=
 
    AndAnd,         // &&
 
    AndEquals,      // &=
 
    OrOr,           // ||
 
    OrEquals,       // |=
 
    EqualEqual,     // ==
 
    NotEqual,       // !=
 
    ShiftLeft,      // <<
 
    LessEquals,     // <=
 
    ShiftRight,     // >>
 
    GreaterEquals,  // >=
 
    // Operator-like (three characters)
 
    ShiftLeftEquals,// <<=
 
    ShiftRightEquals, // >>=
 
    // Special marker token to indicate end of variable-character tokens
 
    SpanEnd,
 
}
 

	
 
impl TokenKind {
 
    /// Returns true if the next expected token is the special `TokenKind::SpanEnd` token. This is
 
    /// the case for tokens of variable length (e.g. an identifier).
 
    fn has_span_end(&self) -> bool {
 
        return *self <= TokenKind::BlockComment
 
    }
 

	
 
    /// Returns the number of characters associated with the token. May only be called on tokens
 
    /// that do not have a variable length.
 
    fn num_characters(&self) -> u32 {
 
        debug_assert!(!self.has_span_end() && *self != TokenKind::SpanEnd);
 
        if *self <= TokenKind::Equal {
 
            1
 
        } else if *self <= TokenKind::GreaterEquals {
 
            2
 
        } else {
 
            3
 
        }
 
    }
 

	
 
    /// Returns the characters that are represented by the token, may only be called on tokens that
 
    /// do not have a variable length.
 
    pub fn token_chars(&self) -> &'static str {
 
        debug_assert!(!self.has_span_end() && *self != TokenKind::SpanEnd);
 
        use TokenKind as TK;
 
        match self {
 
            TK::Exclamation => "!",
 
            TK::Question => "?",
 
            TK::Pound => "#",
 
            TK::OpenAngle => "<",
 
            TK::OpenCurly => "{",
 
            TK::OpenParen => "(",
 
            TK::OpenSquare => "[",
 
            TK::CloseAngle => ">",
 
            TK::CloseCurly => "}",
 
            TK::CloseParen => ")",
 
            TK::CloseSquare => "]",
 
            TK::Colon => ":",
 
            TK::Comma => ",",
 
            TK::Dot => ".",
 
            TK::SemiColon => ";",
 
            TK::At => "@",
 
            TK::Plus => "+",
 
            TK::Minus => "-",
 
            TK::Star => "*",
 
            TK::Slash => "/",
 
            TK::Percent => "%",
 
            TK::Caret => "^",
 
            TK::And => "&",
 
            TK::Or => "|",
 
            TK::Tilde => "~",
 
            TK::Equal => "=",
 
            TK::ColonColon => "::",
 
            TK::DotDot => "..",
 
            TK::ArrowRight => "->",
 
            TK::AtEquals => "@=",
 
            TK::PlusPlus => "++",
 
            TK::PlusEquals => "+=",
 
            TK::MinusMinus => "--",
 
            TK::MinusEquals => "-=",
 
            TK::StarEquals => "*=",
 
            TK::SlashEquals => "/=",
 
            TK::PercentEquals => "%=",
 
            TK::CaretEquals => "^=",
 
            TK::AndAnd => "&&",
 
            TK::AndEquals => "&=",
 
            TK::OrOr => "||",
 
            TK::OrEquals => "|=",
 
            TK::EqualEqual => "==",
 
            TK::NotEqual => "!=",
 
            TK::ShiftLeft => "<<",
 
            TK::LessEquals => "<=",
 
            TK::ShiftRight => ">>",
 
            TK::GreaterEquals => ">=",
 
            TK::ShiftLeftEquals => "<<=",
 
            TK::ShiftRightEquals => ">>=",
 
            // Lets keep these in explicitly for now, in case we want to add more symbols
 
            TK::Ident | TK::Pragma | TK::Integer | TK::String | TK::Character |
 
            TK::LineComment | TK::BlockComment | TK::SpanEnd => unreachable!(),
 
        }
 
    }
 
}
 

	
 
/// Represents a single token at a particular position.
 
pub struct Token {
 
    pub kind: TokenKind,
 
    pub pos: InputPosition,
 
}
 

	
 
impl Token {
 
    pub(crate) fn new(kind: TokenKind, pos: InputPosition) -> Self {
 
        Self{ kind, pos }
 
    }
 
}
 

	
 
/// The kind of token ranges that are specially parsed by the tokenizer.
 
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
 
pub enum TokenRangeKind {
 
    Module,
 
    Pragma,
 
    Import,
 
    Definition,
 
    Code,
 
}
 

	
 
pub const NO_RELATION: i32 = -1;
 
pub const NO_SIBLING: i32 = NO_RELATION;
 

	
 
/// A range of tokens with a specific meaning. Such a range is part of a tree
 
/// where each parent tree envelops all of its children.
 
#[derive(Debug)]
 
pub struct TokenRange {
 
    // Index of parent in `TokenBuffer.ranges`, does not have a parent if the
 
    // range kind is Module, in that case the parent index is -1.
 
    pub parent_idx: i32,
 
    pub range_kind: TokenRangeKind,
 
    pub curly_depth: u32,
 
    // Offsets into `TokenBuffer.ranges`: the tokens belonging to this range.
 
    pub start: u32,             // first token (inclusive index)
 
    pub end: u32,               // last token (exclusive index)
 
    // Child ranges
 
    pub num_child_ranges: u32,  // Number of subranges
 
    pub first_child_idx: i32,   // First subrange (or -1 if no subranges)
 
    pub last_child_idx: i32,    // Last subrange (or -1 if no subranges)
 
    pub next_sibling_idx: i32,  // Next subrange (or -1 if no next subrange)
 
}
 

	
 
pub struct TokenBuffer {
 
    pub tokens: Vec<Token>,
 
    pub ranges: Vec<TokenRange>,
 
}
 

	
 
impl TokenBuffer {
 
    pub(crate) fn new() -> Self {
 
        Self{ tokens: Vec::new(), ranges: Vec::new() }
 
    }
 

	
 
    pub(crate) fn iter_range<'a>(&'a self, range: &TokenRange) -> TokenIter<'a> {
 
        TokenIter::new(self, range.start as usize, range.end as usize)
 
    }
 

	
 
    pub(crate) fn start_pos(&self, range: &TokenRange) -> InputPosition {
 
        self.tokens[range.start as usize].pos
 
    }
 

	
 
    pub(crate) fn end_pos(&self, range: &TokenRange) -> InputPosition {
 
        let last_token = &self.tokens[range.end as usize - 1];
 
        if last_token.kind == TokenKind::SpanEnd {
 
            return last_token.pos
 
        } else {
 
            debug_assert!(!last_token.kind.has_span_end());
 
            return last_token.pos.with_offset(last_token.kind.num_characters());
 
        }
 
    }
 
}
 

	
 
/// Iterator over tokens within a specific `TokenRange`.
 
pub(crate) struct TokenIter<'a> {
 
    tokens: &'a Vec<Token>,
 
    cur: usize,
 
    end: usize,
 
}
 

	
 
impl<'a> TokenIter<'a> {
 
    fn new(buffer: &'a TokenBuffer, start: usize, end: usize) -> Self {
 
        Self{ tokens: &buffer.tokens, cur: start, end }
 
    }
 

	
 
    /// Returns the next token (may include comments), or `None` if at the end
 
    /// of the range.
 
    pub(crate) fn next_including_comments(&self) -> Option<TokenKind> {
 
        if self.cur >= self.end {
 
            return None;
 
        }
 

	
 
        let token = &self.tokens[self.cur];
 
        Some(token.kind)
 
    }
 

	
 
    /// Returns the next token (but skips over comments), or `None` if at the
 
    /// end of the range
 
    pub(crate) fn next(&mut self) -> Option<TokenKind> {
 
        while let Some(token_kind) = self.next_including_comments() {
 
            if token_kind != TokenKind::LineComment && token_kind != TokenKind::BlockComment {
 
                return Some(token_kind);
 
            }
 
            self.consume();
 
        }
 

	
 
        return None
 
    }
 

	
 
    /// Peeks ahead by one token (i.e. the one that comes after `next()`), and
 
    /// skips over comments
 
    pub(crate) fn peek(&self) -> Option<TokenKind> {
 
        for next_idx in self.cur + 1..self.end {
 
            let next_kind = self.tokens[next_idx].kind;
 
            if next_kind != TokenKind::LineComment && next_kind != TokenKind::BlockComment && next_kind != TokenKind::SpanEnd {
 
                return Some(next_kind);
 
            }
 
        }
 

	
 
        return None;
 
    }
 

	
 
    /// Returns the start position belonging to the token returned by `next`. If
 
    /// there is not a next token, then we return the end position of the
 
    /// previous token.
 
    pub(crate) fn last_valid_pos(&self) -> InputPosition {
 
        if self.cur < self.end {
 
            // Return token position
 
            return self.tokens[self.cur].pos
 
        }
 

	
 
        // Return previous token end
 
        let token = &self.tokens[self.cur - 1];
 
        return if token.kind == TokenKind::SpanEnd {
 
            token.pos
 
        } else {
 
            token.pos.with_offset(token.kind.num_characters())
 
        };
 
    }
 

	
 
    /// Returns the token range belonging to the token returned by `next`. This
 
    /// assumes that we're not at the end of the range we're iterating over.
 
    /// TODO: @cleanup Phase out?
 
    pub(crate) fn next_positions(&self) -> (InputPosition, InputPosition) {
 
        debug_assert!(self.cur < self.end);
 
        let token = &self.tokens[self.cur];
 
        if token.kind.has_span_end() {
 
            let span_end = &self.tokens[self.cur + 1];
 
            debug_assert_eq!(span_end.kind, TokenKind::SpanEnd);
 
            (token.pos, span_end.pos)
 
        } else {
 
            let offset = token.kind.num_characters();
 
            (token.pos, token.pos.with_offset(offset))
 
        }
 
    }
 

	
 
    /// See `next_positions`
 
    pub(crate) fn next_span(&self) -> InputSpan {
 
        let (begin, end) = self.next_positions();
 
        return InputSpan::from_positions(begin, end)
 
    }
 

	
 
    /// Advances the iterator to the next (meaningful) token.
 
    pub(crate) fn consume(&mut self) {
 
        if let Some(kind) = self.next_including_comments() {
 
            if kind.has_span_end() {
 
                self.cur += 2;
 
            } else {
 
                self.cur += 1;
 
            }
 
        }
 
    }
 

	
 
    /// Saves the current iteration position, may be passed to `load` to return
 
    /// the iterator to a previous position.
 
    pub(crate) fn save(&self) -> (usize, usize) {
 
        (self.cur, self.end)
 
    }
 

	
 
    pub(crate) fn load(&mut self, saved: (usize, usize)) {
 
        self.cur = saved.0;
 
        self.end = saved.1;
 
    }
 
}
 
\ No newline at end of file
src/protocol/parser/type_table.rs
Show inline comments
 
@@ -494,769 +494,769 @@ enum MemoryLayoutResult {
 
    TypeExists(usize, usize), // (size, alignment)
 
    PushBreadcrumb(MemoryBreadcrumb),
 
}
 

	
 
// TODO: @Optimize, initial memory-unoptimized implementation
 
struct TypeLoopEntry {
 
    definition_id: DefinitionId,
 
    monomorph_idx: usize,
 
    is_union: bool,
 
}
 

	
 
struct TypeLoop {
 
    members: Vec<TypeLoopEntry>
 
}
 

	
 
pub struct TypeTable {
 
    /// Lookup from AST DefinitionId to a defined type. Considering possible
 
    /// polymorphs is done inside the `DefinedType` struct.
 
    lookup: HashMap<DefinitionId, DefinedType>,
 
    /// Breadcrumbs left behind while trying to find type loops. Also used to
 
    /// determine sizes of types when all type loops are detected.
 
    type_loop_breadcrumbs: Vec<TypeLoopBreadcrumb>,
 
    type_loops: Vec<TypeLoop>,
 
    /// Stores all encountered types during type loop detection. Used afterwards
 
    /// to iterate over all types in order to compute size/alignment.
 
    encountered_types: Vec<TypeLoopEntry>,
 
    /// Breadcrumbs and temporary storage during memory layout computation.
 
    memory_layout_breadcrumbs: Vec<MemoryBreadcrumb>,
 
    size_alignment_stack: Vec<(usize, usize)>,
 
}
 

	
 
impl TypeTable {
 
    /// Construct a new type table without any resolved types.
 
    pub(crate) fn new() -> Self {
 
        Self{ 
 
            lookup: HashMap::new(), 
 
            type_loop_breadcrumbs: Vec::with_capacity(32),
 
            type_loops: Vec::with_capacity(8),
 
            encountered_types: Vec::with_capacity(32),
 
            memory_layout_breadcrumbs: Vec::with_capacity(32),
 
            size_alignment_stack: Vec::with_capacity(64),
 
        }
 
    }
 

	
 
    /// Iterates over all defined types (polymorphic and non-polymorphic) and
 
    /// add their types in two passes. In the first pass we will just add the
 
    /// base types (we will not consider monomorphs, and we will not compute
 
    /// byte sizes). In the second pass we will compute byte sizes of
 
    /// non-polymorphic types, and potentially the monomorphs that are embedded
 
    /// in those types.
 
    pub(crate) fn build_base_types(&mut self, modules: &mut [Module], ctx: &mut PassCtx) -> Result<(), ParseError> {
 
        // Make sure we're allowed to cast root_id to index into ctx.modules
 
        debug_assert!(modules.iter().all(|m| m.phase >= ModuleCompilationPhase::DefinitionsParsed));
 
        debug_assert!(self.lookup.is_empty());
 

	
 
        if cfg!(debug_assertions) {
 
            for (index, module) in modules.iter().enumerate() {
 
                debug_assert_eq!(index, module.root_id.index as usize);
 
            }
 
        }
 

	
 
        // Use context to guess hashmap size of the base types
 
        let reserve_size = ctx.heap.definitions.len();
 
        self.lookup.reserve(reserve_size);
 

	
 
        // Resolve all base types
 
        for definition_idx in 0..ctx.heap.definitions.len() {
 
            let definition_id = ctx.heap.definitions.get_id(definition_idx);
 
            let definition = &ctx.heap[definition_id];
 

	
 
            match definition {
 
                Definition::Enum(_) => self.build_base_enum_definition(modules, ctx, definition_id)?,
 
                Definition::Union(_) => self.build_base_union_definition(modules, ctx, definition_id)?,
 
                Definition::Struct(_) => self.build_base_struct_definition(modules, ctx, definition_id)?,
 
                Definition::Function(_) => self.build_base_function_definition(modules, ctx, definition_id)?,
 
                Definition::Component(_) => self.build_base_component_definition(modules, ctx, definition_id)?,
 
            }
 
        }
 

	
 
        debug_assert_eq!(self.lookup.len(), reserve_size, "mismatch in reserved size of type table"); // NOTE: Temp fix for builtin functions
 
        for module in modules.iter_mut() {
 
            module.phase = ModuleCompilationPhase::TypesAddedToTable;
 
        }
 

	
 
        // Go through all types again, lay out all types that are not
 
        // polymorphic. This might cause us to lay out types that are monomorphs
 
        // of polymorphic types.
 
        for definition_idx in 0..ctx.heap.definitions.len() {
 
            let definition_id = ctx.heap.definitions.get_id(definition_idx);
 
            let poly_type = self.lookup.get(&definition_id).unwrap();
 

	
 
            // Here we explicitly want to instantiate types which have no
 
            // polymorphic arguments (even if it has phantom polymorphic
 
            // arguments) because otherwise the user will see very weird
 
            // error messages.
 
            if poly_type.definition.type_class().is_data_type() && poly_type.poly_vars.is_empty() && poly_type.num_monomorphs() == 0 {
 
                self.detect_and_resolve_type_loops_for(
 
                    modules, ctx.heap,
 
                    ConcreteType{
 
                        parts: vec![ConcreteTypePart::Instance(definition_id, 0)]
 
                    },
 
                )?;
 
                self.lay_out_memory_for_encountered_types(ctx.arch);
 
            }
 
        }
 

	
 
        Ok(())
 
    }
 

	
 
    /// Retrieves base definition from type table. We must be able to retrieve
 
    /// it as we resolve all base types upon type table construction (for now).
 
    /// However, in the future we might do on-demand type resolving, so return
 
    /// an option anyway
 
    pub(crate) fn get_base_definition(&self, definition_id: &DefinitionId) -> Option<&DefinedType> {
 
        self.lookup.get(&definition_id)
 
    }
 

	
 
    /// Returns the index into the monomorph type array if the procedure type
 
    /// already has a (reserved) monomorph.
 
    pub(crate) fn get_procedure_monomorph_index(&self, definition_id: &DefinitionId, types: &ConcreteType) -> Option<i32> {
 
        let def = self.lookup.get(definition_id).unwrap();
 
        let monos = def.definition.procedure_monomorphs();
 
        return monos.iter()
 
            .position(|v| v.concrete_type == *types)
 
            .map(|v| v as i32);
 
    }
 

	
 
    /// Returns a mutable reference to a procedure's monomorph expression data.
 
    /// Used by typechecker to fill in previously reserved type information
 
    pub(crate) fn get_procedure_expression_data_mut(&mut self, definition_id: &DefinitionId, monomorph_idx: i32) -> &mut ProcedureMonomorph {
 
        debug_assert!(monomorph_idx >= 0);
 
        let def = self.lookup.get_mut(definition_id).unwrap();
 
        let monomorphs = def.definition.procedure_monomorphs_mut();
 
        return &mut monomorphs[monomorph_idx as usize];
 
    }
 

	
 
    pub(crate) fn get_procedure_expression_data(&self, definition_id: &DefinitionId, monomorph_idx: i32) -> &ProcedureMonomorph {
 
        debug_assert!(monomorph_idx >= 0);
 
        let def = self.lookup.get(definition_id).unwrap();
 
        let monomorphs = def.definition.procedure_monomorphs();
 
        return &monomorphs[monomorph_idx as usize];
 
    }
 

	
 
    /// Reserves space for a monomorph of a polymorphic procedure. The index
 
    /// will point into a (reserved) slot of the array of expression types. The
 
    /// monomorph may NOT exist yet (because the reservation implies that we're
 
    /// going to be performing typechecking on it, and we don't want to
 
    /// check the same monomorph twice)
 
    pub(crate) fn reserve_procedure_monomorph_index(&mut self, definition_id: &DefinitionId, concrete_type: ConcreteType) -> i32 {
 
        let def = self.lookup.get_mut(definition_id).unwrap();
 
        let mono_types = def.definition.procedure_monomorphs_mut();
 
        debug_assert!(def.is_polymorph == (concrete_type.parts.len() != 1));
 
        debug_assert!(!mono_types.iter().any(|v| v.concrete_type == concrete_type));
 

	
 
        let mono_idx = mono_types.len();
 
        mono_types.push(ProcedureMonomorph{
 
            concrete_type,
 
            arg_types: Vec::new(),
 
            expr_data: Vec::new(),
 
        });
 

	
 
        return mono_idx as i32;
 
    }
 

	
 
    /// Adds a datatype polymorph to the type table. Will not add the
 
    /// monomorph if it is already present, or if the type's polymorphic
 
    /// variables are all unused.
 
    /// TODO: Fix signature
 
    pub(crate) fn add_data_monomorph(
 
        &mut self, modules: &[Module], heap: &Heap, arch: &TargetArch, definition_id: DefinitionId, concrete_type: ConcreteType
 
    ) -> Result<i32, ParseError> {
 
        debug_assert_eq!(definition_id, get_concrete_type_definition(&concrete_type));
 

	
 
        // Check if the monomorph already exists
 
        let poly_type = self.lookup.get_mut(&definition_id).unwrap();
 
        if let Some(idx) = poly_type.get_monomorph_index(&concrete_type) {
 
            return Ok(idx as i32);
 
        }
 

	
 
        // Doesn't exist, so instantiate a monomorph and determine its memory
 
        // layout.
 
        self.detect_and_resolve_type_loops_for(modules, heap, concrete_type)?;
 
        debug_assert_eq!(self.encountered_types[0].definition_id, definition_id);
 
        let mono_idx = self.encountered_types[0].monomorph_idx;
 
        self.lay_out_memory_for_encountered_types(arch);
 

	
 
        return Ok(mono_idx as i32);
 
    }
 

	
 
    //--------------------------------------------------------------------------
 
    // Building base types
 
    //--------------------------------------------------------------------------
 

	
 
    /// Builds the base type for an enum. Will not compute byte sizes
 
    fn build_base_enum_definition(&mut self, modules: &[Module], ctx: &mut PassCtx, definition_id: DefinitionId) -> Result<(), ParseError> {
 
        debug_assert!(!self.lookup.contains_key(&definition_id), "base enum already built");
 
        let definition = ctx.heap[definition_id].as_enum();
 
        let root_id = definition.defined_in;
 

	
 
        // Determine enum variants
 
        let mut enum_value = -1;
 
        let mut variants = Vec::with_capacity(definition.variants.len());
 

	
 
        for variant in &definition.variants {
 
            if enum_value == i64::MAX {
 
                let source = &modules[definition.defined_in.index as usize].source;
 
                return Err(ParseError::new_error_str_at_span(
 
                    source, variant.identifier.span,
 
                    "this enum variant has an integer value that is too large"
 
                ));
 
            }
 

	
 
            enum_value += 1;
 
            if let EnumVariantValue::Integer(explicit_value) = variant.value {
 
                enum_value = explicit_value;
 
            }
 

	
 
            variants.push(EnumVariant{
 
                identifier: variant.identifier.clone(),
 
                value: enum_value,
 
            });
 
        }
 

	
 
        // Determine tag size
 
        let mut min_enum_value = 0;
 
        let mut max_enum_value = 0;
 
        if !variants.is_empty() {
 
            min_enum_value = variants[0].value;
 
            max_enum_value = variants[0].value;
 
            for variant in variants.iter().skip(1) {
 
                min_enum_value = min_enum_value.min(variant.value);
 
                max_enum_value = max_enum_value.max(variant.value);
 
            }
 
        }
 

	
 
        let (tag_type, size_and_alignment) = Self::variant_tag_type_from_values(min_enum_value, max_enum_value);
 

	
 
        // Enum names and polymorphic args do not conflict
 
        Self::check_identifier_collision(
 
            modules, root_id, &variants, |variant| &variant.identifier, "enum variant"
 
        )?;
 

	
 
        // Polymorphic arguments cannot appear as embedded types, because
 
        // they can only consist of integer variants.
 
        Self::check_poly_args_collision(modules, ctx, root_id, &definition.poly_vars)?;
 
        let poly_vars = Self::create_polymorphic_variables(&definition.poly_vars);
 

	
 
        self.lookup.insert(definition_id, DefinedType {
 
            ast_root: root_id,
 
            ast_definition: definition_id,
 
            definition: DefinedTypeVariant::Enum(EnumType{
 
                variants,
 
                monomorphs: Vec::new(),
 
                minimum_tag_value: min_enum_value,
 
                maximum_tag_value: max_enum_value,
 
                tag_type,
 
                size: size_and_alignment,
 
                alignment: size_and_alignment
 
            }),
 
            poly_vars,
 
            is_polymorph: false,
 
        });
 

	
 
        return Ok(());
 
    }
 

	
 
    /// Builds the base type for a union. Will compute byte sizes.
 
    fn build_base_union_definition(&mut self, modules: &[Module], ctx: &mut PassCtx, definition_id: DefinitionId) -> Result<(), ParseError> {
 
        debug_assert!(!self.lookup.contains_key(&definition_id), "base union already built");
 
        let definition = ctx.heap[definition_id].as_union();
 
        let root_id = definition.defined_in;
 

	
 
        // Check all variants and their embedded types
 
        let mut variants = Vec::with_capacity(definition.variants.len());
 
        let mut tag_counter = 0;
 
        for variant in &definition.variants {
 
            for embedded in &variant.value {
 
                Self::check_member_parser_type(
 
                    modules, ctx, root_id, embedded, false
 
                )?;
 
            }
 

	
 
            variants.push(UnionVariant{
 
                identifier: variant.identifier.clone(),
 
                embedded: variant.value.clone(),
 
                tag_value: tag_counter,
 
            });
 
            tag_counter += 1;
 
        }
 

	
 
        let mut max_tag_value = 0;
 
        if tag_counter != 0 {
 
            max_tag_value = tag_counter - 1
 
        }
 

	
 
        let (tag_type, tag_size) = Self::variant_tag_type_from_values(0, max_tag_value);
 

	
 
        // Make sure there are no conflicts in identifiers
 
        Self::check_identifier_collision(
 
            modules, root_id, &variants, |variant| &variant.identifier, "union variant"
 
        )?;
 
        Self::check_poly_args_collision(modules, ctx, root_id, &definition.poly_vars)?;
 

	
 
        // Construct internal representation of union
 
        let mut poly_vars = Self::create_polymorphic_variables(&definition.poly_vars);
 
        for variant in &definition.variants {
 
            for embedded in &variant.value {
 
                Self::mark_used_polymorphic_variables(&mut poly_vars, embedded);
 
            }
 
        }
 

	
 
        let is_polymorph = poly_vars.iter().any(|arg| arg.is_in_use);
 

	
 
        self.lookup.insert(definition_id, DefinedType{
 
            ast_root: root_id,
 
            ast_definition: definition_id,
 
            definition: DefinedTypeVariant::Union(UnionType{
 
                variants,
 
                monomorphs: Vec::new(),
 
                tag_type,
 
                tag_size,
 
            }),
 
            poly_vars,
 
            is_polymorph
 
        });
 

	
 
        return Ok(());
 
    }
 

	
 
    /// Builds base struct type. Will not compute byte sizes.
 
    fn build_base_struct_definition(&mut self, modules: &[Module], ctx: &mut PassCtx, definition_id: DefinitionId) -> Result<(), ParseError> {
 
        debug_assert!(!self.lookup.contains_key(&definition_id), "base struct already built");
 
        let definition = ctx.heap[definition_id].as_struct();
 
        let root_id = definition.defined_in;
 

	
 
        // Check all struct fields and construct internal representation
 
        let mut fields = Vec::with_capacity(definition.fields.len());
 

	
 
        for field in &definition.fields {
 
            Self::check_member_parser_type(
 
                modules, ctx, root_id, &field.parser_type, false
 
            )?;
 

	
 
            fields.push(StructField{
 
                identifier: field.field.clone(),
 
                parser_type: field.parser_type.clone(),
 
            });
 
        }
 

	
 
        // Make sure there are no conflicting variables
 
        Self::check_identifier_collision(
 
            modules, root_id, &fields, |field| &field.identifier, "struct field"
 
        )?;
 
        Self::check_poly_args_collision(modules, ctx, root_id, &definition.poly_vars)?;
 

	
 
        // Construct base type in table
 
        let mut poly_vars = Self::create_polymorphic_variables(&definition.poly_vars);
 
        for field in &fields {
 
            Self::mark_used_polymorphic_variables(&mut poly_vars, &field.parser_type);
 
        }
 

	
 
        let is_polymorph = poly_vars.iter().any(|arg| arg.is_in_use);
 

	
 
        self.lookup.insert(definition_id, DefinedType{
 
            ast_root: root_id,
 
            ast_definition: definition_id,
 
            definition: DefinedTypeVariant::Struct(StructType{
 
                fields,
 
                monomorphs: Vec::new(),
 
            }),
 
            poly_vars,
 
            is_polymorph
 
        });
 

	
 
        return Ok(())
 
    }
 

	
 
    /// Builds base function type.
 
    fn build_base_function_definition(&mut self, modules: &[Module], ctx: &mut PassCtx, definition_id: DefinitionId) -> Result<(), ParseError> {
 
        debug_assert!(!self.lookup.contains_key(&definition_id), "base function already built");
 
        let definition = ctx.heap[definition_id].as_function();
 
        let root_id = definition.defined_in;
 

	
 
        // Check and construct return types and argument types.
 
        debug_assert_eq!(definition.return_types.len(), 1, "not one return type"); // TODO: @ReturnValues
 
        debug_assert_eq!(definition.return_types.len(), 1, "not one return type");
 
        for return_type in &definition.return_types {
 
            Self::check_member_parser_type(
 
                modules, ctx, root_id, return_type, definition.builtin
 
            )?;
 
        }
 

	
 
        let mut arguments = Vec::with_capacity(definition.parameters.len());
 
        for parameter_id in &definition.parameters {
 
            let parameter = &ctx.heap[*parameter_id];
 
            Self::check_member_parser_type(
 
                modules, ctx, root_id, &parameter.parser_type, definition.builtin
 
            )?;
 

	
 
            arguments.push(FunctionArgument{
 
                identifier: parameter.identifier.clone(),
 
                parser_type: parameter.parser_type.clone(),
 
            });
 
        }
 

	
 
        // Check conflict of identifiers
 
        Self::check_identifier_collision(
 
            modules, root_id, &arguments, |arg| &arg.identifier, "function argument"
 
        )?;
 
        Self::check_poly_args_collision(modules, ctx, root_id, &definition.poly_vars)?;
 

	
 
        // Construct internal representation of function type
 
        let mut poly_vars = Self::create_polymorphic_variables(&definition.poly_vars);
 
        for return_type in &definition.return_types {
 
            Self::mark_used_polymorphic_variables(&mut poly_vars, return_type);
 
        }
 
        for argument in &arguments {
 
            Self::mark_used_polymorphic_variables(&mut poly_vars, &argument.parser_type);
 
        }
 

	
 
        let is_polymorph = poly_vars.iter().any(|arg| arg.is_in_use);
 

	
 
        self.lookup.insert(definition_id, DefinedType{
 
            ast_root: root_id,
 
            ast_definition: definition_id,
 
            definition: DefinedTypeVariant::Function(FunctionType{
 
                return_types: definition.return_types.clone(),
 
                arguments,
 
                monomorphs: Vec::new(),
 
            }),
 
            poly_vars,
 
            is_polymorph
 
        });
 

	
 
        return Ok(());
 
    }
 

	
 
    /// Builds base component type.
 
    fn build_base_component_definition(&mut self, modules: &[Module], ctx: &mut PassCtx, definition_id: DefinitionId) -> Result<(), ParseError> {
 
        debug_assert!(!self.lookup.contains_key(&definition_id), "base component already built");
 

	
 
        let definition = &ctx.heap[definition_id].as_component();
 
        let root_id = definition.defined_in;
 

	
 
        // Check the argument types
 
        let mut arguments = Vec::with_capacity(definition.parameters.len());
 
        for parameter_id in &definition.parameters {
 
            let parameter = &ctx.heap[*parameter_id];
 
            Self::check_member_parser_type(
 
                modules, ctx, root_id, &parameter.parser_type, false
 
            )?;
 

	
 
            arguments.push(FunctionArgument{
 
                identifier: parameter.identifier.clone(),
 
                parser_type: parameter.parser_type.clone(),
 
            });
 
        }
 

	
 
        // Check conflict of identifiers
 
        Self::check_identifier_collision(
 
            modules, root_id, &arguments, |arg| &arg.identifier, "connector argument"
 
        )?;
 
        Self::check_poly_args_collision(modules, ctx, root_id, &definition.poly_vars)?;
 

	
 
        // Construct internal representation of component
 
        let mut poly_vars = Self::create_polymorphic_variables(&definition.poly_vars);
 
        for argument in &arguments {
 
            Self::mark_used_polymorphic_variables(&mut poly_vars, &argument.parser_type);
 
        }
 

	
 
        let is_polymorph = poly_vars.iter().any(|arg| arg.is_in_use);
 

	
 
        self.lookup.insert(definition_id, DefinedType{
 
            ast_root: root_id,
 
            ast_definition: definition_id,
 
            definition: DefinedTypeVariant::Component(ComponentType{
 
                variant: definition.variant,
 
                arguments,
 
                monomorphs: Vec::new()
 
            }),
 
            poly_vars,
 
            is_polymorph
 
        });
 

	
 
        Ok(())
 
    }
 

	
 
    /// Will check if the member type (field of a struct, embedded type in a
 
    /// union variant) is valid.
 
    fn check_member_parser_type(
 
        modules: &[Module], ctx: &PassCtx, base_definition_root_id: RootId,
 
        member_parser_type: &ParserType, allow_special_compiler_types: bool
 
    ) -> Result<(), ParseError> {
 
        use ParserTypeVariant as PTV;
 

	
 
        for element in &member_parser_type.elements {
 
            match element.variant {
 
                // Special cases
 
                PTV::Void | PTV::InputOrOutput | PTV::ArrayLike | PTV::IntegerLike => {
 
                    if !allow_special_compiler_types {
 
                        unreachable!("compiler-only ParserTypeVariant in member type");
 
                    }
 
                },
 
                // Builtin types, always valid
 
                PTV::Message | PTV::Bool |
 
                PTV::UInt8 | PTV::UInt16 | PTV::UInt32 | PTV::UInt64 |
 
                PTV::SInt8 | PTV::SInt16 | PTV::SInt32 | PTV::SInt64 |
 
                PTV::Character | PTV::String |
 
                PTV::Array | PTV::Input | PTV::Output |
 
                // Likewise, polymorphic variables are always valid
 
                PTV::PolymorphicArgument(_, _) => {},
 
                // Types that are not constructable, or types that are not
 
                // allowed (and checked earlier)
 
                PTV::IntegerLiteral | PTV::Inferred => {
 
                    unreachable!("illegal ParserTypeVariant within type definition");
 
                },
 
                // Finally, user-defined types
 
                PTV::Definition(definition_id, _) => {
 
                    let definition = &ctx.heap[definition_id];
 
                    if !(definition.is_struct() || definition.is_enum() || definition.is_union()) {
 
                        let source = &modules[base_definition_root_id.index as usize].source;
 
                        return Err(ParseError::new_error_str_at_span(
 
                            source, element.element_span, "expected a datatype (a struct, enum or union)"
 
                        ));
 
                    }
 

	
 
                    // Otherwise, we're fine
 
                }
 
            }
 
        }
 

	
 
        // If here, then all elements check out
 
        return Ok(());
 
    }
 

	
 
    /// Go through a list of identifiers and ensure that all identifiers have
 
    /// unique names
 
    fn check_identifier_collision<T: Sized, F: Fn(&T) -> &Identifier>(
 
        modules: &[Module], root_id: RootId, items: &[T], getter: F, item_name: &'static str
 
    ) -> Result<(), ParseError> {
 
        for (item_idx, item) in items.iter().enumerate() {
 
            let item_ident = getter(item);
 
            for other_item in &items[0..item_idx] {
 
                let other_item_ident = getter(other_item);
 
                if item_ident == other_item_ident {
 
                    let module_source = &modules[root_id.index as usize].source;
 
                    return Err(ParseError::new_error_at_span(
 
                        module_source, item_ident.span, format!("This {} is defined more than once", item_name)
 
                    ).with_info_at_span(
 
                        module_source, other_item_ident.span, format!("The other {} is defined here", item_name)
 
                    ));
 
                }
 
            }
 
        }
 

	
 
        Ok(())
 
    }
 

	
 
    /// Go through a list of polymorphic arguments and make sure that the
 
    /// arguments all have unique names, and the arguments do not conflict with
 
    /// any symbols defined at the module scope.
 
    fn check_poly_args_collision(
 
        modules: &[Module], ctx: &PassCtx, root_id: RootId, poly_args: &[Identifier]
 
    ) -> Result<(), ParseError> {
 
        // Make sure polymorphic arguments are unique and none of the
 
        // identifiers conflict with any imported scopes
 
        for (arg_idx, poly_arg) in poly_args.iter().enumerate() {
 
            for other_poly_arg in &poly_args[..arg_idx] {
 
                if poly_arg == other_poly_arg {
 
                    let module_source = &modules[root_id.index as usize].source;
 
                    return Err(ParseError::new_error_str_at_span(
 
                        module_source, poly_arg.span,
 
                        "This polymorphic argument is defined more than once"
 
                    ).with_info_str_at_span(
 
                        module_source, other_poly_arg.span,
 
                        "It conflicts with this polymorphic argument"
 
                    ));
 
                }
 
            }
 

	
 
            // Check if identifier conflicts with a symbol defined or imported
 
            // in the current module
 
            if let Some(symbol) = ctx.symbols.get_symbol_by_name(SymbolScope::Module(root_id), poly_arg.value.as_bytes()) {
 
                // We have a conflict
 
                let module_source = &modules[root_id.index as usize].source;
 
                let introduction_span = symbol.variant.span_of_introduction(ctx.heap);
 
                return Err(ParseError::new_error_str_at_span(
 
                    module_source, poly_arg.span,
 
                    "This polymorphic argument conflicts with another symbol"
 
                ).with_info_str_at_span(
 
                    module_source, introduction_span,
 
                    "It conflicts due to this symbol"
 
                ));
 
            }
 
        }
 

	
 
        // All arguments are fine
 
        Ok(())
 
    }
 

	
 
    //--------------------------------------------------------------------------
 
    // Detecting type loops
 
    //--------------------------------------------------------------------------
 

	
 
    /// Internal function that will detect type loops and check if they're
 
    /// resolvable. If so then the appropriate union variants will be marked as
 
    /// "living on heap". If not then a `ParseError` will be returned
 
    fn detect_and_resolve_type_loops_for(&mut self, modules: &[Module], heap: &Heap, concrete_type: ConcreteType) -> Result<(), ParseError> {
 
        use DefinedTypeVariant as DTV;
 

	
 
        debug_assert!(self.type_loop_breadcrumbs.is_empty());
 
        debug_assert!(self.type_loops.is_empty());
 
        debug_assert!(self.encountered_types.is_empty());
 

	
 
        // Push the initial breadcrumb
 
        let initial_breadcrumb = self.check_member_for_type_loops(&concrete_type);
 
        if let TypeLoopResult::PushBreadcrumb(definition_id, concrete_type) = initial_breadcrumb {
 
            self.handle_new_breadcrumb_for_type_loops(definition_id, concrete_type);
 
        } else {
 
            unreachable!();
 
        }
 

	
 
        // Enter into the main resolving loop
 
        while !self.type_loop_breadcrumbs.is_empty() {
 
            // Because we might be modifying the breadcrumb array we need to
 
            let breadcrumb_idx = self.type_loop_breadcrumbs.len() - 1;
 
            let mut breadcrumb = self.type_loop_breadcrumbs[breadcrumb_idx].clone();
 

	
 
            let poly_type = self.lookup.get(&breadcrumb.definition_id).unwrap();
 

	
 
            let resolve_result = match &poly_type.definition {
 
                DTV::Enum(_) => {
 
                    TypeLoopResult::TypeExists
 
                },
 
                DTV::Union(definition) => {
 
                    let monomorph = &definition.monomorphs[breadcrumb.monomorph_idx];
 
                    let num_variants = monomorph.variants.len();
 

	
 
                    let mut union_result = TypeLoopResult::TypeExists;
 

	
 
                    'member_loop: while breadcrumb.next_member < num_variants {
 
                        let mono_variant = &monomorph.variants[breadcrumb.next_member];
 
                        let num_embedded = mono_variant.embedded.len();
 

	
 
                        while breadcrumb.next_embedded < num_embedded {
 
                            let mono_embedded = &mono_variant.embedded[breadcrumb.next_embedded];
 
                            union_result = self.check_member_for_type_loops(&mono_embedded.concrete_type);
 

	
 
                            if union_result != TypeLoopResult::TypeExists {
 
                                // In type loop or new breadcrumb pushed, so
 
                                // break out of the resolving loop
 
                                break 'member_loop;
 
                            }
 

	
 
                            breadcrumb.next_embedded += 1;
 
                        }
 

	
 
                        breadcrumb.next_embedded = 0;
 
                        breadcrumb.next_member += 1
 
                    }
 

	
 
                    union_result
 
                },
 
                DTV::Struct(definition) => {
 
                    let monomorph = &definition.monomorphs[breadcrumb.monomorph_idx];
 
                    let num_fields = monomorph.fields.len();
 

	
 
                    let mut struct_result = TypeLoopResult::TypeExists;
 
                    while breadcrumb.next_member < num_fields {
 
                        let mono_field = &monomorph.fields[breadcrumb.next_member];
 
                        struct_result = self.check_member_for_type_loops(&mono_field.concrete_type);
 

	
 
                        if struct_result != TypeLoopResult::TypeExists {
 
                            // Type loop or breadcrumb pushed, so break out of
 
                            // the resolving loop
 
                            break;
 
                        }
 

	
 
                        breadcrumb.next_member += 1;
 
                    }
 

	
 
                    struct_result
 
                },
 
                DTV::Function(_) | DTV::Component(_) => unreachable!(),
 
            };
 

	
 
            // Handle the result of attempting to resolve the current breadcrumb
 
            match resolve_result {
 
                TypeLoopResult::TypeExists => {
 
                    // We finished parsing the type
 
                    self.type_loop_breadcrumbs.pop();
 
                },
 
                TypeLoopResult::PushBreadcrumb(definition_id, concrete_type) => {
 
                    // We recurse into the member type.
 
                    self.type_loop_breadcrumbs[breadcrumb_idx] = breadcrumb;
 
                    self.handle_new_breadcrumb_for_type_loops(definition_id, concrete_type);
 
                },
 
                TypeLoopResult::TypeLoop(first_idx) => {
 
                    // Because we will be modifying breadcrumbs within the
 
                    // type-loop handling code, put back the modified breadcrumb
 
                    self.type_loop_breadcrumbs[breadcrumb_idx] = breadcrumb;
 

	
 
                    // We're in a type loop. Add the type loop
 
                    let mut loop_members = Vec::with_capacity(self.type_loop_breadcrumbs.len() - first_idx);
 
                    let mut contains_union = false;
 

	
 
                    for breadcrumb_idx in first_idx..self.type_loop_breadcrumbs.len() {
 
                        let breadcrumb = &mut self.type_loop_breadcrumbs[breadcrumb_idx];
 
                        let mut is_union = false;
 

	
 
                        let entry = self.lookup.get_mut(&breadcrumb.definition_id).unwrap();
 
                        match &mut entry.definition {
 
                            DTV::Union(definition) => {
 
                                // Mark the currently processed variant as requiring heap
 
                                // allocation, then advance the *embedded* type. The loop above
 
                                // will then take care of advancing it to the next *member*.
 
                                let monomorph = &mut definition.monomorphs[breadcrumb.monomorph_idx];
 
                                let variant = &mut monomorph.variants[breadcrumb.next_member];
 
                                variant.lives_on_heap = true;
 
                                breadcrumb.next_embedded += 1;
 
                                is_union = true;
 
                                contains_union = true;
 
                            },
 
                            _ => {}, // else: we don't care for now
 
                        }
 

	
 
                        loop_members.push(TypeLoopEntry{
 
                            definition_id: breadcrumb.definition_id,
 
                            monomorph_idx: breadcrumb.monomorph_idx,
 
                            is_union
 
                        });
 
                    }
 

	
 
                    let new_type_loop = TypeLoop{ members: loop_members };
 
                    if !contains_union {
 
                        // No way to (potentially) break the union. So return a
 
                        // type loop error. This is because otherwise our
 
                        // breadcrumb resolver ends up in an infinite loop.
 
                        return Err(construct_type_loop_error(
 
                            self, &new_type_loop, modules, heap
 
                        ));
 
                    }
 

	
 
                    self.type_loops.push(new_type_loop);
 
                }
 
            }
 
        }
 

	
 
        // All breadcrumbs have been cleared. So now `type_loops` contains all
 
        // of the encountered type loops, and `encountered_types` contains a
 
        // list of all unique monomorphs we encountered.
 

	
 
        // The next step is to figure out if all of the type loops can be
 
        // broken. A type loop can be broken if at least one union exists in the
 
        // loop and that union ended up having variants that are not part of
 
        // a type loop.
 
        fn type_loop_source_span_and_message<'a>(
 
            modules: &'a [Module], heap: &Heap, defined_type: &DefinedType, monomorph_idx: usize, index_in_loop: usize
 
        ) -> (&'a InputSource, InputSpan, String) {
 
            // Note: because we will discover the type loop the *first* time we
 
            // instantiate a monomorph with the provided polymorphic arguments
 
            // (not all arguments are actually used in the type). We don't have
 
            // to care about a second instantiation where certain unused
 
            // polymorphic arguments are different.
 
            let monomorph_type = match &defined_type.definition {
 
                DTV::Union(definition) => &definition.monomorphs[monomorph_idx].concrete_type,
 
                DTV::Struct(definition) => &definition.monomorphs[monomorph_idx].concrete_type,
 
                DTV::Enum(_) | DTV::Function(_) | DTV::Component(_) =>
 
                    unreachable!(), // impossible to have an enum/procedure in a type loop
 
            };
src/protocol/parser/visitor.rs
Show inline comments
 
use crate::protocol::ast::*;
 
use crate::protocol::input_source::ParseError;
 
use crate::protocol::parser::{type_table::*, Module};
 
use crate::protocol::symbol_table::{SymbolTable};
 

	
 
type Unit = ();
 
pub(crate) type VisitorResult = Result<Unit, ParseError>;
 

	
 
/// Globally configured vector capacity for statement buffers in visitor 
 
/// implementations
 
pub(crate) const STMT_BUFFER_INIT_CAPACITY: usize = 256;
 
/// Globally configured vector capacity for expression buffers in visitor
 
/// implementations
 
pub(crate) const EXPR_BUFFER_INIT_CAPACITY: usize = 256;
 

	
 
/// General context structure that is used while traversing the AST.
 
/// TODO: Revise, visitor abstraction is starting to get in the way of programming
 
pub(crate) struct Ctx<'p> {
 
    pub heap: &'p mut Heap,
 
    pub modules: &'p mut [Module],
 
    pub module_idx: usize, // currently considered module
 
    pub symbols: &'p mut SymbolTable,
 
    pub types: &'p mut TypeTable,
 
    pub arch: &'p crate::protocol::TargetArch,
 
}
 

	
 
impl<'p> Ctx<'p> {
 
    /// Returns module `modules[module_idx]`
 
    pub(crate) fn module(&self) -> &Module {
 
        &self.modules[self.module_idx]
 
    }
 

	
 
    pub(crate) fn module_mut(&mut self) -> &mut Module {
 
        &mut self.modules[self.module_idx]
 
    }
 
}
 

	
 
/// Visitor is a generic trait that will fully walk the AST. The default
 
/// implementation of the visitors is to not recurse. The exception is the
 
/// top-level `visit_definition`, `visit_stmt` and `visit_expr` methods, which
 
/// call the appropriate visitor function.
 
pub(crate) trait Visitor {
 
    // Entry point
 
    fn visit_module(&mut self, ctx: &mut Ctx) -> VisitorResult {
 
        let mut def_index = 0;
 
        let module_root_id = ctx.modules[ctx.module_idx].root_id;
 
        loop {
 
            let definition_id = {
 
                let root = &ctx.heap[module_root_id];
 
                if def_index >= root.definitions.len() {
 
                    return Ok(())
 
                }
 

	
 
                root.definitions[def_index]
 
            };
 

	
 
            self.visit_definition(ctx, definition_id)?;
 
            def_index += 1;
 
        }
 
    }
 

	
 
    // Definitions
 
    // --- enum matching
 
    fn visit_definition(&mut self, ctx: &mut Ctx, id: DefinitionId) -> VisitorResult {
 
        match &ctx.heap[id] {
 
            Definition::Enum(def) => {
 
                let def = def.this;
 
                self.visit_enum_definition(ctx, def)
 
            },
 
            Definition::Union(def) => {
 
                let def = def.this;
 
                self.visit_union_definition(ctx, def)
 
            }
 
            Definition::Struct(def) => {
 
                let def = def.this;
 
                self.visit_struct_definition(ctx, def)
 
            },
 
            Definition::Component(def) => {
 
                let def = def.this;
 
                self.visit_component_definition(ctx, def)
 
            },
 
            Definition::Function(def) => {
 
                let def = def.this;
 
                self.visit_function_definition(ctx, def)
 
            }
 
        }
 
    }
 

	
 
    // --- enum variant handling
 
    fn visit_enum_definition(&mut self, _ctx: &mut Ctx, _id: EnumDefinitionId) -> VisitorResult { Ok(()) }
 
    fn visit_union_definition(&mut self, _ctx: &mut Ctx, _id: UnionDefinitionId) -> VisitorResult{ Ok(()) }
 
    fn visit_struct_definition(&mut self, _ctx: &mut Ctx, _id: StructDefinitionId) -> VisitorResult { Ok(()) }
 
    fn visit_component_definition(&mut self, _ctx: &mut Ctx, _id: ComponentDefinitionId) -> VisitorResult { Ok(()) }
 
    fn visit_function_definition(&mut self, _ctx: &mut Ctx, _id: FunctionDefinitionId) -> VisitorResult { Ok(()) }
 

	
 
    // Statements
 
    // --- enum matching
 
    fn visit_stmt(&mut self, ctx: &mut Ctx, id: StatementId) -> VisitorResult {
 
        match &ctx.heap[id] {
 
            Statement::Block(stmt) => {
 
                let this = stmt.this;
 
                self.visit_block_stmt(ctx, this)
 
            },
 
            Statement::EndBlock(_stmt) => Ok(()),
 
            Statement::Local(stmt) => {
 
                let this = stmt.this();
 
                self.visit_local_stmt(ctx, this)
 
            },
 
            Statement::Labeled(stmt) => {
 
                let this = stmt.this;
 
                self.visit_labeled_stmt(ctx, this)
 
            },
 
            Statement::If(stmt) => {
 
                let this = stmt.this;
 
                self.visit_if_stmt(ctx, this)
 
            },
 
            Statement::EndIf(_stmt) => Ok(()),
 
            Statement::While(stmt) => {
 
                let this = stmt.this;
 
                self.visit_while_stmt(ctx, this)
 
            },
 
            Statement::EndWhile(_stmt) => Ok(()),
 
            Statement::Break(stmt) => {
 
                let this = stmt.this;
 
                self.visit_break_stmt(ctx, this)
 
            },
 
            Statement::Continue(stmt) => {
 
                let this = stmt.this;
 
                self.visit_continue_stmt(ctx, this)
 
            },
 
            Statement::Synchronous(stmt) => {
 
                let this = stmt.this;
 
                self.visit_synchronous_stmt(ctx, this)
 
            },
 
            Statement::EndSynchronous(_stmt) => Ok(()),
 
            Statement::Fork(stmt) => {
 
                let this = stmt.this;
 
                self.visit_fork_stmt(ctx, this)
 
            },
 
            Statement::EndFork(_stmt) => Ok(()),
 
            Statement::Return(stmt) => {
 
                let this = stmt.this;
 
                self.visit_return_stmt(ctx, this)
 
            },
 
            Statement::Goto(stmt) => {
 
                let this = stmt.this;
 
                self.visit_goto_stmt(ctx, this)
 
            },
 
            Statement::New(stmt) => {
 
                let this = stmt.this;
 
                self.visit_new_stmt(ctx, this)
 
            },
 
            Statement::Expression(stmt) => {
 
                let this = stmt.this;
 
                self.visit_expr_stmt(ctx, this)
 
            }
 
        }
 
    }
 

	
 
    fn visit_local_stmt(&mut self, ctx: &mut Ctx, id: LocalStatementId) -> VisitorResult {
 
        match &ctx.heap[id] {
 
            LocalStatement::Channel(stmt) => {
 
                let this = stmt.this;
 
                self.visit_local_channel_stmt(ctx, this)
 
            },
 
            LocalStatement::Memory(stmt) => {
 
                let this = stmt.this;
 
                self.visit_local_memory_stmt(ctx, this)
 
            },
 
        }
 
    }
 

	
 
    // --- enum variant handling
 
    fn visit_block_stmt(&mut self, _ctx: &mut Ctx, _id: BlockStatementId) -> VisitorResult { Ok(()) }
 
    fn visit_local_memory_stmt(&mut self, _ctx: &mut Ctx, _id: MemoryStatementId) -> VisitorResult { Ok(()) }
 
    fn visit_local_channel_stmt(&mut self, _ctx: &mut Ctx, _id: ChannelStatementId) -> VisitorResult { Ok(()) }
 
    fn visit_labeled_stmt(&mut self, _ctx: &mut Ctx, _id: LabeledStatementId) -> VisitorResult { Ok(()) }
 
    fn visit_if_stmt(&mut self, _ctx: &mut Ctx, _id: IfStatementId) -> VisitorResult { Ok(()) }
 
    fn visit_while_stmt(&mut self, _ctx: &mut Ctx, _id: WhileStatementId) -> VisitorResult { Ok(()) }
 
    fn visit_break_stmt(&mut self, _ctx: &mut Ctx, _id: BreakStatementId) -> VisitorResult { Ok(()) }
 
    fn visit_continue_stmt(&mut self, _ctx: &mut Ctx, _id: ContinueStatementId) -> VisitorResult { Ok(()) }
 
    fn visit_synchronous_stmt(&mut self, _ctx: &mut Ctx, _id: SynchronousStatementId) -> VisitorResult { Ok(()) }
 
    fn visit_fork_stmt(&mut self, _ctx: &mut Ctx, _id: ForkStatementId) -> VisitorResult { Ok(()) }
 
    fn visit_return_stmt(&mut self, _ctx: &mut Ctx, _id: ReturnStatementId) -> VisitorResult { Ok(()) }
 
    fn visit_goto_stmt(&mut self, _ctx: &mut Ctx, _id: GotoStatementId) -> VisitorResult { Ok(()) }
 
    fn visit_new_stmt(&mut self, _ctx: &mut Ctx, _id: NewStatementId) -> VisitorResult { Ok(()) }
 
    fn visit_expr_stmt(&mut self, _ctx: &mut Ctx, _id: ExpressionStatementId) -> VisitorResult { Ok(()) }
 

	
 
    // Expressions
 
    // --- enum matching
 
    fn visit_expr(&mut self, ctx: &mut Ctx, id: ExpressionId) -> VisitorResult {
 
        match &ctx.heap[id] {
 
            Expression::Assignment(expr) => {
 
                let this = expr.this;
 
                self.visit_assignment_expr(ctx, this)
 
            },
 
            Expression::Binding(expr) => {
 
                let this = expr.this;
 
                self.visit_binding_expr(ctx, this)
 
            }
 
            Expression::Conditional(expr) => {
 
                let this = expr.this;
 
                self.visit_conditional_expr(ctx, this)
 
            }
 
            Expression::Binary(expr) => {
 
                let this = expr.this;
 
                self.visit_binary_expr(ctx, this)
 
            }
 
            Expression::Unary(expr) => {
 
                let this = expr.this;
 
                self.visit_unary_expr(ctx, this)
 
            }
 
            Expression::Indexing(expr) => {
 
                let this = expr.this;
 
                self.visit_indexing_expr(ctx, this)
 
            }
 
            Expression::Slicing(expr) => {
 
                let this = expr.this;
 
                self.visit_slicing_expr(ctx, this)
 
            }
 
            Expression::Select(expr) => {
 
                let this = expr.this;
 
                self.visit_select_expr(ctx, this)
 
            }
 
            Expression::Literal(expr) => {
 
                let this = expr.this;
 
                self.visit_literal_expr(ctx, this)
 
            }
 
            Expression::Cast(expr) => {
 
                let this = expr.this;
 
                self.visit_cast_expr(ctx, this)
 
            }
 
            Expression::Call(expr) => {
 
                let this = expr.this;
 
                self.visit_call_expr(ctx, this)
 
            }
 
            Expression::Variable(expr) => {
 
                let this = expr.this;
 
                self.visit_variable_expr(ctx, this)
 
            }
 
        }
 
    }
 

	
 
    fn visit_assignment_expr(&mut self, _ctx: &mut Ctx, _id: AssignmentExpressionId) -> VisitorResult { Ok(()) }
 
    fn visit_binding_expr(&mut self, _ctx: &mut Ctx, _id: BindingExpressionId) -> VisitorResult { Ok(()) }
 
    fn visit_conditional_expr(&mut self, _ctx: &mut Ctx, _id: ConditionalExpressionId) -> VisitorResult { Ok(()) }
 
    fn visit_binary_expr(&mut self, _ctx: &mut Ctx, _id: BinaryExpressionId) -> VisitorResult { Ok(()) }
 
    fn visit_unary_expr(&mut self, _ctx: &mut Ctx, _id: UnaryExpressionId) -> VisitorResult { Ok(()) }
 
    fn visit_indexing_expr(&mut self, _ctx: &mut Ctx, _id: IndexingExpressionId) -> VisitorResult { Ok(()) }
 
    fn visit_slicing_expr(&mut self, _ctx: &mut Ctx, _id: SlicingExpressionId) -> VisitorResult { Ok(()) }
 
    fn visit_select_expr(&mut self, _ctx: &mut Ctx, _id: SelectExpressionId) -> VisitorResult { Ok(()) }
 
    fn visit_literal_expr(&mut self, _ctx: &mut Ctx, _id: LiteralExpressionId) -> VisitorResult { Ok(()) }
 
    fn visit_cast_expr(&mut self, _ctx: &mut Ctx, _id: CastExpressionId) -> VisitorResult { Ok(()) }
 
    fn visit_call_expr(&mut self, _ctx: &mut Ctx, _id: CallExpressionId) -> VisitorResult { Ok(()) }
 
    fn visit_variable_expr(&mut self, _ctx: &mut Ctx, _id: VariableExpressionId) -> VisitorResult { Ok(()) }
 
}
 
\ No newline at end of file
src/protocol/tests/parser_monomorphs.rs
Show inline comments
 
/// parser_monomorphs.rs
 
///
 
/// Simple tests to make sure that all of the appropriate monomorphs are 
 
/// instantiated
 

	
 
use super::*;
 

	
 
#[test]
 
fn test_struct_monomorphs() {
 
    Tester::new_single_source_expect_ok(
 
        "no polymorph",
 
        "struct Integer{ s32 field }"
 
    ).for_struct("Integer", |s| { s
 
        .assert_num_monomorphs(1)
 
        .assert_has_monomorph("Integer");
 
    });
 

	
 
    Tester::new_single_source_expect_ok(
 
        "single polymorph",
 
        "
 
        struct Number<T>{ T number }
 
        func instantiator() -> s32 {
 
            auto a = Number<s8>{ number: 0 };
 
            auto b = Number<s8>{ number: 1 };
 
            auto c = Number<s32>{ number: 2 };
 
            auto d = Number<s64>{ number: 3 };
 
            auto e = Number<Number<s16>>{ number: Number{ number: 4 }};
 
            return 0;
 
        }
 
        "
 
    ).for_struct("Number", |s| { s
 
        .assert_has_monomorph("Number<s8>")
 
        .assert_has_monomorph("Number<s16>")
 
        .assert_has_monomorph("Number<s32>")
 
        .assert_has_monomorph("Number<s64>")
 
        .assert_has_monomorph("Number<Number<s16>>")
 
        .assert_num_monomorphs(5);
 
    }).for_function("instantiator", |f| { f
 
        .for_variable("a", |v| {v.assert_concrete_type("Number<s8>");} )
 
        .for_variable("e", |v| {v.assert_concrete_type("Number<Number<s16>>");} );
 
    });
 
}
 

	
 
#[test]
 
fn test_enum_monomorphs() {
 
    Tester::new_single_source_expect_ok(
 
        "no polymorph",
 
        "
 
        enum Answer{ Yes, No }
 
        func do_it() -> s32 { auto a = Answer::Yes; return 0; }
 
        "
 
    ).for_enum("Answer", |e| { e
 
        .assert_num_monomorphs(1)
 
        .assert_has_monomorph("Answer");
 
    });
 

	
 
    // Note for reader: because the enum doesn't actually use the polymorphic
 
    // variable, we expect to have 1 monomorph: the type only has to be laid
 
    // out once.
 
    Tester::new_single_source_expect_ok(
 
        "single polymorph",
 
        "
 
        enum Answer<T> { Yes, No }
 
        func instantiator() -> s32 {
 
            auto a = Answer<s8>::Yes;
 
            auto b = Answer<s8>::No;
 
            auto c = Answer<s32>::Yes;
 
            auto d = Answer<Answer<Answer<s64>>>::No;
 
            return 0;
 
        }
 
        "
 
    ).for_enum("Answer", |e| { e
 
        .assert_num_monomorphs(1)
 
        .assert_has_monomorph("Answer<s8>");
 
    });
 
}
 

	
 
#[test]
 
fn test_union_monomorphs() {
 
    Tester::new_single_source_expect_ok(
 
        "no polymorph",
 
        "
 
        union Trinary { Undefined, Value(bool) }
 
        func do_it() -> s32 { auto a = Trinary::Value(true); return 0; }
 
        "
 
    ).for_union("Trinary", |e| { e
 
        .assert_num_monomorphs(1)
 
        .assert_has_monomorph("Trinary");
 
    });
 

	
 
    // TODO: Does this do what we want? Or do we expect the embedded monomorph
 
    //  Result<s8,s32> to be instantiated as well? I don't think so.
 
    Tester::new_single_source_expect_ok(
 
        "polymorphs",
 
        "
 
        union Result<T, E>{ Ok(T), Err(E) }
 
        func instantiator() -> s32 {
 
            s16 a_s16 = 5;
 
            auto a = Result<s8, bool>::Ok(0);
 
            auto b = Result<bool, s8>::Ok(true);
 
            auto c = Result<Result<s8, s32>, Result<s16, s64>>::Err(Result::Ok(5));
 
            auto d = Result<Result<s8, s32>, auto>::Err(Result<auto, s64>::Ok(a_s16));
 
            return 0;
 
        }
 
        "
 
    ).for_union("Result", |e| { e
 
        .assert_num_monomorphs(5)
 
        .assert_has_monomorph("Result<s8,bool>")
 
        .assert_has_monomorph("Result<bool,s8>")
 
        .assert_has_monomorph("Result<Result<s8,s32>,Result<s16,s64>>")
 
        .assert_has_monomorph("Result<s8,s32>")
 
        .assert_has_monomorph("Result<s16,s64>");
 
    }).for_function("instantiator", |f| { f
 
        .for_variable("d", |v| { v
 
            .assert_parser_type("auto")
 
            .assert_concrete_type("Result<Result<s8,s32>,Result<s16,s64>>");
 
        });
 
    });
 
}
 
\ No newline at end of file
src/runtime2/branch.rs
Show inline comments
 
use std::collections::HashMap;
 
use std::ops::{Index, IndexMut};
 

	
 
use crate::protocol::eval::{Prompt, Value, ValueGroup};
 

	
 
use super::port::PortIdLocal;
 

	
 
// To share some logic between the FakeTree and ExecTree implementation
 
trait BranchListItem {
 
    fn get_id(&self) -> BranchId;
 
    fn set_next_id(&mut self, id: BranchId);
 
    fn get_next_id(&self) -> BranchId;
 
}
 

	
 
/// Generic branch ID. A component will always have one branch: the
 
/// non-speculative branch. This branch has ID 0. Hence in a speculative context
 
/// we use this fact to let branch ID 0 denote the ID being invalid.
 
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
 
pub struct BranchId {
 
    pub index: u32
 
}
 

	
 
impl BranchId {
 
    #[inline]
 
    pub(crate) fn new_invalid() -> Self {
 
        return Self{ index: 0 };
 
    }
 

	
 
    #[inline]
 
    fn new(index: u32) -> Self {
 
        debug_assert!(index != 0);
 
        return Self{ index };
 
    }
 

	
 
    #[inline]
 
    pub(crate) fn is_valid(&self) -> bool {
 
        return self.index != 0;
 
    }
 
}
 

	
 
#[derive(Debug, PartialEq, Eq)]
 
pub(crate) enum SpeculativeState {
 
    // Non-synchronous variants
 
    RunningNonSync,         // regular execution of code
 
    Error,                  // encountered a runtime error
 
    Finished,               // finished executing connector's code
 
    // Synchronous variants
 
    RunningInSync,          // running within a sync block
 
    HaltedAtBranchPoint,    // at a branching point (at a `get` call)
 
    ReachedSyncEnd,         // reached end of sync block, branch represents a local solution
 
    Inconsistent,           // branch can never represent a local solution, so halted
 
}
 

	
 
#[derive(Debug)]
 
pub(crate) enum PreparedStatement {
 
    CreatedChannel((Value, Value)),
 
    ForkedExecution(bool),
 
    PerformedPut,
 
    PerformedGet(ValueGroup),
 
    None,
 
}
 

	
 
impl PreparedStatement {
 
    pub(crate) fn is_none(&self) -> bool {
 
        if let PreparedStatement::None = self {
 
            return true;
 
        } else {
 
            return false;
 
        }
 
    }
 

	
 
    pub(crate) fn take(&mut self) -> PreparedStatement {
 
        if let PreparedStatement::None = self {
 
            return PreparedStatement::None;
 
        } else {
 
            let mut replacement = PreparedStatement::None;
 
            std::mem::swap(self, &mut replacement);
 
            return replacement;
 
        }
 
    }
 
}
 

	
 
/// The execution state of a branch. This envelops the PDL code and the
 
/// execution state. And derived from that: if we're ready to keep running the
 
/// code, or if we're halted for some reason (e.g. waiting for a message).
 
pub(crate) struct Branch {
 
    pub id: BranchId,
 
    pub parent_id: BranchId,
 
    // Execution state
 
    pub code_state: Prompt,
 
    pub sync_state: SpeculativeState,
 
    pub awaiting_port: PortIdLocal, // only valid if in "awaiting message" queue. TODO: Maybe put in enum
 
    pub awaiting_port: PortIdLocal, // only valid if in "awaiting message" queue.
 
    pub next_in_queue: BranchId, // used by `ExecTree`/`BranchQueue`
 
    pub prepared: PreparedStatement,
 
}
 

	
 
impl BranchListItem for Branch {
 
    #[inline] fn get_id(&self) -> BranchId { return self.id; }
 
    #[inline] fn set_next_id(&mut self, id: BranchId) { self.next_in_queue = id; }
 
    #[inline] fn get_next_id(&self) -> BranchId { return self.next_in_queue; }
 
}
 

	
 
impl Branch {
 
    /// Creates a new non-speculative branch
 
    pub(crate) fn new_non_sync(component_state: Prompt) -> Self {
 
        Branch {
 
            id: BranchId::new_invalid(),
 
            parent_id: BranchId::new_invalid(),
 
            code_state: component_state,
 
            sync_state: SpeculativeState::RunningNonSync,
 
            awaiting_port: PortIdLocal::new_invalid(),
 
            next_in_queue: BranchId::new_invalid(),
 
            prepared: PreparedStatement::None,
 
        }
 
    }
 

	
 
    /// Constructs a sync branch. The provided branch is assumed to be the
 
    /// parent of the new branch within the execution tree.
 
    fn new_sync(new_index: u32, parent_branch: &Branch) -> Self {
 
        // debug_assert!(
 
        //     (parent_branch.sync_state == SpeculativeState::RunningNonSync && !parent_branch.parent_id.is_valid()) ||
 
        //     (parent_branch.sync_state == SpeculativeState::HaltedAtBranchPoint)
 
        // ); // forking from non-sync, or forking from a branching point
 
        debug_assert!(parent_branch.prepared.is_none());
 

	
 
        Branch {
 
            id: BranchId::new(new_index),
 
            parent_id: parent_branch.id,
 
            code_state: parent_branch.code_state.clone(),
 
            sync_state: SpeculativeState::RunningInSync,
 
            awaiting_port: parent_branch.awaiting_port,
 
            next_in_queue: BranchId::new_invalid(),
 
            prepared: PreparedStatement::None,
 
        }
 
    }
 
}
 

	
 
/// Queue of branches. Just a little helper.
 
#[derive(Copy, Clone)]
 
struct BranchQueue {
 
    first: BranchId,
 
    last: BranchId,
 
}
 

	
 
impl BranchQueue {
 
    #[inline]
 
    fn new() -> Self {
 
        Self{
 
            first: BranchId::new_invalid(),
 
            last: BranchId::new_invalid()
 
        }
 
    }
 

	
 
    #[inline]
 
    fn is_empty(&self) -> bool {
 
        debug_assert!(self.first.is_valid() == self.last.is_valid());
 
        return !self.first.is_valid();
 
    }
 
}
 

	
 
const NUM_QUEUES: usize = 3;
 

	
 
#[derive(Debug, PartialEq, Eq, Clone, Copy)]
 
pub(crate) enum QueueKind {
 
    Runnable,
 
    AwaitingMessage,
 
    FinishedSync,
 
}
 

	
 
impl QueueKind {
 
    fn as_index(&self) -> usize {
 
        return match self {
 
            QueueKind::Runnable => 0,
 
            QueueKind::AwaitingMessage => 1,
 
            QueueKind::FinishedSync => 2,
 
        }
 
    }
 
}
 

	
 
// -----------------------------------------------------------------------------
 
// ExecTree
 
// -----------------------------------------------------------------------------
 

	
 
/// Execution tree of branches. Tries to keep the extra information stored
 
/// herein to a minimum. So the execution tree is aware of the branches, their
 
/// execution state and the way they're dependent on each other, but the
 
/// execution tree should not be aware of e.g. sync algorithms.
 
///
 
/// Note that the tree keeps track of multiple lists of branches. Each list
 
/// contains branches that ended up in a particular execution state. The lists
 
/// are described by the various `BranchQueue` instances and the `next_in_queue`
 
/// field in each branch.
 
pub(crate) struct ExecTree {
 
    // All branches. the `parent_id` field in each branch implies the shape of
 
    // the tree. Branches are index stable throughout a sync round.
 
    pub branches: Vec<Branch>,
 
    queues: [BranchQueue; NUM_QUEUES]
 
}
 

	
 
impl ExecTree {
 
    /// Constructs a new execution tree with a single non-sync branch.
 
    pub fn new(component: Prompt) -> Self {
 
        return Self {
 
            branches: vec![Branch::new_non_sync(component)],
 
            queues: [BranchQueue::new(); 3]
 
        }
 
    }
 

	
 
    // --- Generic branch (queue) management
 

	
 
    /// Returns if tree is in speculative mode
 
    pub fn is_in_sync(&self) -> bool {
 
        return self.branches.len() != 1;
 
    }
 

	
 
    /// Returns true if the particular queue is empty
 
    pub fn queue_is_empty(&self, kind: QueueKind) -> bool {
 
        return self.queues[kind.as_index()].is_empty();
 
    }
 

	
 
    /// Pops a branch (ID) from a queue.
 
    pub fn pop_from_queue(&mut self, kind: QueueKind) -> Option<BranchId> {
 
        debug_assert_ne!(kind, QueueKind::FinishedSync); // for purposes of logic we expect the queue to grow during a sync round
 
        return pop_from_queue(&mut self.queues[kind.as_index()], &mut self.branches);
 
    }
 

	
 
    /// Pushes a branch (ID) into a queue.
 
    pub fn push_into_queue(&mut self, kind: QueueKind, id: BranchId) {
 
        push_into_queue(&mut self.queues[kind.as_index()], &mut self.branches, id);
 
    }
 

	
 
    /// Returns the non-sync branch (TODO: better name?)
 
    pub fn base_branch_mut(&mut self) -> &mut Branch {
 
        debug_assert!(!self.is_in_sync());
 
        return &mut self.branches[0];
 
    }
 

	
 
    /// Returns the branch ID of the first branch in a particular queue.
 
    pub fn get_queue_first(&self, kind: QueueKind) -> Option<BranchId> {
 
        let queue = &self.queues[kind.as_index()];
 
        if queue.first.is_valid() {
 
            return Some(queue.first);
 
        } else {
 
            return None;
 
        }
 
    }
 

	
 
    /// Returns the next branch ID of a branch (assumed to be in a particular
 
    /// queue.
 
    pub fn get_queue_next(&self, branch_id: BranchId) -> Option<BranchId> {
 
        let branch = &self.branches[branch_id.index as usize];
 
        if branch.next_in_queue.is_valid() {
 
            return Some(branch.next_in_queue);
 
        } else {
 
            return None;
 
        }
 
    }
 

	
 
    /// Returns an iterator that starts with the provided branch, and then
 
    /// continues to visit all of the branch's parents.
 
    pub fn iter_parents(&self, branch_id: BranchId) -> BranchParentIter {
 
        return BranchParentIter{
 
            branches: self.branches.as_slice(),
 
            index: branch_id.index as usize,
 
        }
 
    }
 

	
 
    // --- Preparing and finishing a speculative round
 

	
 
    /// Starts a synchronous round by cloning the non-sync branch and marking it
 
    /// as the root of the speculative tree. The id of this root sync branch is
 
    /// returned.
 
    pub fn start_sync(&mut self) -> BranchId {
 
        debug_assert!(!self.is_in_sync());
 
        let sync_branch = Branch::new_sync(1, &self.branches[0]);
 
        let sync_branch_id = sync_branch.id;
 
        self.branches.push(sync_branch);
 

	
 
        return sync_branch_id;
 
    }
 

	
 
    /// Creates a new speculative branch based on the provided one. The index to
 
    /// retrieve this new branch will be returned.
 
    pub fn fork_branch(&mut self, parent_branch_id: BranchId) -> BranchId {
 
        debug_assert!(self.is_in_sync());
 
        let parent_branch = &self[parent_branch_id];
 
        let new_branch = Branch::new_sync(self.branches.len() as u32, parent_branch);
 
        let new_branch_id = new_branch.id;
 
        self.branches.push(new_branch);
 

	
 
        return new_branch_id;
 
    }
 

	
 
    /// Collapses the speculative execution tree back into a deterministic one,
 
    /// using the provided branch as the final sync result.
 
    pub fn end_sync(&mut self, branch_id: BranchId) {
 
        debug_assert!(self.is_in_sync());
 

	
 
        // Swap indicated branch into the first position
 
        self.branches.swap(0, branch_id.index as usize);
 
        self.branches.truncate(1);
 

	
 
        // Reset all values to non-sync defaults
 
        let branch = &mut self.branches[0];
 
        branch.id = BranchId::new_invalid();
 
        branch.parent_id = BranchId::new_invalid();
 
        branch.sync_state = SpeculativeState::RunningNonSync;
 
        debug_assert!(!branch.awaiting_port.is_valid());
 
        branch.next_in_queue = BranchId::new_invalid();
 
        debug_assert!(branch.prepared.is_none());
 

	
 
        // Clear out all the queues
 
        for queue_idx in 0..NUM_QUEUES {
 
            self.queues[queue_idx] = BranchQueue::new();
 
        }
 
    }
 
}
 

	
 
impl Index<BranchId> for ExecTree {
 
    type Output = Branch;
 

	
 
    fn index(&self, index: BranchId) -> &Self::Output {
 
        debug_assert!(index.is_valid());
 
        return &self.branches[index.index as usize];
 
    }
 
}
 

	
 
impl IndexMut<BranchId> for ExecTree {
 
    fn index_mut(&mut self, index: BranchId) -> &mut Self::Output {
 
        debug_assert!(index.is_valid());
 
        return &mut self.branches[index.index as usize];
 
    }
 
}
 

	
 
/// Iterator over the parents of an `ExecTree` branch.
 
pub(crate) struct BranchParentIter<'a> {
 
    branches: &'a [Branch],
 
    index: usize,
 
}
 

	
 
impl<'a> Iterator for BranchParentIter<'a> {
 
    type Item = &'a Branch;
 

	
 
    fn next(&mut self) -> Option<Self::Item> {
 
        if self.index == 0 {
 
            return None;
 
        }
 

	
 
        let branch = &self.branches[self.index];
 
        self.index = branch.parent_id.index as usize;
 
        return Some(branch);
 
    }
 
}
 

	
 
// -----------------------------------------------------------------------------
 
// FakeTree
 
// -----------------------------------------------------------------------------
 

	
 
/// Generic fake branch. This is supposed to be used in conjunction with the
 
/// fake tree. The purpose is to have a branching-like tree to use in
 
/// combination with a consensus algorithm in places where we don't have PDL
 
/// code.
 
pub(crate) struct FakeBranch {
 
    pub id: BranchId,
 
    pub parent_id: BranchId,
 
    pub sync_state: SpeculativeState,
 
    pub awaiting_port: PortIdLocal,
 
    pub next_in_queue: BranchId,
 
    pub inbox: HashMap<PortIdLocal, ValueGroup>,
 
}
 

	
 
impl BranchListItem for FakeBranch {
 
    #[inline] fn get_id(&self) -> BranchId { return self.id; }
 
    #[inline] fn set_next_id(&mut self, id: BranchId) { self.next_in_queue = id; }
 
    #[inline] fn get_next_id(&self) -> BranchId { return self.next_in_queue; }
 
}
 

	
 
impl FakeBranch {
 
    fn new_root(_index: u32) -> FakeBranch {
 
        debug_assert!(_index == 1);
 
        return FakeBranch{
 
            id: BranchId::new(1),
 
            parent_id: BranchId::new_invalid(),
 
            sync_state: SpeculativeState::RunningInSync,
 
            awaiting_port: PortIdLocal::new_invalid(),
 
            next_in_queue: BranchId::new_invalid(),
 
            inbox: HashMap::new(),
 
        }
 
    }
 

	
 
    fn new_branching(index: u32, parent_branch: &FakeBranch) -> FakeBranch {
 
        return FakeBranch {
 
            id: BranchId::new(index),
 
            parent_id: parent_branch.id,
 
            sync_state: SpeculativeState::RunningInSync,
 
            awaiting_port: parent_branch.awaiting_port,
 
            next_in_queue: BranchId::new_invalid(),
 
            inbox: parent_branch.inbox.clone(),
 
        }
 
    }
 

	
 
    pub fn insert_message(&mut self, target_port: PortIdLocal, contents: ValueGroup) {
 
        debug_assert!(target_port.is_valid());
 
        debug_assert!(self.awaiting_port == target_port);
 
        self.awaiting_port = PortIdLocal::new_invalid();
 
        self.inbox.insert(target_port, contents);
 
    }
 
}
 

	
 
/// A little helper for native components that don't have a set of branches that
 
/// are actually executing code, but just have to manage the idea of branches
 
/// due to them performing the equivalent of a branching `get` call.
 
pub(crate) struct FakeTree {
 
    pub branches: Vec<FakeBranch>,
 
    queues: [BranchQueue; NUM_QUEUES],
 
}
 

	
 
impl FakeTree {
 
    pub fn new() -> Self {
 
        // TODO: Don't like this? Cause is that now we don't have a non-sync
 
        //  branch. But we assumed BranchId=0 means the branch is invalid. We
 
        //  can do the rusty Option<BranchId> stuff. But we still need a token
 
        //  value within the protocol to signify no-branch-id. Maybe the high
 
        //  bit? Branches are crazy expensive, no-one is going to have 2^32
 
        //  branches anyway. 2^31 isn't too bad.
 
        return Self {
 
            branches: vec![FakeBranch{
 
                id: BranchId::new_invalid(),
 
                parent_id: BranchId::new_invalid(),
 
                sync_state: SpeculativeState::RunningNonSync,
 
                awaiting_port: PortIdLocal::new_invalid(),
 
                next_in_queue: BranchId::new_invalid(),
 
                inbox: HashMap::new(),
 
            }],
 
            queues: [BranchQueue::new(); 3]
 
        }
 
    }
 

	
 
    fn is_in_sync(&self) -> bool {
 
        return self.branches.len() > 1;
 
    }
 

	
 
    pub fn queue_is_empty(&self, kind: QueueKind) -> bool {
 
        return self.queues[kind.as_index()].is_empty();
 
    }
 

	
 
    pub fn pop_from_queue(&mut self, kind: QueueKind) -> Option<BranchId> {
 
        debug_assert_ne!(kind, QueueKind::FinishedSync);
 
        return pop_from_queue(&mut self.queues[kind.as_index()], &mut self.branches);
 
    }
 

	
 
    pub fn push_into_queue(&mut self, kind: QueueKind, id: BranchId) {
 
        push_into_queue(&mut self.queues[kind.as_index()], &mut self.branches, id);
 
    }
 

	
 
    pub fn get_queue_first(&self, kind: QueueKind) -> Option<BranchId> {
 
        let queue = &self.queues[kind.as_index()];
 
        if queue.first.is_valid() {
 
            return Some(queue.first)
 
        } else {
 
            return None;
 
        }
 
    }
 

	
 
    pub fn get_queue_next(&self, branch_id: BranchId) -> Option<BranchId> {
 
        let branch = &self.branches[branch_id.index as usize];
 
        if branch.next_in_queue.is_valid() {
 
            return Some(branch.next_in_queue);
 
        } else {
 
            return None;
 
        }
 
    }
 

	
 
    pub fn start_sync(&mut self) -> BranchId {
 
        debug_assert!(!self.is_in_sync());
 

	
src/runtime2/connector.rs
Show inline comments
 
// connector.rs
 
//
 
// Represents a component. A component (and the scheduler that is running it)
 
// has many properties that are not easy to subdivide into aspects that are
 
// conceptually handled by particular data structures. That is to say: the code
 
// that we run governs: running PDL code, keeping track of ports, instantiating
 
// new components and transports (i.e. interacting with the runtime), running
 
// a consensus algorithm, etc. But on the other hand, our data is rather
 
// simple: we have a speculative execution tree, a set of ports that we own,
 
// and a bit of code that we should run.
 
//
 
// So currently the code is organized as following:
 
// - The scheduler that is running the component is the authoritative source on
 
//     ports during *non-sync* mode. The consensus algorithm is the
 
//     authoritative source during *sync* mode. They retrieve each other's
 
//     state during the transitions. Hence port data exists duplicated between
 
//     these two datastructures.
 
// - The execution tree is where executed branches reside. But the execution
 
//     tree is only aware of the tree shape itself (and keeps track of some
 
//     queues of branches that are in a particular state), and tends to store
 
//     the PDL program state. The consensus algorithm is also somewhat aware
 
//     of the execution tree, but only in terms of what is needed to complete
 
//     a sync round (for now, that means the port mapping in each branch).
 
//     Hence once more we have properties conceptually associated with branches
 
//     in two places.
 
// - TODO: Write about handling messages, consensus wrapping data
 
// - TODO: Write about way information is exchanged between PDL/component and scheduler through ctx
 

	
 
use std::sync::atomic::AtomicBool;
 

	
 
use crate::{PortId, ProtocolDescription};
 
use crate::protocol::eval::{EvalContinuation, EvalError, Prompt, Value, ValueGroup};
 
use crate::protocol::RunContext;
 

	
 
use super::branch::{BranchId, ExecTree, QueueKind, SpeculativeState, PreparedStatement};
 
use super::consensus::{Consensus, Consistency, RoundConclusion, find_ports_in_value_group};
 
use super::inbox::{DataMessage, Message, SyncCompMessage, SyncPortMessage, SyncControlMessage, PublicInbox};
 
use super::native::Connector;
 
use super::port::{PortKind, PortIdLocal};
 
use super::scheduler::{ComponentCtx, SchedulerCtx, MessageTicket};
 

	
 
pub(crate) struct ConnectorPublic {
 
    pub inbox: PublicInbox,
 
    pub sleeping: AtomicBool,
 
}
 

	
 
impl ConnectorPublic {
 
    pub fn new(initialize_as_sleeping: bool) -> Self {
 
        ConnectorPublic{
 
            inbox: PublicInbox::new(),
 
            sleeping: AtomicBool::new(initialize_as_sleeping),
 
        }
 
    }
 
}
 

	
 
#[derive(Debug, PartialEq, Eq, Clone, Copy)]
 
enum Mode {
 
    NonSync,    // running non-sync code
 
    Sync,       // running sync code (in potentially multiple branches)
 
    SyncError,  // encountered an unrecoverable error in sync mode
 
    Error,      // encountered an error in non-sync mode (or finished handling the sync mode error).
 
}
 

	
 
#[derive(Debug)]
 
pub(crate) enum ConnectorScheduling {
 
    Immediate,          // Run again, immediately
 
    Later,              // Schedule for running, at some later point in time
 
    NotNow,             // Do not reschedule for running
 
    Exit,               // Connector has exited
 
}
 

	
 
pub(crate) struct ConnectorPDL {
 
    mode: Mode,
 
    eval_error: Option<EvalError>,
 
    tree: ExecTree,
 
    consensus: Consensus,
 
    last_finished_handled: Option<BranchId>,
 
}
 

	
 
// TODO: Remove remaining fields once 'fires()' is removed from language.
 
struct ConnectorRunContext<'a> {
 
    branch_id: BranchId,
 
    consensus: &'a Consensus,
 
    prepared: PreparedStatement,
 
}
 

	
 
impl<'a> RunContext for ConnectorRunContext<'a>{
 
    fn performed_put(&mut self, _port: PortId) -> bool {
 
        return match self.prepared.take() {
 
            PreparedStatement::None => false,
 
            PreparedStatement::PerformedPut => true,
 
            taken => unreachable!("prepared statement is '{:?}' during 'performed_put()'", taken)
 
        };
 
    }
 

	
 
    fn performed_get(&mut self, _port: PortId) -> Option<ValueGroup> {
 
        return match self.prepared.take() {
 
            PreparedStatement::None => None,
 
            PreparedStatement::PerformedGet(value) => Some(value),
 
            taken => unreachable!("prepared statement is '{:?}' during 'performed_get()'", taken),
 
        };
 
    }
 

	
 
    fn fires(&mut self, port: PortId) -> Option<Value> {
 
        todo!("Remove fires() now");
 
        let port_id = PortIdLocal::new(port.0.u32_suffix);
 
        let annotation = self.consensus.get_annotation(self.branch_id, port_id);
 
        return annotation.expected_firing.map(|v| Value::Bool(v));
 
    }
 

	
 
    fn created_channel(&mut self) -> Option<(Value, Value)> {
 
        return match self.prepared.take() {
 
            PreparedStatement::None => None,
 
            PreparedStatement::CreatedChannel(ports) => Some(ports),
 
            taken => unreachable!("prepared statement is '{:?}' during 'created_channel()'", taken),
 
        };
 
    }
 

	
 
    fn performed_fork(&mut self) -> Option<bool> {
 
        return match self.prepared.take() {
 
            PreparedStatement::None => None,
 
            PreparedStatement::ForkedExecution(path) => Some(path),
 
            taken => unreachable!("prepared statement is '{:?}' during 'performed_fork()'", taken),
 
        };
 
    }
 
}
 

	
 
impl Connector for ConnectorPDL {
 
    fn run(&mut self, sched_ctx: SchedulerCtx, comp_ctx: &mut ComponentCtx) -> ConnectorScheduling {
 
        if let Some(scheduling) = self.handle_new_messages(comp_ctx) {
 
            return scheduling;
 
        }
 

	
 
        match self.mode {
 
            Mode::Sync => {
 
                // Run in sync mode
 
                let scheduling = self.run_in_sync_mode(sched_ctx, comp_ctx);
 

	
 
                // Handle any new finished branches
 
                let mut iter_id = self.last_finished_handled.or(self.tree.get_queue_first(QueueKind::FinishedSync));
 
                while let Some(branch_id) = iter_id {
 
                    iter_id = self.tree.get_queue_next(branch_id);
 
                    self.last_finished_handled = Some(branch_id);
 

	
 
                    if let Some(round_conclusion) = self.consensus.handle_new_finished_sync_branch(branch_id, comp_ctx) {
 
                        // Actually found a solution
 
                        return self.enter_non_sync_mode(round_conclusion, comp_ctx);
 
                    }
 

	
 
                    self.last_finished_handled = Some(branch_id);
 
                }
 

	
 
                return scheduling;
 
            },
 
            Mode::NonSync => {
 
                let scheduling = self.run_in_deterministic_mode(sched_ctx, comp_ctx);
 
                return scheduling;
 
            },
 
            Mode::SyncError => {
 
                let scheduling = self.run_in_sync_mode(sched_ctx, comp_ctx);
 
                return scheduling;
 
            },
 
            Mode::Error => {
 
                // This shouldn't really be called. Because when we reach exit
 
                // mode the scheduler should not run the component anymore
 
                unreachable!("called component run() during error-mode");
 
            },
 
        }
 
    }
 
}
 

	
 
impl ConnectorPDL {
 
    pub fn new(initial: Prompt) -> Self {
 
        Self{
 
            mode: Mode::NonSync,
 
            eval_error: None,
 
            tree: ExecTree::new(initial),
 
            consensus: Consensus::new(),
 
            last_finished_handled: None,
 
        }
 
    }
 

	
 
    // --- Handling messages
 

	
 
    pub fn handle_new_messages(&mut self, ctx: &mut ComponentCtx) -> Option<ConnectorScheduling> {
 
        while let Some(ticket) = ctx.get_next_message_ticket() {
 
            let message = ctx.read_message_using_ticket(ticket);
 
            let immediate_result = if let Message::Data(_) = message {
 
                self.handle_new_data_message(ticket, ctx);
 
                None
 
            } else {
 
                match ctx.take_message_using_ticket(ticket) {
 
                    Message::Data(_) => unreachable!(),
 
                    Message::SyncComp(message) => {
 
                        self.handle_new_sync_comp_message(message, ctx)
 
                    },
 
                    Message::SyncPort(message) => {
 
                        self.handle_new_sync_port_message(message, ctx);
 
                        None
 
                    },
 
                    Message::SyncControl(message) => {
 
                        self.handle_new_sync_control_message(message, ctx)
 
                    },
 
                    Message::Control(_) => unreachable!("control message in component"),
 
                }
 
            };
 

	
 
            if let Some(result) = immediate_result {
 
                return Some(result);
 
            }
 
        }
 

	
 
        return None;
 
    }
 

	
 
    pub fn handle_new_data_message(&mut self, ticket: MessageTicket, ctx: &mut ComponentCtx) {
 
        // Go through all branches that are awaiting new messages and see if
 
        // there is one that can receive this message.
 
        if !self.consensus.handle_new_data_message(ticket, ctx) {
 
            // Message should not be handled now
 
            return;
 
        }
 

	
 
        let message = ctx.read_message_using_ticket(ticket).as_data();
 
        let mut iter_id = self.tree.get_queue_first(QueueKind::AwaitingMessage);
 
        while let Some(branch_id) = iter_id {
 
            iter_id = self.tree.get_queue_next(branch_id);
 

	
 
            let branch = &self.tree[branch_id];
 
            if branch.awaiting_port != message.data_header.target_port { continue; }
 
            if !self.consensus.branch_can_receive(branch_id, &message) { continue; }
 

	
 
            // This branch can receive, so fork and given it the message
 
            let receiving_branch_id = self.tree.fork_branch(branch_id);
 
            self.consensus.notify_of_new_branch(branch_id, receiving_branch_id);
 
            let receiving_branch = &mut self.tree[receiving_branch_id];
 

	
 
            debug_assert!(receiving_branch.awaiting_port == message.data_header.target_port);
 
            receiving_branch.awaiting_port = PortIdLocal::new_invalid();
 
            receiving_branch.prepared = PreparedStatement::PerformedGet(message.content.clone());
 
            self.consensus.notify_of_received_message(receiving_branch_id, &message, ctx);
 

	
 
            // And prepare the branch for running
 
            self.tree.push_into_queue(QueueKind::Runnable, receiving_branch_id);
 
        }
 
    }
 

	
 
    pub fn handle_new_sync_comp_message(&mut self, message: SyncCompMessage, ctx: &mut ComponentCtx) -> Option<ConnectorScheduling> {
 
        println!("DEBUG: Actually really handling {:?}", message);
 
        if let Some(round_conclusion) = self.consensus.handle_new_sync_comp_message(message, ctx) {
 
            return Some(self.enter_non_sync_mode(round_conclusion, ctx));
 
        }
 

	
 
        return None;
 
    }
 

	
 
    pub fn handle_new_sync_port_message(&mut self, message: SyncPortMessage, ctx: &mut ComponentCtx) {
 
        self.consensus.handle_new_sync_port_message(message, ctx);
 
    }
 

	
 
    pub fn handle_new_sync_control_message(&mut self, message: SyncControlMessage, ctx: &mut ComponentCtx) -> Option<ConnectorScheduling> {
 
        if let Some(round_conclusion) = self.consensus.handle_new_sync_control_message(message, ctx) {
 
            return Some(self.enter_non_sync_mode(round_conclusion, ctx));
 
        }
 

	
 
        return None;
 
    }
 

	
 
    // --- Running code
 

	
 
    pub fn run_in_sync_mode(&mut self, sched_ctx: SchedulerCtx, comp_ctx: &mut ComponentCtx) -> ConnectorScheduling {
 
        // Check if we have any branch that needs running
 
        debug_assert!(self.tree.is_in_sync() && self.consensus.is_in_sync());
 
        let branch_id = self.tree.pop_from_queue(QueueKind::Runnable);
 
        if branch_id.is_none() {
 
            return ConnectorScheduling::NotNow;
 
        }
 

	
 
        // Retrieve the branch and run it
 
        let branch_id = branch_id.unwrap();
 
        let branch = &mut self.tree[branch_id];
 

	
 
        let mut run_context = ConnectorRunContext{
 
            branch_id,
 
            consensus: &self.consensus,
 
            prepared: branch.prepared.take(),
 
        };
 

	
 
        let run_result = Self::run_prompt(&mut branch.code_state, &sched_ctx.runtime.protocol_description, &mut run_context);
 
        if let Err(eval_error) = run_result {
 
            self.eval_error = Some(eval_error);
 
            self.mode = Mode::SyncError;
 
            if let Some(conclusion) = self.consensus.notify_of_fatal_branch(branch_id, comp_ctx) {
 
                // We can exit immediately
 
                return self.enter_non_sync_mode(conclusion, comp_ctx);
 
            } else {
 
                // Current branch failed. But we may have other things that are
 
                // running.
 
                return ConnectorScheduling::Immediate;
 
            }
 
        }
 
        let run_result = run_result.unwrap();
 

	
 
        // Handle the returned result. Note that this match statement contains
 
        // explicit returns in case the run result requires that the component's
 
        // code is ran again immediately
 
        match run_result {
 
            EvalContinuation::BranchInconsistent => {
 
                // Branch became inconsistent
 
                branch.sync_state = SpeculativeState::Inconsistent;
 
            },
 
            EvalContinuation::BlockFires(port_id) => {
 
                // Branch called `fires()` on a port that has not been used yet.
 
                let port_id = PortIdLocal::new(port_id.0.u32_suffix);
 

	
 
                // Create two forks, one that assumes the port will fire, and
 
                // one that assumes the port remains silent
 
                branch.sync_state = SpeculativeState::HaltedAtBranchPoint;
 

	
 
                let firing_branch_id = self.tree.fork_branch(branch_id);
 
                let silent_branch_id = self.tree.fork_branch(branch_id);
 
                self.consensus.notify_of_new_branch(branch_id, firing_branch_id);
 
                let _result = self.consensus.notify_of_speculative_mapping(firing_branch_id, port_id, true, comp_ctx);
 
                debug_assert_eq!(_result, Consistency::Valid);
 
                self.consensus.notify_of_new_branch(branch_id, silent_branch_id);
 
                let _result = self.consensus.notify_of_speculative_mapping(silent_branch_id, port_id, false, comp_ctx);
 
                debug_assert_eq!(_result, Consistency::Valid);
 

	
 
                // Somewhat important: we push the firing one first, such that
 
                // that branch is ran again immediately.
 
                self.tree.push_into_queue(QueueKind::Runnable, firing_branch_id);
 
                self.tree.push_into_queue(QueueKind::Runnable, silent_branch_id);
 

	
 
                return ConnectorScheduling::Immediate;
 
            },
 
            EvalContinuation::BlockGet(port_id) => {
 
                // Branch performed a `get()` on a port that does not have a
 
                // received message on that port.
 
                let port_id = PortIdLocal::new(port_id.0.u32_suffix);
 

	
 
                branch.sync_state = SpeculativeState::HaltedAtBranchPoint;
 
                branch.awaiting_port = port_id;
 
                self.tree.push_into_queue(QueueKind::AwaitingMessage, branch_id);
 

	
 
                // Note: we only know that a branch is waiting on a message when
 
                // it reaches the `get` call. But we might have already received
 
                // a message that targets this branch, so check now.
 
                let mut any_message_received = false;
 
                for message in comp_ctx.get_read_data_messages(port_id) {
 
                    if self.consensus.branch_can_receive(branch_id, &message) {
 
                        // This branch can receive the message, so we do the
 
                        // fork-and-receive dance
 
                        let receiving_branch_id = self.tree.fork_branch(branch_id);
 
                        let branch = &mut self.tree[receiving_branch_id];
 
                        branch.awaiting_port = PortIdLocal::new_invalid();
 
                        branch.prepared = PreparedStatement::PerformedGet(message.content.clone());
 

	
 
                        self.consensus.notify_of_new_branch(branch_id, receiving_branch_id);
 
                        self.consensus.notify_of_received_message(receiving_branch_id, &message, comp_ctx);
 
                        self.tree.push_into_queue(QueueKind::Runnable, receiving_branch_id);
 

	
 
                        any_message_received = true;
 
                    }
 
                }
 

	
 
                if any_message_received {
 
                    return ConnectorScheduling::Immediate;
 
                }
 
            }
 
            EvalContinuation::SyncBlockEnd => {
 
                let consistency = self.consensus.notify_of_finished_branch(branch_id);
 
                if consistency == Consistency::Valid {
 
                    branch.sync_state = SpeculativeState::ReachedSyncEnd;
 
                    self.tree.push_into_queue(QueueKind::FinishedSync, branch_id);
 
                } else {
 
                    branch.sync_state = SpeculativeState::Inconsistent;
 
                }
 
            },
 
            EvalContinuation::NewFork => {
 
                // Like the `NewChannel` result. This means we're setting up
 
                // a branch and putting a marker inside the RunContext for the
 
                // next time we run the PDL code
 
                let left_id = branch_id;
 
                let right_id = self.tree.fork_branch(left_id);
 
                self.consensus.notify_of_new_branch(left_id, right_id);
 
                self.tree.push_into_queue(QueueKind::Runnable, left_id);
 
                self.tree.push_into_queue(QueueKind::Runnable, right_id);
 

	
 
                let left_branch = &mut self.tree[left_id];
 
                left_branch.prepared = PreparedStatement::ForkedExecution(true);
 
                let right_branch = &mut self.tree[right_id];
 
                right_branch.prepared = PreparedStatement::ForkedExecution(false);
 
            }
 
            EvalContinuation::Put(port_id, content) => {
 
                // Branch is attempting to send data
 
                let port_id = PortIdLocal::new(port_id.0.u32_suffix);
 
                let (sync_header, data_header) = self.consensus.handle_message_to_send(branch_id, port_id, &content, comp_ctx);
 
                let message = DataMessage{ sync_header, data_header, content };
 
                match comp_ctx.submit_message(Message::Data(message)) {
 
                    Ok(_) => {
 
                        // Message is underway
 
                        branch.prepared = PreparedStatement::PerformedPut;
 
                        self.tree.push_into_queue(QueueKind::Runnable, branch_id);
 
                        return ConnectorScheduling::Immediate;
 
                    },
 
                    Err(_) => {
 
                        // We don't own the port
 
                        let pd = &sched_ctx.runtime.protocol_description;
 
                        let eval_error = branch.code_state.new_error_at_expr(
 
                            &pd.modules, &pd.heap,
 
                            String::from("attempted to 'put' on port that is no longer owned")
 
                        );
 
                        self.eval_error = Some(eval_error);
 
                        self.mode = Mode::SyncError;
 

	
 
                        println!("DEBUGERINO: Notify of fatal branch");
 
                        if let Some(conclusion) = self.consensus.notify_of_fatal_branch(branch_id, comp_ctx) {
 
                            println!("DEBUGERINO: Actually got {:?}", conclusion);
 
                            return self.enter_non_sync_mode(conclusion, comp_ctx);
 
                        }
 
                    }
 
                }
 
            },
 
            _ => unreachable!("unexpected run result {:?} in sync mode", run_result),
 
        }
 

	
 
        // If here then the run result did not require a particular action. We
 
        // return whether we have more active branches to run or not.
 
        if self.tree.queue_is_empty(QueueKind::Runnable) {
 
            return ConnectorScheduling::NotNow;
 
        } else {
 
            return ConnectorScheduling::Later;
 
        }
 
    }
 

	
 
    pub fn run_in_deterministic_mode(&mut self, sched_ctx: SchedulerCtx, comp_ctx: &mut ComponentCtx) -> ConnectorScheduling {
 
        debug_assert!(!self.tree.is_in_sync() && !self.consensus.is_in_sync());
 

	
 
        let branch = self.tree.base_branch_mut();
 
        debug_assert!(branch.sync_state == SpeculativeState::RunningNonSync);
 

	
 
        let mut run_context = ConnectorRunContext{
 
            branch_id: branch.id,
 
            consensus: &self.consensus,
 
            prepared: branch.prepared.take(),
 
        };
 
        let run_result = Self::run_prompt(&mut branch.code_state, &sched_ctx.runtime.protocol_description, &mut run_context);
 
        if let Err(eval_error) = run_result {
 
            comp_ctx.push_error(eval_error);
 
            return ConnectorScheduling::Exit
 
        }
 
        let run_result = run_result.unwrap();
 

	
 
        match run_result {
 
            EvalContinuation::ComponentTerminated => {
 
                branch.sync_state = SpeculativeState::Finished;
 
                return ConnectorScheduling::Exit;
 
            },
 
            EvalContinuation::SyncBlockStart => {
 
                comp_ctx.notify_sync_start();
 
                let sync_branch_id = self.tree.start_sync();
 
                debug_assert!(self.last_finished_handled.is_none());
 
                self.consensus.start_sync(comp_ctx);
 
                self.consensus.notify_of_new_branch(BranchId::new_invalid(), sync_branch_id);
src/runtime2/consensus.rs
Show inline comments
 
use crate::collections::VecSet;
 

	
 
use crate::protocol::eval::ValueGroup;
 

	
 
use super::ConnectorId;
 
use super::branch::BranchId;
 
use super::port::{ChannelId, PortIdLocal, PortState};
 
use super::inbox::{
 
    Message, DataHeader, SyncHeader, ChannelAnnotation, BranchMarker,
 
    DataMessage,
 
    SyncCompMessage, SyncCompContent,
 
    SyncPortMessage, SyncPortContent,
 
    SyncControlMessage, SyncControlContent
 
};
 
use super::scheduler::{ComponentCtx, ComponentPortChange, MessageTicket};
 

	
 
struct BranchAnnotation {
 
    channel_mapping: Vec<ChannelAnnotation>,
 
    cur_marker: BranchMarker,
 
}
 

	
 
#[derive(Debug)]
 
pub(crate) struct LocalSolution {
 
    component: ConnectorId,
 
    final_branch_id: BranchId,
 
    sync_round_number: u32,
 
    port_mapping: Vec<(ChannelId, BranchMarker)>,
 
}
 

	
 
#[derive(Debug, Clone)]
 
pub(crate) struct GlobalSolution {
 
    component_branches: Vec<(ConnectorId, BranchId, u32)>,
 
    channel_mapping: Vec<(ChannelId, BranchMarker)>, // TODO: This can go, is debugging info
 
}
 

	
 
#[derive(Debug, PartialEq, Eq)]
 
pub enum RoundConclusion {
 
    Failure,
 
    Success(BranchId),
 
}
 

	
 
// -----------------------------------------------------------------------------
 
// Consensus
 
// -----------------------------------------------------------------------------
 

	
 
#[derive(Debug)]
 
struct Peer {
 
    id: ConnectorId,
 
    encountered_this_round: bool,
 
    expected_sync_round: u32,
 
}
 

	
 
/// The consensus algorithm. Currently only implemented to find the component
 
/// with the highest ID within the sync region and letting it handle all the
 
/// local solutions.
 
///
 
/// The type itself serves as an experiment to see how code should be organized.
 
// TODO: Flatten all datastructures
 
// TODO: Have a "branch+port position hint" in case multiple operations are
 
//  performed on the same port to prevent repeated lookups
 
// TODO: A lot of stuff should be batched. Like checking all the sync headers
 
//  and sending "I have a higher ID" messages. Should reduce locking by quite a
 
//  bit.
 
// TODO: Needs a refactor. Firstly we have cases where we don't have a branch ID
 
//  but we do want to enumerate all current ports. So put that somewhere in a
 
//  central place. Secondly. Error handling and regular message handling is
 
//  becoming a mess.
 
pub(crate) struct Consensus {
 
    // --- State that is cleared after each round
 
    // Local component's state
 
    highest_connector_id: ConnectorId,
 
    branch_annotations: Vec<BranchAnnotation>, // index is branch ID
 
    branch_markers: Vec<BranchId>, // index is branch marker, maps to branch
 
    // Gathered state from communication
 
    encountered_ports: VecSet<PortIdLocal>, // to determine if we should send "port remains silent" messages.
 
    solution_combiner: SolutionCombiner,
 
    handled_wave: bool, // encountered notification wave in this round
 
    conclusion: Option<RoundConclusion>,
 
    ack_remaining: u32,
 
    // --- Persistent state
 
    peers: Vec<Peer>,
 
    sync_round: u32,
 
    // --- Workspaces
 
    workspace_ports: Vec<PortIdLocal>,
 
}
 

	
 
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
 
pub(crate) enum Consistency {
 
    Valid,
 
    Inconsistent,
 
}
 

	
 
#[derive(Debug, PartialEq, Eq)]
 
pub(crate) enum MessageOrigin {
 
    Past,
 
    Present,
 
    Future
 
}
 

	
 
impl Consensus {
 
    pub fn new() -> Self {
 
        return Self {
 
            highest_connector_id: ConnectorId::new_invalid(),
 
            branch_annotations: Vec::new(),
 
            branch_markers: Vec::new(),
 
            encountered_ports: VecSet::new(),
 
            solution_combiner: SolutionCombiner::new(),
 
            handled_wave: false,
 
            conclusion: None,
 
            ack_remaining: 0,
 
            peers: Vec::new(),
 
            sync_round: 0,
 
            workspace_ports: Vec::new(),
 
        }
 
    }
 

	
 
    // --- Controlling sync round and branches
 

	
 
    /// Returns whether the consensus algorithm is running in sync mode
 
    pub fn is_in_sync(&self) -> bool {
 
        return !self.branch_annotations.is_empty();
 
    }
 

	
 
    /// TODO: Remove this once multi-fire is in place
 
    #[deprecated]
 
    pub fn get_annotation(&self, branch_id: BranchId, channel_id: PortIdLocal) -> &ChannelAnnotation {
 
        let branch = &self.branch_annotations[branch_id.index as usize];
 
        let port = branch.channel_mapping.iter().find(|v| v.channel_id.index == channel_id.index).unwrap();
 
        return port;
 
    }
 

	
 
    /// Sets up the consensus algorithm for a new synchronous round. The
 
    /// provided ports should be the ports the component owns at the start of
 
    /// the sync round.
 
    pub fn start_sync(&mut self, ctx: &ComponentCtx) {
 
        debug_assert!(!self.highest_connector_id.is_valid());
 
        debug_assert!(self.branch_annotations.is_empty());
 
        debug_assert!(self.solution_combiner.local.is_empty());
 

	
 
        // We'll use the first "branch" (the non-sync one) to store our ports,
 
        // this allows cloning if we created a new branch.
 
        self.branch_annotations.push(BranchAnnotation{
 
            channel_mapping: ctx.get_ports().iter()
 
                .map(|v| ChannelAnnotation {
 
                    channel_id: v.channel_id,
 
                    registered_id: None,
 
                    expected_firing: None,
 
                })
 
                .collect(),
 
            cur_marker: BranchMarker::new_invalid(),
 
        });
 
        self.branch_markers.push(BranchId::new_invalid());
 

	
 
        self.highest_connector_id = ctx.id;
 

	
 
    }
 

	
 
    /// Notifies the consensus algorithm that a new branch has appeared. Must be
 
    /// called for each forked branch in the execution tree.
 
    pub fn notify_of_new_branch(&mut self, parent_branch_id: BranchId, new_branch_id: BranchId) {
 
        // If called correctly. Then each time we are notified the new branch's
 
        // index is the length in `branch_annotations`.
 
        debug_assert!(self.branch_annotations.len() == new_branch_id.index as usize);
 
        let parent_branch_annotations = &self.branch_annotations[parent_branch_id.index as usize];
 
        let new_marker = BranchMarker::new(self.branch_markers.len() as u32);
 
        let new_branch_annotations = BranchAnnotation{
 
            channel_mapping: parent_branch_annotations.channel_mapping.clone(),
 
            cur_marker: new_marker,
 
        };
 
        self.branch_annotations.push(new_branch_annotations);
 
        self.branch_markers.push(new_branch_id);
 
    }
 

	
 
    /// Notifies the consensus algorithm that a particular branch has
 
    /// encountered an unrecoverable error.
 
    pub fn notify_of_fatal_branch(&mut self, failed_branch_id: BranchId, ctx: &mut ComponentCtx) -> Option<RoundConclusion> {
 
        debug_assert!(self.is_in_sync());
 

	
 
        // Check for trivial case, where branch has not yet communicated within
 
        // the consensus algorithm
 
        let branch = &self.branch_annotations[failed_branch_id.index as usize];
 
        if branch.channel_mapping.iter().all(|v| v.registered_id.is_none()) {
 
            println!("DEBUG: Failure everything silent");
 
            return Some(RoundConclusion::Failure);
 
        }
 

	
 
        // We're not in the trivial case: since we've communicated we need to
 
        // let everyone know that this round is probably not going to end well.
 
        return self.initiate_sync_failure(ctx);
 
    }
 

	
 
    /// Notifies the consensus algorithm that a branch has reached the end of
 
    /// the sync block. A final check for consistency will be performed that the
 
    /// caller has to handle. Note that
 
    pub fn notify_of_finished_branch(&self, branch_id: BranchId) -> Consistency {
 
        debug_assert!(self.is_in_sync());
 
        let branch = &self.branch_annotations[branch_id.index as usize];
 
        for mapping in &branch.channel_mapping {
 
            match mapping.expected_firing {
 
                Some(expected) => {
 
                    if expected != mapping.registered_id.is_some() {
 
                        // Inconsistent speculative state and actual state
 
                        debug_assert!(mapping.registered_id.is_none()); // because if we did fire on a silent port, we should've caught that earlier
 
                        return Consistency::Inconsistent;
 
                    }
 
                },
 
                None => {},
 
            }
 
        }
 

	
 
        return Consistency::Valid;
 
    }
 

	
 
    /// Notifies the consensus algorithm that a particular branch has assumed
 
    /// a speculative value for its port mapping.
 
    pub fn notify_of_speculative_mapping(&mut self, branch_id: BranchId, port_id: PortIdLocal, does_fire: bool, ctx: &ComponentCtx) -> Consistency {
 
        debug_assert!(self.is_in_sync());
 

	
 
        let port_desc = ctx.get_port_by_id(port_id).unwrap();
 
        let channel_id = port_desc.channel_id;
 
        let branch = &mut self.branch_annotations[branch_id.index as usize];
 
        for mapping in &mut branch.channel_mapping {
 
            if mapping.channel_id == channel_id {
 
                match mapping.expected_firing {
 
                    None => {
 
                        // Not yet mapped, perform speculative mapping
 
                        mapping.expected_firing = Some(does_fire);
 
                        return Consistency::Valid;
 
                    },
 
                    Some(current) => {
 
                        // Already mapped
 
                        if current == does_fire {
 
                            return Consistency::Valid;
 
                        } else {
 
                            return Consistency::Inconsistent;
 
                        }
 
                    }
 
                }
 
            }
 
        }
 

	
 
        unreachable!("notify_of_speculative_mapping called with unowned port");
 
    }
 

	
 
    /// Generates a new local solution from a finished branch. If the component
 
    /// is not the leader of the sync region then it will be sent to the
 
    /// appropriate component. If it is the leader then there is a chance that
 
    /// this solution completes a global solution. In that case the solution
 
    /// branch ID will be returned.
 
    pub(crate) fn handle_new_finished_sync_branch(&mut self, branch_id: BranchId, ctx: &mut ComponentCtx) -> Option<RoundConclusion> {
 
        // Turn the port mapping into a local solution
 
        let source_mapping = &self.branch_annotations[branch_id.index as usize].channel_mapping;
 
        let mut target_mapping = Vec::with_capacity(source_mapping.len());
 

	
 
        for port in source_mapping {
 
            // Note: if the port is silent, and we've never communicated
 
            // over the port, then we need to do so now, to let the peer
 
            // component know about our sync leader state.
 
            let port_desc = ctx.get_port_by_channel_id(port.channel_id).unwrap();
 
            let self_port_id = port_desc.self_id;
 
            let peer_port_id = port_desc.peer_id;
 
            let channel_id = port_desc.channel_id;
 

	
 
            if !self.encountered_ports.contains(&self_port_id) {
 
                let message = SyncPortMessage {
 
                    sync_header: SyncHeader{
 
                        sending_component_id: ctx.id,
 
                        highest_component_id: self.highest_connector_id,
 
                        sync_round: self.sync_round
 
                    },
 
                    source_port: self_port_id,
 
                    target_port: peer_port_id,
 
                    content: SyncPortContent::SilentPortNotification,
 
                };
 
                match ctx.submit_message(Message::SyncPort(message)) {
 
                    Ok(_) => {
 
                        self.encountered_ports.push(self_port_id);
 
                    },
 
                    Err(_) => {
 
                        // Seems like we were done with this branch, but one of
 
                        // the silent ports (in scope) is actually closed
 
                        return self.notify_of_fatal_branch(branch_id, ctx);
 
                    }
 
                }
 
            }
 

	
 
            target_mapping.push((
 
                channel_id,
 
                port.registered_id.unwrap_or(BranchMarker::new_invalid())
 
            ));
 
        }
 

	
 
        let local_solution = LocalSolution{
 
            component: ctx.id,
 
            sync_round_number: self.sync_round,
 
            final_branch_id: branch_id,
 
            port_mapping: target_mapping,
 
        };
 
        let maybe_conclusion = self.send_to_leader_or_handle_as_leader(SyncCompContent::LocalSolution(local_solution), ctx);
 
        return maybe_conclusion;
 
    }
 

	
 
    /// Notifies the consensus algorithm about the chosen branch to commit to
 
    /// memory (may be the invalid "start" branch)
 
    pub fn end_sync(&mut self, branch_id: BranchId, final_ports: &mut Vec<ComponentPortChange>) {
 
        debug_assert!(self.is_in_sync());
 

	
 
        // TODO: Handle sending and receiving ports
 
        // Set final ports
 
        let branch = &self.branch_annotations[branch_id.index as usize];
 

	
 
        // Clear out internal storage to defaults
 
        println!("DEBUG: ***** Incrementing sync round stuff");
 
        self.highest_connector_id = ConnectorId::new_invalid();
 
        self.branch_annotations.clear();
 
        self.branch_markers.clear();
 
        self.encountered_ports.clear();
 
        self.solution_combiner.clear();
 
        self.handled_wave = false;
 
        self.conclusion = None;
 
        self.ack_remaining = 0;
 

	
 
        // And modify persistent storage
 
        self.sync_round += 1;
 

	
 
        for peer in self.peers.iter_mut() {
 
            peer.encountered_this_round = false;
 
            peer.expected_sync_round += 1;
 
        }
 

	
 
        println!("DEBUG: ***** Peers post round are:\n{:#?}", &self.peers)
 
    }
 

	
 
    // --- Handling messages
 

	
 
    /// Prepares a message for sending. Caller should have made sure that
 
    /// sending the message is consistent with the speculative state.
 
    pub fn handle_message_to_send(&mut self, branch_id: BranchId, source_port_id: PortIdLocal, content: &ValueGroup, ctx: &mut ComponentCtx) -> (SyncHeader, DataHeader) {
 
        debug_assert!(self.is_in_sync());
 
        let branch = &mut self.branch_annotations[branch_id.index as usize];
 
        let port_info = ctx.get_port_by_id(source_port_id).unwrap();
 

	
 
        if cfg!(debug_assertions) {
 
            // Check for consistent mapping
 
            let port = branch.channel_mapping.iter()
 
                .find(|v| v.channel_id == port_info.channel_id)
 
                .unwrap();
 
            debug_assert!(port.expected_firing == None || port.expected_firing == Some(true));
 
        }
 

	
 
        // Check for ports that are being sent
 
        debug_assert!(self.workspace_ports.is_empty());
 
        find_ports_in_value_group(content, &mut self.workspace_ports);
 
        if !self.workspace_ports.is_empty() {
 
            todo!("handle sending ports");
 
            self.workspace_ports.clear();
 
        }
 

	
 
        // Construct data header
 
        let data_header = DataHeader{
 
            expected_mapping: branch.channel_mapping.iter()
 
                .filter(|v| v.registered_id.is_some() || v.channel_id == port_info.channel_id)
 
                .copied()
 
                .collect(),
 
            sending_port: port_info.self_id,
 
            target_port: port_info.peer_id,
 
            new_mapping: branch.cur_marker,
 
        };
 

	
 
        // Update port mapping
 
        for mapping in &mut branch.channel_mapping {
 
            if mapping.channel_id == port_info.channel_id {
 
                mapping.expected_firing = Some(true);
 
                mapping.registered_id = Some(branch.cur_marker);
 
            }
 
        }
 

	
 
        // Update branch marker
 
        let new_marker = BranchMarker::new(self.branch_markers.len() as u32);
 
        branch.cur_marker = new_marker;
 
        self.branch_markers.push(branch_id);
 

	
 
        self.encountered_ports.push(source_port_id);
 

	
 
        return (self.create_sync_header(ctx), data_header);
 
    }
 

	
 
    /// Handles a new data message by handling the sync header. The caller is
 
    /// responsible for checking for branches that might be able to receive
 
    /// the message.
 
    pub fn handle_new_data_message(&mut self, ticket: MessageTicket, ctx: &mut ComponentCtx) -> bool {
 
        let message = ctx.read_message_using_ticket(ticket).as_data();
 
        let target_port = message.data_header.target_port;
 
        match self.handle_received_sync_header(message.sync_header, ctx) {
 
            MessageOrigin::Past => return false,
 
            MessageOrigin::Present => {
 
                self.encountered_ports.push(target_port);
 
                return true;
 
            },
 
            MessageOrigin::Future => {
 
                let message = ctx.take_message_using_ticket(ticket);
 
                ctx.put_back_message(message);
 
                return false;
 
            }
 
        }
 
    }
 

	
 
    /// Handles a new sync message by handling the sync header and the contents
 
    /// of the message. Returns `Some` with the branch ID of the global solution
 
    /// if the sync solution has been found.
 
    pub fn handle_new_sync_comp_message(&mut self, message: SyncCompMessage, ctx: &mut ComponentCtx) -> Option<RoundConclusion> {
 
        match self.handle_received_sync_header(message.sync_header, ctx) {
 
            MessageOrigin::Past => return None,
 
            MessageOrigin::Present => {},
 
            MessageOrigin::Future => {
 
                ctx.put_back_message(Message::SyncComp(message));
 
                return None
 
            }
 
        }
 

	
 
        // And handle the contents
 
        debug_assert_eq!(message.target_component_id, ctx.id);
 

	
 
        match &message.content {
 
            SyncCompContent::LocalFailure |
 
            SyncCompContent::LocalSolution(_) |
 
            SyncCompContent::PartialSolution(_) |
 
            SyncCompContent::AckFailure |
 
            SyncCompContent::Presence(_) => {
 
                // Needs to be handled by the leader
 
                return self.send_to_leader_or_handle_as_leader(message.content, ctx);
 
            },
 
            SyncCompContent::GlobalSolution(solution) => {
 
                // Found a global solution
 
                debug_assert_ne!(self.highest_connector_id, ctx.id); // not the leader
 
                let (_, branch_id, _) = solution.component_branches.iter()
 
                    .find(|(component_id, _, _)| *component_id == ctx.id)
 
                    .unwrap();
 
                return Some(RoundConclusion::Success(*branch_id));
 
            },
 
            SyncCompContent::GlobalFailure => {
 
                // Global failure of round, send Ack to leader
 
                println!("DEBUGERINO: Got GlobalFailure, sending Ack in response");
 
                debug_assert_ne!(self.highest_connector_id, ctx.id); // not the leader
 
                let _result = self.send_to_leader_or_handle_as_leader(SyncCompContent::AckFailure, ctx);
 
                debug_assert!(_result.is_none());
 
                return Some(RoundConclusion::Failure);
 
            },
 
            SyncCompContent::Notification => {
 
                // We were just interested in the sync header we handled above
 
                return None;
 
            }
 
        }
 
    }
 

	
 
    pub fn handle_new_sync_port_message(&mut self, message: SyncPortMessage, ctx: &mut ComponentCtx) -> Option<RoundConclusion> {
 
        match self.handle_received_sync_header(message.sync_header, ctx) {
 
            MessageOrigin::Past => return None,
 
            MessageOrigin::Present => {},
 
            MessageOrigin::Future => {
 
                ctx.put_back_message(Message::SyncPort(message));
 
                return None;
 
            }
 
        }
 

	
 
        debug_assert!(self.is_in_sync());
 
        debug_assert!(ctx.get_port_by_id(message.target_port).is_some());
 
        match message.content {
 
            SyncPortContent::SilentPortNotification => {
 
                // The point here is to let us become part of the sync round and
 
                // take note of the leader in case all of our ports are silent.
 
                self.encountered_ports.push(message.target_port);
 
                return None
 
            }
 
            SyncPortContent::NotificationWave => {
 
                // Wave to discover everyone in the network, handling sync
 
                // header takes care of leader discovery, here we need to make
 
                // sure we propagate the wave
 
                if self.handled_wave {
 
                    return None;
 
                }
 

	
 
                self.handled_wave = true;
 

	
 
                // Propagate wave to all peers except the one that has sent us
 
                // the wave.
 
                for mapping in &self.branch_annotations[0].channel_mapping {
 
                    let channel_id = mapping.channel_id;
 
                    let port_desc = ctx.get_port_by_channel_id(channel_id).unwrap();
 
                    if port_desc.self_id == message.target_port {
 
                        // Wave came from this port, no need to send one back
 
                        continue;
 
                    }
 

	
 
                    let message = SyncPortMessage{
 
                        sync_header: self.create_sync_header(ctx),
 
                        source_port: port_desc.self_id,
 
                        target_port: port_desc.peer_id,
 
                        content: SyncPortContent::NotificationWave,
 
                    };
 
                    // As with the other SyncPort where we throw away the
 
                    // result: we're dealing with an error here anyway
 
                    let _unused = ctx.submit_message(Message::SyncPort(message));
 
                }
 

	
 
                // And let the leader know about our port state
 
                let annotations = &self.branch_annotations[0];
 
                let mut channels = Vec::with_capacity(annotations.channel_mapping.len());
src/runtime2/inbox.rs
Show inline comments
 
use std::sync::Mutex;
 
use std::collections::VecDeque;
 

	
 
use crate::protocol::eval::ValueGroup;
 
use crate::runtime2::consensus::{ComponentPresence, SolutionCombiner};
 
use crate::runtime2::port::ChannelId;
 

	
 
use super::ConnectorId;
 
use super::consensus::{GlobalSolution, LocalSolution};
 
use super::port::PortIdLocal;
 

	
 
// TODO: Remove Debug derive from all types
 

	
 
#[derive(Debug, Copy, Clone)]
 
pub(crate) struct ChannelAnnotation {
 
    pub channel_id: ChannelId,
 
    pub registered_id: Option<BranchMarker>,
 
    pub expected_firing: Option<bool>,
 
}
 

	
 
/// Marker for a branch in a port mapping. A marker is, like a branch ID, a
 
/// unique identifier for a branch, but differs in that a branch only has one
 
/// branch ID, but might have multiple associated markers (i.e. one branch
 
/// performing a `put` three times will generate three markers.
 
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
 
pub(crate) struct BranchMarker{
 
    marker: u32,
 
}
 

	
 
impl BranchMarker {
 
    #[inline]
 
    pub(crate) fn new(marker: u32) -> Self {
 
        debug_assert!(marker != 0);
 
        return Self{ marker };
 
    }
 

	
 
    #[inline]
 
    pub(crate) fn new_invalid() -> Self {
 
        return Self{ marker: 0 }
 
    }
 
}
 

	
 
/// The header added by the synchronization algorithm to all.
 
#[derive(Debug, Clone, Copy)]
 
pub(crate) struct SyncHeader {
 
    pub sending_component_id: ConnectorId,
 
    pub highest_component_id: ConnectorId,
 
    pub sync_round: u32,
 
}
 

	
 
/// The header added to data messages
 
#[derive(Debug, Clone)]
 
pub(crate) struct DataHeader {
 
    pub expected_mapping: Vec<ChannelAnnotation>,
 
    pub sending_port: PortIdLocal,
 
    pub target_port: PortIdLocal,
 
    pub new_mapping: BranchMarker,
 
}
 

	
 
/// A data message is a message that is intended for the receiver's PDL code,
 
/// but will also be handled by the consensus algorithm
 
#[derive(Debug, Clone)]
 
pub(crate) struct DataMessage {
 
    pub sync_header: SyncHeader,
 
    pub data_header: DataHeader,
 
    pub content: ValueGroup,
 
}
 

	
 
#[derive(Debug)]
 
pub(crate) enum SyncCompContent {
 
    LocalFailure, // notifying leader that component has failed (e.g. timeout, whatever)
 
    LocalSolution(LocalSolution), // sending a local solution to the leader
 
    PartialSolution(SolutionCombiner), // when new leader is detected, forward all local results
 
    GlobalSolution(GlobalSolution), // broadcasting to everyone
 
    GlobalFailure, // broadcasting to everyone
 
    AckFailure, // acknowledgement of failure to leader
 
    Notification, // just a notification (so purpose of message is to send the SyncHeader)
 
    Presence(ComponentPresence), // notifying leader of component presence (needed to ensure failing a round involves all components in a sync round)
 
}
 

	
 
/// A sync message is a message that is intended only for the consensus
 
/// algorithm. The message goes directly to a component.
 
#[derive(Debug)]
 
pub(crate) struct SyncCompMessage {
 
    pub sync_header: SyncHeader,
 
    pub target_component_id: ConnectorId,
 
    pub content: SyncCompContent,
 
}
 

	
 
#[derive(Debug)]
 
pub(crate) enum SyncPortContent {
 
    SilentPortNotification,
 
    NotificationWave,
 
}
 

	
 
/// A sync message intended for the consensus algorithm. This message does not
 
/// go to a component, but through a channel (and results in potential
 
/// rerouting) because we're not sure about the ID of the component that holds
 
/// the other end of the channel.
 
#[derive(Debug)]
 
pub(crate) struct SyncPortMessage {
 
    pub sync_header: SyncHeader,
 
    pub source_port: PortIdLocal,
 
    pub target_port: PortIdLocal,
 
    pub content: SyncPortContent,
 
}
 

	
 
#[derive(Debug)]
 
pub(crate) enum SyncControlContent {
 
    ChannelIsClosed(PortIdLocal), // contains port that is owned by the recipient of the message
 
}
 

	
 
/// A sync control message: originating from the scheduler, but intended for the
 
/// current sync round of the recipient. Every kind of consensus algorithm must
 
/// be able to handle such a message.
 
#[derive(Debug)]
 
pub(crate) struct SyncControlMessage {
 
    // For now these control messages are only aimed at components. Might change
 
    // in the future. But for now we respond to messages from components that
 
    // have, because of that message, published their ID.
 
    pub in_response_to_sync_round: u32,
 
    pub target_component_id: ConnectorId,
 
    pub content: SyncControlContent,
 
}
 

	
 
/// A control message is a message intended for the scheduler that is executing
 
/// a component.
 
#[derive(Debug)]
 
pub(crate) struct ControlMessage {
 
    pub id: u32, // generic identifier, used to match request to response
 
    pub sending_component_id: ConnectorId,
 
    pub content: ControlContent,
 
}
 

	
 
#[derive(Debug)]
 
pub(crate) enum ControlContent {
 
    PortPeerChanged(PortIdLocal, ConnectorId),
 
    CloseChannel(PortIdLocal),
 
    Ack,
 
    Ping,
 
}
 

	
 
/// Combination of data message and control messages.
 
#[derive(Debug)]
 
pub(crate) enum Message {
 
    Data(DataMessage),
 
    SyncComp(SyncCompMessage),
 
    SyncPort(SyncPortMessage),
 
    SyncControl(SyncControlMessage),
 
    Control(ControlMessage),
 
}
 

	
 
impl Message {
 
    /// If the message is sent through a particular channel, then this function
 
    /// returns the port through which the message was sent.
 
    pub(crate) fn source_port(&self) -> Option<PortIdLocal> {
 
        // Currently only data messages have a source port
 
        match self {
 
            Message::Data(message) => return Some(message.data_header.sending_port),
 
            Message::SyncPort(message) => return Some(message.source_port),
 
            Message::SyncComp(_) => return None,
 
            Message::SyncControl(_) => return None,
 
            Message::Control(_) => return None,
 
        }
 
    }
 

	
 
    /// If the message is sent through a particular channel, then this function
 
    /// returns the target port through which the message was sent.
 
    pub(crate) fn target_port(&self) -> Option<PortIdLocal> {
 
        match self {
 
            Message::Data(message) => return Some(message.data_header.target_port),
 
            Message::SyncPort(message) => return Some(message.target_port),
 
            Message::SyncComp(_) => return None,
 
            Message::SyncControl(_) => return None,
 
            Message::Control(message) => {
 
                match &message.content {
 
                    ControlContent::PortPeerChanged(port_id, _) => return Some(*port_id),
 
                    ControlContent::CloseChannel(port_id) => return Some(*port_id),
 
                    ControlContent::Ping => return None,
 
                    ControlContent::Ack => return None,
 
                }
 
            }
 
        }
 
    }
 

	
 
    pub(crate) fn source_component(&self) -> Option<ConnectorId> {
 
        match self {
 
            Message::Data(message) => Some(message.sync_header.sending_component_id),
 
            Message::SyncPort(message) => Some(message.sync_header.sending_component_id),
 
            Message::SyncComp(message) => Some(message.sync_header.sending_component_id),
 
            Message::SyncControl(_) => None,
 
            Message::Control(message) => Some(message.sending_component_id)
 
        }
 
    }
 

	
 
    pub(crate) fn as_data(&self) -> &DataMessage {
 
        match self {
 
            Message::Data(v) => v,
 
            _ => unreachable!(),
 
        }
 
    }
 
}
 

	
 
/// The public inbox of a connector. The thread running the connector that owns
 
/// this inbox may retrieved from it. Non-owning threads may only put new
 
/// messages inside of it.
 
// TODO: @Optimize, lazy concurrency. Probably ringbuffer with read/write heads.
 
//  Should behave as a MPSC queue.
 
pub struct PublicInbox {
 
    messages: Mutex<VecDeque<Message>>,
 
}
 

	
 
impl PublicInbox {
 
    pub fn new() -> Self {
 
        Self{
 
            messages: Mutex::new(VecDeque::new()),
 
        }
 
    }
 

	
 
    pub(crate) fn insert_message(&self, message: Message) {
 
        let mut lock = self.messages.lock().unwrap();
 
        lock.push_back(message);
 
    }
 

	
 
    pub(crate) fn take_message(&self) -> Option<Message> {
 
        let mut lock = self.messages.lock().unwrap();
 
        return lock.pop_front();
 
    }
 

	
 
    pub fn is_empty(&self) -> bool {
 
        let lock = self.messages.lock().unwrap();
 
        return lock.is_empty();
 
    }
 

	
 
    pub fn clear(&self) {
 
        let mut lock = self.messages.lock().unwrap();
 
        lock.clear();
 
    }
 
}
 
\ No newline at end of file
src/runtime2/mod.rs
Show inline comments
 
@@ -128,514 +128,513 @@ pub(crate) struct ScheduledConnector {
 
}
 

	
 
// -----------------------------------------------------------------------------
 
// Runtime
 
// -----------------------------------------------------------------------------
 

	
 
/// Externally facing runtime.
 
pub struct Runtime {
 
    inner: Arc<RuntimeInner>,
 
}
 

	
 
impl Runtime {
 
    pub fn new(num_threads: u32, protocol_description: ProtocolDescription) -> Runtime {
 
        // Setup global state
 
        assert!(num_threads > 0, "need a thread to run connectors");
 
        let runtime_inner = Arc::new(RuntimeInner{
 
            protocol_description,
 
            port_counter: AtomicU32::new(0),
 
            connectors: RwLock::new(ConnectorStore::with_capacity(32)),
 
            connector_queue: Mutex::new(VecDeque::with_capacity(32)),
 
            schedulers: Mutex::new(Vec::new()),
 
            scheduler_notifier: Condvar::new(),
 
            active_connectors: AtomicU32::new(0),
 
            active_interfaces: AtomicU32::new(1), // this `Runtime` instance
 
            should_exit: AtomicBool::new(false),
 
        });
 

	
 
        // Launch threads
 
        {
 
            let mut schedulers = Vec::with_capacity(num_threads as usize);
 
            for thread_index in 0..num_threads {
 
                let cloned_runtime_inner = runtime_inner.clone();
 
                let thread = thread::Builder::new()
 
                    .name(format!("thread-{}", thread_index))
 
                    .spawn(move || {
 
                        let mut scheduler = Scheduler::new(cloned_runtime_inner, thread_index);
 
                        scheduler.run();
 
                    })
 
                    .unwrap();
 

	
 
                schedulers.push(thread);
 
            }
 

	
 
            let mut lock = runtime_inner.schedulers.lock().unwrap();
 
            *lock = schedulers;
 
        }
 

	
 
        // Return runtime
 
        return Runtime{ inner: runtime_inner };
 
    }
 

	
 
    /// Returns a new interface through which channels and connectors can be
 
    /// created.
 
    pub fn create_interface(&self) -> ApplicationInterface {
 
        self.inner.increment_active_interfaces();
 
        let (connector, mut interface) = ConnectorApplication::new(self.inner.clone());
 
        let connector_key = self.inner.create_interface_component(connector);
 
        interface.set_connector_id(connector_key.downcast());
 

	
 
        // Note that we're not scheduling. That is done by the interface in case
 
        // it is actually needed.
 
        return interface;
 
    }
 
}
 

	
 
impl Drop for Runtime {
 
    fn drop(&mut self) {
 
        self.inner.decrement_active_interfaces();
 
        let mut lock = self.inner.schedulers.lock().unwrap();
 
        for handle in lock.drain(..) {
 
            handle.join().unwrap();
 
        }
 
    }
 
}
 

	
 
// -----------------------------------------------------------------------------
 
// RuntimeInner
 
// -----------------------------------------------------------------------------
 

	
 
pub(crate) struct RuntimeInner {
 
    // Protocol
 
    pub(crate) protocol_description: ProtocolDescription,
 
    // Regular counter for port IDs
 
    port_counter: AtomicU32,
 
    // Storage of connectors and the work queue
 
    connectors: RwLock<ConnectorStore>,
 
    connector_queue: Mutex<VecDeque<ConnectorKey>>,
 
    schedulers: Mutex<Vec<JoinHandle<()>>>,
 
    // Conditions to determine whether the runtime can exit
 
    scheduler_notifier: Condvar,  // coupled to mutex on `connector_queue`.
 
    // TODO: Figure out if we can simply merge the counters?
 
    active_connectors: AtomicU32, // active connectors (if sleeping, then still considered active)
 
    active_interfaces: AtomicU32, // active API interfaces that can add connectors/channels
 
    should_exit: AtomicBool,
 
}
 

	
 
impl RuntimeInner {
 
    // --- Managing the components queued for execution
 

	
 
    /// Wait until there is a connector to run. If there is one, then `Some`
 
    /// will be returned. If there is no more work, then `None` will be
 
    /// returned.
 
    pub(crate) fn wait_for_work(&self) -> Option<ConnectorKey> {
 
        let mut lock = self.connector_queue.lock().unwrap();
 
        while lock.is_empty() && !self.should_exit.load(Ordering::Acquire) {
 
            lock = self.scheduler_notifier.wait(lock).unwrap();
 
        }
 

	
 
        return lock.pop_front();
 
    }
 

	
 
    pub(crate) fn push_work(&self, key: ConnectorKey) {
 
        let mut lock = self.connector_queue.lock().unwrap();
 
        lock.push_back(key);
 
        self.scheduler_notifier.notify_one();
 
    }
 

	
 
    // --- Creating/using ports
 

	
 
    /// Creates a new port pair. Note that these are stored globally like the
 
    /// connectors are. Ports stored by components belong to those components.
 
    pub(crate) fn create_channel(&self, creating_connector: ConnectorId) -> (Port, Port) {
 
        use port::{PortIdLocal, PortKind};
 

	
 
        let getter_id = self.port_counter.fetch_add(2, Ordering::SeqCst);
 
        let channel_id = ChannelId::new(getter_id);
 
        let putter_id = PortIdLocal::new(getter_id + 1);
 
        let getter_id = PortIdLocal::new(getter_id);
 

	
 
        let getter_port = Port{
 
            self_id: getter_id,
 
            peer_id: putter_id,
 
            channel_id,
 
            kind: PortKind::Getter,
 
            state: PortState::Open,
 
            peer_connector: creating_connector,
 
        };
 
        let putter_port = Port{
 
            self_id: putter_id,
 
            peer_id: getter_id,
 
            channel_id,
 
            kind: PortKind::Putter,
 
            state: PortState::Open,
 
            peer_connector: creating_connector,
 
        };
 

	
 
        return (getter_port, putter_port);
 
    }
 

	
 
    /// Sends a message directly (without going through the port) to a
 
    /// component. This is slightly less efficient then sending over a port, but
 
    /// might be preferable for some algorithms. If the component was sleeping
 
    /// then it is scheduled for execution.
 
    pub(crate) fn send_message_maybe_destroyed(&self, target_id: ConnectorId, message: Message) -> bool {
 
        let target = {
 
            let mut lock = self.connectors.read().unwrap();
 
            lock.get(target_id.index)
 
        };
 

	
 
        // Do a CAS on the number of users. Most common case the component is
 
        // alive and we're the only one sending the message. Note that if we
 
        // finish this block, we're sure that no-one has set the `num_users`
 
        // value to 0. This is essential! When at 0, the component is added to
 
        // the freelist and the generation counter will be incremented.
 
        let mut cur_num_users = 1;
 
        while let Err(old_num_users) = target.num_users.compare_exchange(cur_num_users, cur_num_users + 1, Ordering::SeqCst, Ordering::Acquire) {
 
            if old_num_users == 0 {
 
                // Cannot send message. Whatever the component state is
 
                // (destroyed, at a different generation number, busy being
 
                // destroyed, etc.) we cannot send the message and will not
 
                // modify the component
 
                return false;
 
            }
 

	
 
            cur_num_users = old_num_users;
 
        }
 

	
 
        // We incremented the counter. But we might still be at the wrong
 
        // generation number. The generation number is a monotonically
 
        // increasing value. Since it only increases when someone gets the
 
        // `num_users` counter to 0, we can simply load the generation number.
 
        let generation = target.generation.load(Ordering::Acquire);
 
        if generation != target_id.generation {
 
            // We're at the wrong generation, so we cannot send the message.
 
            // However, since we incremented the `num_users` counter, the moment
 
            // we decrement it we might be the one that are supposed to handle
 
            // the destruction of the component. Note that all users of the
 
            // component do an increment-followed-by-decrement, we can simply
 
            // do a `fetch_sub`.
 
            let old_num_users = target.num_users.fetch_sub(1, Ordering::SeqCst);
 
            if old_num_users == 1 {
 
                // We're the one that got the counter to 0, so we're the ones
 
                // that are supposed to handle component exit
 
                self.finish_component_destruction(target_id);
 
            }
 

	
 
            return false;
 
        }
 

	
 
        // The generation is correct, and since we incremented the `num_users`
 
        // counter we're now sure that we can send the message and it will be
 
        // handled by the receiver
 
        target.connector.public.inbox.insert_message(message);
 

	
 
        // Finally, do the same as above: decrement number of users, if at gets
 
        // to 0 we're the ones who should handle the exit condition.
 
        let old_num_users = target.num_users.fetch_sub(1, Ordering::SeqCst);
 
        if old_num_users == 1 {
 
            // We're allowed to destroy the component.
 
            self.finish_component_destruction(target_id);
 
        } else {
 
            // Message is sent. If the component is sleeping, then we're sure
 
            // it is not scheduled and it has not initiated the destruction of
 
            // the component (because of the way
 
            // `initiate_component_destruction` does not set sleeping to true).
 
            // So we can safely schedule it.
 
            let should_wake_up = target.connector.public.sleeping
 
                .compare_exchange(true, false, Ordering::SeqCst, Ordering::Acquire)
 
                .is_ok();
 

	
 
            if should_wake_up {
 
                let key = unsafe{ ConnectorKey::from_id(target_id) };
 
                self.push_work(key);
 
            }
 
        }
 

	
 
        return true
 
    }
 

	
 
    /// Sends a message to a particular component, assumed to occur over a port.
 
    /// If the component happened to be sleeping then it will be scheduled for
 
    /// execution. Because of the port management system we may assumed that
 
    /// we're always accessing the component at the right generation number.
 
    pub(crate) fn send_message_assumed_alive(&self, target_id: ConnectorId, message: Message) {
 
        let target = {
 
            let lock = self.connectors.read().unwrap();
 
            let entry = lock.get(target_id.index);
 
            debug_assert_eq!(entry.generation.load(Ordering::Acquire), target_id.generation);
 
            &mut entry.connector.public
 
        };
 

	
 
        target.inbox.insert_message(message);
 

	
 
        let should_wake_up = target.sleeping
 
            .compare_exchange(true, false, Ordering::SeqCst, Ordering::Acquire)
 
            .is_ok();
 

	
 
        if should_wake_up {
 
            let key = unsafe{ ConnectorKey::from_id(target_id) };
 
            self.push_work(key);
 
        }
 
    }
 

	
 
    // --- Creating/retrieving/destroying components
 

	
 
    /// Creates an initially sleeping application connector.
 
    fn create_interface_component(&self, component: ConnectorApplication) -> ConnectorKey {
 
        // Initialize as sleeping, as it will be scheduled by the programmer.
 
        let mut lock = self.connectors.write().unwrap();
 
        let key = lock.create(ConnectorVariant::Native(Box::new(component)), true);
 

	
 
        self.increment_active_components();
 
        return key;
 
    }
 

	
 
    /// Creates a new PDL component. This function just creates the component.
 
    /// If you create it initially awake, then you must add it to the work
 
    /// queue. Other aspects of correctness (i.e. setting initial ports) are
 
    /// relinquished to the caller!
 
    pub(crate) fn create_pdl_component(&self, connector: ConnectorPDL, initially_sleeping: bool) -> ConnectorKey {
 
        // Create as not sleeping, as we'll schedule it immediately
 
        let key = {
 
            let mut lock = self.connectors.write().unwrap();
 
            lock.create(ConnectorVariant::UserDefined(connector), initially_sleeping)
 
        };
 

	
 
        self.increment_active_components();
 
        return key;
 
    }
 

	
 
    /// Retrieve private access to the component through its key.
 
    #[inline]
 
    pub(crate) fn get_component_private(&self, connector_key: &ConnectorKey) -> &'static mut ScheduledConnector {
 
        let entry = {
 
            let lock = self.connectors.read().unwrap();
 
            lock.get(connector_key.index)
 
        };
 

	
 
        debug_assert_eq!(entry.generation.load(Ordering::Acquire), connector_key.generation, "private access to {:?}", connector_key);
 
        return &mut entry.connector;
 
    }
 

	
 
    // --- Managing component destruction
 

	
 
    /// Start component destruction, may only be done by the scheduler that is
 
    /// executing the component. This might not actually destroy the component,
 
    /// since other components might be sending it messages.
 
    fn initiate_component_destruction(&self, connector_key: ConnectorKey) {
 
        // Most of the time no-one will be sending messages, so try
 
        // immediate destruction
 
        let mut lock = self.connectors.write().unwrap();
 
        let entry = lock.get(connector_key.index);
 
        debug_assert_eq!(entry.generation.load(Ordering::Acquire), connector_key.generation);
 
        debug_assert_eq!(entry.connector.public.sleeping.load(Ordering::Acquire), false); // not sleeping: caller is executing this component
 
        let old_num_users = entry.num_users.fetch_sub(1, Ordering::SeqCst);
 
        if old_num_users == 1 {
 
            // We just brought the number of users down to 0. Destroy the
 
            // component
 
            entry.connector.public.inbox.clear();
 
            entry.generation.fetch_add(1, Ordering::SeqCst);
 
            lock.destroy(connector_key);
 
            self.decrement_active_components();
 
        }
 
    }
 

	
 
    fn finish_component_destruction(&self, connector_id: ConnectorId) {
 
        let mut lock = self.connectors.write().unwrap();
 
        let entry = lock.get(connector_id.index);
 
        debug_assert_eq!(entry.num_users.load(Ordering::Acquire), 0);
 
        let _old_generation = entry.generation.fetch_add(1, Ordering::SeqCst);
 
        debug_assert_eq!(_old_generation, connector_id.generation);
 

	
 
        // TODO: In the future we should not only clear out the inbox, but send
 
        //  messages back to the senders indicating the messages did not arrive.
 
        entry.connector.public.inbox.clear();
 

	
 
        // Invariant of only one thread being able to handle the internals of
 
        // component is preserved by the fact that only one thread can decrement
 
        // `num_users` to 0.
 
        lock.destroy(unsafe{ ConnectorKey::from_id(connector_id) });
 
        self.decrement_active_components();
 
    }
 

	
 
    // --- Managing exit condition
 

	
 
    #[inline]
 
    pub(crate) fn increment_active_interfaces(&self) {
 
        let _old_num = self.active_interfaces.fetch_add(1, Ordering::SeqCst);
 
        debug_assert_ne!(_old_num, 0); // once it hits 0, it stays zero
 
    }
 

	
 
    pub(crate) fn decrement_active_interfaces(&self) {
 
        let old_num = self.active_interfaces.fetch_sub(1, Ordering::SeqCst);
 
        debug_assert!(old_num > 0);
 
        if old_num == 1 { // such that active interfaces is now 0
 
            let num_connectors = self.active_connectors.load(Ordering::Acquire);
 
            if num_connectors == 0 {
 
                self.signal_for_shutdown();
 
            }
 
        }
 
    }
 

	
 
    #[inline]
 
    fn increment_active_components(&self) {
 
        let _old_num = self.active_connectors.fetch_add(1, Ordering::SeqCst);
 
    }
 

	
 
    fn decrement_active_components(&self) {
 
        let old_num = self.active_connectors.fetch_sub(1, Ordering::SeqCst);
 
        debug_assert!(old_num > 0);
 
        if old_num == 1 { // such that we have no more active connectors (for now!)
 
            let num_interfaces = self.active_interfaces.load(Ordering::Acquire);
 
            if num_interfaces == 0 {
 
                self.signal_for_shutdown();
 
            }
 
        }
 
    }
 

	
 
    #[inline]
 
    fn signal_for_shutdown(&self) {
 
        debug_assert_eq!(self.active_interfaces.load(Ordering::Acquire), 0);
 
        debug_assert_eq!(self.active_connectors.load(Ordering::Acquire), 0);
 

	
 
        let _lock = self.connector_queue.lock().unwrap();
 
        let should_signal = self.should_exit
 
            .compare_exchange(false, true, Ordering::SeqCst, Ordering::Acquire)
 
            .is_ok();
 

	
 
        if should_signal {
 
            self.scheduler_notifier.notify_all();
 
        }
 
    }
 
}
 

	
 
// TODO: Come back to this at some point
 
unsafe impl Send for RuntimeInner {}
 
unsafe impl Sync for RuntimeInner {}
 

	
 
// -----------------------------------------------------------------------------
 
// ConnectorStore
 
// -----------------------------------------------------------------------------
 

	
 
struct StoreEntry {
 
    connector: ScheduledConnector,
 
    generation: std::sync::atomic::AtomicU32,
 
    num_users: std::sync::atomic::AtomicU32,
 
}
 

	
 
struct ConnectorStore {
 
    // Freelist storage of connectors. Storage should be pointer-stable as
 
    // someone might be mutating the vector while we're executing one of the
 
    // connectors.
 
    entries: RawVec<*mut StoreEntry>,
 
    free: Vec<usize>,
 
}
 

	
 
impl ConnectorStore {
 
    fn with_capacity(capacity: usize) -> Self {
 
        Self {
 
            entries: RawVec::with_capacity(capacity),
 
            free: Vec::with_capacity(capacity),
 
        }
 
    }
 

	
 
    /// Directly retrieves an entry. There be dragons here. The `connector`
 
    /// might have its destructor already executed. Accessing it might then lead
 
    /// to memory corruption.
 
    fn get(&self, index: u32) -> &'static mut StoreEntry {
 
        unsafe {
 
            let entry = self.entries.get_mut(index as usize);
 
            return &mut **entry;
 
        }
 
    }
 

	
 
    /// Creates a new connector. Caller should ensure ports are set up correctly
 
    /// and the connector is queued for execution if needed.
 
    fn create(&mut self, connector: ConnectorVariant, initially_sleeping: bool) -> ConnectorKey {
 
        let mut connector = ScheduledConnector {
 
            connector,
 
            ctx: ComponentCtx::new_empty(),
 
            public: ConnectorPublic::new(initially_sleeping),
 
            router: ControlMessageHandler::new(),
 
            shutting_down: false,
 
        };
 

	
 
        let index;
 
        let key;
 

	
 
        if self.free.is_empty() {
 
            // No free entries, allocate new entry
 
            index = self.entries.len();
 
            key = ConnectorKey{
 
                index: index as u32, generation: 0
 
            };
 
            connector.ctx.id = key.downcast();
 

	
 
            let connector = Box::into_raw(Box::new(StoreEntry{
 
                connector,
 
                generation: AtomicU32::new(0),
 
                num_users: AtomicU32::new(1),
 
            }));
 
            self.entries.push(connector);
 
        } else {
 
            // Free spot available
 
            index = self.free.pop().unwrap();
 

	
 
            unsafe {
 
                let target = &mut **self.entries.get_mut(index);
 
                std::ptr::write(&mut target.connector as *mut _, connector);
 
                let _old_num_users = target.num_users.fetch_add(1, Ordering::SeqCst);
 
                debug_assert_eq!(_old_num_users, 0);
 

	
 
                let generation = target.generation.load(Ordering::Acquire);
 
                key = ConnectorKey{ index: index as u32, generation };
 
                target.connector.ctx.id = key.downcast();
 
            }
 
        }
 

	
 
        println!("DEBUG [ global store  ] Created component at {}", key.index);
 
        return key;
 
    }
 

	
 
    /// Destroys a connector. Caller should make sure it is not scheduled for
 
    /// execution. Otherwise one experiences "bad stuff" (tm).
 
    fn destroy(&mut self, key: ConnectorKey) {
 
        unsafe {
 
            let target = self.entries.get_mut(key.index as usize);
 
            (**target).generation.fetch_add(1, Ordering::SeqCst);
 
            std::ptr::drop_in_place(*target);
 
            // Note: but not deallocating!
 
        }
 

	
 
        println!("DEBUG [ global store  ] Destroyed component at {}", key.index);
 
        self.free.push(key.index as usize);
 
    }
 
}
 

	
 
impl Drop for ConnectorStore {
 
    fn drop(&mut self) {
 
        // Everything in the freelist already had its destructor called, so only
 
        // has to be deallocated
 
        for free_idx in self.free.iter().copied() {
 
            unsafe {
 
                let memory = self.entries.get_mut(free_idx);
 
                let layout = std::alloc::Layout::for_value(&**memory);
 
                std::alloc::dealloc(*memory as *mut u8, layout);
 

	
 
                // mark as null for the remainder
 
                *memory = std::ptr::null_mut();
 
            }
 
        }
 

	
 
        // With the deallocated stuff marked as null, clear the remainder that
 
        // is not null
 
        for idx in 0..self.entries.len() {
 
            unsafe {
 
                let memory = *self.entries.get_mut(idx);
 
                if !memory.is_null() {
 
                    let _ = Box::from_raw(memory); // take care of deallocation, bit dirty, but meh
 
                }
 
            }
 
        }
 
    }
 
}
 
\ No newline at end of file
src/runtime2/native.rs
Show inline comments
 
@@ -82,479 +82,478 @@ impl Connector for ConnectorApplication {
 
            }
 

	
 
            return scheduling;
 
        } else {
 
            return self.run_in_deterministic_mode(sched_ctx, comp_ctx);
 
        }
 
    }
 
}
 

	
 
impl ConnectorApplication {
 
    pub(crate) fn new(runtime: Arc<RuntimeInner>) -> (Self, ApplicationInterface) {
 
        let sync_done = Arc::new(( Mutex::new(None), Condvar::new() ));
 
        let job_queue = Arc::new(Mutex::new(VecDeque::with_capacity(32)));
 

	
 
        let connector = ConnectorApplication {
 
            sync_done: sync_done.clone(),
 
            job_queue: job_queue.clone(),
 
            is_in_sync: false,
 
            sync_desc: Vec::new(),
 
            tree: FakeTree::new(),
 
            consensus: Consensus::new(),
 
            last_finished_handled: None,
 
            branch_extra: vec![0],
 
        };
 
        let interface = ApplicationInterface::new(sync_done, job_queue, runtime);
 

	
 
        return (connector, interface);
 
    }
 

	
 
    fn handle_new_messages(&mut self, comp_ctx: &mut ComponentCtx) {
 
        while let Some(ticket) = comp_ctx.get_next_message_ticket() {
 
            let message = comp_ctx.read_message_using_ticket(ticket);
 
            if let Message::Data(_) = message {
 
                self.handle_new_data_message(ticket, comp_ctx)
 
            } else {
 
                match comp_ctx.take_message_using_ticket(ticket) {
 
                    Message::Data(message) => unreachable!(),
 
                    Message::SyncComp(message) => self.handle_new_sync_comp_message(message, comp_ctx),
 
                    Message::SyncPort(message) => self.handle_new_sync_port_message(message, comp_ctx),
 
                    Message::SyncControl(message) => todo!("implement"),
 
                    Message::Control(_) => unreachable!("control message in native API component"),
 
                }
 
            }
 
        }
 
    }
 

	
 
    pub(crate) fn handle_new_data_message(&mut self, ticket: MessageTicket, ctx: &mut ComponentCtx) {
 
        // Go through all branches that are awaiting new messages and see if
 
        // there is one that can receive this message.
 
        if !self.consensus.handle_new_data_message(ticket, ctx) {
 
            // Old message, so drop it
 
            return;
 
        }
 

	
 
        let mut iter_id = self.tree.get_queue_first(QueueKind::AwaitingMessage);
 
        while let Some(branch_id) = iter_id {
 
            let message = ctx.read_message_using_ticket(ticket).as_data();
 
            iter_id = self.tree.get_queue_next(branch_id);
 

	
 
            let branch = &self.tree[branch_id];
 
            if branch.awaiting_port != message.data_header.target_port { continue; }
 
            if !self.consensus.branch_can_receive(branch_id, &message) { continue; }
 

	
 
            // This branch can receive, so fork and given it the message
 
            let receiving_branch_id = self.tree.fork_branch(branch_id);
 
            debug_assert!(receiving_branch_id.index as usize == self.branch_extra.len());
 
            self.branch_extra.push(self.branch_extra[branch_id.index as usize]); // copy instruction index
 
            self.consensus.notify_of_new_branch(branch_id, receiving_branch_id);
 
            let receiving_branch = &mut self.tree[receiving_branch_id];
 

	
 
            receiving_branch.insert_message(message.data_header.target_port, message.content.clone());
 
            self.consensus.notify_of_received_message(receiving_branch_id, &message, ctx);
 

	
 
            // And prepare the branch for running
 
            self.tree.push_into_queue(QueueKind::Runnable, receiving_branch_id);
 
        }
 
    }
 

	
 
    pub(crate) fn handle_new_sync_comp_message(&mut self, message: SyncCompMessage, ctx: &mut ComponentCtx) {
 
        if let Some(conclusion) = self.consensus.handle_new_sync_comp_message(message, ctx) {
 
            self.collapse_sync_to_conclusion(conclusion, ctx);
 
        }
 
    }
 

	
 
    pub(crate) fn handle_new_sync_port_message(&mut self, message: SyncPortMessage, ctx: &mut ComponentCtx) {
 
        self.consensus.handle_new_sync_port_message(message, ctx);
 
    }
 

	
 
    fn run_in_sync_mode(&mut self, _sched_ctx: SchedulerCtx, comp_ctx: &mut ComponentCtx) -> ConnectorScheduling {
 
        debug_assert!(self.is_in_sync);
 

	
 
        self.handle_new_messages(comp_ctx);
 

	
 
        let branch_id = self.tree.pop_from_queue(QueueKind::Runnable);
 
        if branch_id.is_none() {
 
            return ConnectorScheduling::NotNow;
 
        }
 

	
 
        let branch_id = branch_id.unwrap();
 
        let branch = &mut self.tree[branch_id];
 
        let mut instruction_idx = self.branch_extra[branch_id.index as usize];
 

	
 
        if instruction_idx >= self.sync_desc.len() {
 
            // Performed last instruction, so this branch is officially at the
 
            // end of the synchronous interaction.
 
            let consistency = self.consensus.notify_of_finished_branch(branch_id);
 
            if consistency == Consistency::Valid {
 
                branch.sync_state = SpeculativeState::ReachedSyncEnd;
 
                self.tree.push_into_queue(QueueKind::FinishedSync, branch_id);
 
            } else {
 
                branch.sync_state = SpeculativeState::Inconsistent;
 
            }
 
        } else {
 
            // We still have instructions to perform
 
            let cur_instruction = &self.sync_desc[instruction_idx];
 
            self.branch_extra[branch_id.index as usize] += 1;
 

	
 
            match &cur_instruction {
 
                ApplicationSyncAction::Put(port_id, content) => {
 
                    let port_id = *port_id;
 

	
 
                    let (sync_header, data_header) = self.consensus.handle_message_to_send(branch_id, port_id, &content, comp_ctx);
 
                    let message = Message::Data(DataMessage {
 
                        sync_header,
 
                        data_header,
 
                        content: content.clone(),
 
                    });
 
                    comp_ctx.submit_message(message);
 
                    self.tree.push_into_queue(QueueKind::Runnable, branch_id);
 
                    return ConnectorScheduling::Immediate;
 
                },
 
                ApplicationSyncAction::Get(port_id) => {
 
                    let port_id = *port_id;
 

	
 
                    branch.sync_state = SpeculativeState::HaltedAtBranchPoint;
 
                    branch.awaiting_port = port_id;
 
                    self.tree.push_into_queue(QueueKind::AwaitingMessage, branch_id);
 

	
 
                    let mut any_message_received = false;
 
                    for message in comp_ctx.get_read_data_messages(port_id) {
 
                        if self.consensus.branch_can_receive(branch_id, &message) {
 
                            // This branch can receive the message, so we do the
 
                            // fork-and-receive dance
 
                            let receiving_branch_id = self.tree.fork_branch(branch_id);
 
                            let branch = &mut self.tree[receiving_branch_id];
 
                            debug_assert!(receiving_branch_id.index as usize == self.branch_extra.len());
 
                            self.branch_extra.push(instruction_idx + 1);
 

	
 
                            branch.insert_message(port_id, message.content.clone());
 

	
 
                            self.consensus.notify_of_new_branch(branch_id, receiving_branch_id);
 
                            self.consensus.notify_of_received_message(receiving_branch_id, &message, comp_ctx);
 
                            self.tree.push_into_queue(QueueKind::Runnable, receiving_branch_id);
 

	
 
                            any_message_received = true;
 
                        }
 
                    }
 

	
 
                    if any_message_received {
 
                        return ConnectorScheduling::Immediate;
 
                    }
 
                }
 
            }
 
        }
 

	
 
        if self.tree.queue_is_empty(QueueKind::Runnable) {
 
            return ConnectorScheduling::NotNow;
 
        } else {
 
            return ConnectorScheduling::Later;
 
        }
 
    }
 

	
 
    fn run_in_deterministic_mode(&mut self, _sched_ctx: SchedulerCtx, comp_ctx: &mut ComponentCtx) -> ConnectorScheduling {
 
        debug_assert!(!self.is_in_sync);
 

	
 
        // In non-sync mode the application component doesn't really do anything
 
        // except performing jobs submitted from the API. This is the only
 
        // case where we expect to be woken up.
 
        // Note that we have to communicate to the scheduler when we've received
 
        // ports or created components (hence: given away ports) *before* we
 
        // enter a sync round.
 
        let mut queue = self.job_queue.lock().unwrap();
 
        while let Some(job) = queue.pop_front() {
 
            match job {
 
                ApplicationJob::NewChannel((endpoint_a, endpoint_b)) => {
 
                    comp_ctx.push_port(endpoint_a);
 
                    comp_ctx.push_port(endpoint_b);
 

	
 
                    return ConnectorScheduling::Immediate;
 
                }
 
                ApplicationJob::NewConnector(connector, initial_ports) => {
 
                    comp_ctx.push_component(connector, initial_ports);
 

	
 
                    return ConnectorScheduling::Later;
 
                },
 
                ApplicationJob::SyncRound(mut description) => {
 
                    // Entering sync mode
 
                    comp_ctx.notify_sync_start();
 
                    self.sync_desc = description;
 
                    self.is_in_sync = true;
 
                    debug_assert!(self.last_finished_handled.is_none());
 
                    debug_assert!(self.branch_extra.len() == 1);
 

	
 
                    let first_branch_id = self.tree.start_sync();
 
                    self.tree.push_into_queue(QueueKind::Runnable, first_branch_id);
 
                    debug_assert!(first_branch_id.index == 1);
 
                    self.consensus.start_sync(comp_ctx);
 
                    self.consensus.notify_of_new_branch(BranchId::new_invalid(), first_branch_id);
 
                    self.branch_extra.push(0); // set first branch to first instruction
 

	
 
                    return ConnectorScheduling::Immediate;
 
                },
 
                ApplicationJob::Shutdown => {
 
                    debug_assert!(queue.is_empty());
 

	
 
                    return ConnectorScheduling::Exit;
 
                }
 
            }
 
        }
 

	
 
        // Queue was empty
 
        return ConnectorScheduling::NotNow;
 
    }
 

	
 
    fn collapse_sync_to_conclusion(&mut self, conclusion: RoundConclusion, comp_ctx: &mut ComponentCtx) {
 
        // Notifying tree, consensus algorithm and context of ending sync
 
        let mut fake_vec = Vec::new();
 

	
 
        let (branch_id, success) = match conclusion {
 
            RoundConclusion::Success(branch_id) => {
 
                debug_assert!(self.branch_extra[branch_id.index as usize] >= self.sync_desc.len()); // finished program provided by API
 
                (branch_id, true)
 
            },
 
            RoundConclusion::Failure => (BranchId::new_invalid(), false),
 
        };
 

	
 
        let mut solution_branch = self.tree.end_sync(branch_id);
 
        self.consensus.end_sync(branch_id, &mut fake_vec);
 
        debug_assert!(fake_vec.is_empty());
 

	
 
        comp_ctx.notify_sync_end(&[]);
 

	
 
        // Turning hashmapped inbox into vector of values
 
        let mut inbox = Vec::with_capacity(solution_branch.inbox.len());
 
        for action in &self.sync_desc {
 
            match action {
 
                ApplicationSyncAction::Put(_, _) => {},
 
                ApplicationSyncAction::Get(port_id) => {
 
                    debug_assert!(solution_branch.inbox.contains_key(port_id));
 
                    inbox.push(solution_branch.inbox.remove(port_id).unwrap());
 
                },
 
            }
 
        }
 

	
 
        // Notifying interface of ending sync
 
        self.is_in_sync = false;
 
        self.sync_desc.clear();
 
        self.branch_extra.truncate(1);
 
        self.last_finished_handled = None;
 

	
 
        let (results, notification) = &*self.sync_done;
 
        let mut results = results.lock().unwrap();
 
        *results = Some(FinishedSync{ success, inbox });
 
        notification.notify_one();
 
    }
 
}
 

	
 
// -----------------------------------------------------------------------------
 
// ApplicationInterface
 
// -----------------------------------------------------------------------------
 

	
 
#[derive(Debug)]
 
pub enum ChannelCreationError {
 
    InSync,
 
}
 

	
 
#[derive(Debug)]
 
pub enum ApplicationStartSyncError {
 
    AlreadyInSync,
 
    NoSyncActions,
 
    IncorrectPortKind,
 
    UnownedPort,
 
}
 

	
 
#[derive(Debug)]
 
pub enum ApplicationEndSyncError {
 
    NotInSync,
 
    Failure,
 
}
 

	
 
pub enum ApplicationSyncAction {
 
    Put(PortIdLocal, ValueGroup),
 
    Get(PortIdLocal),
 
}
 

	
 
/// The interface to a `ApplicationConnector`. This allows setting up the
 
/// interactions the `ApplicationConnector` performs within a synchronous round.
 
pub struct ApplicationInterface {
 
    sync_done: SyncDone,
 
    job_queue: JobQueue,
 
    runtime: Arc<RuntimeInner>,
 
    is_in_sync: bool,
 
    connector_id: ConnectorId,
 
    owned_ports: Vec<(PortKind, PortIdLocal)>,
 
}
 

	
 
impl ApplicationInterface {
 
    fn new(sync_done: SyncDone, job_queue: JobQueue, runtime: Arc<RuntimeInner>) -> Self {
 
        return Self{
 
            sync_done, job_queue, runtime,
 
            is_in_sync: false,
 
            connector_id: ConnectorId::new_invalid(),
 
            owned_ports: Vec::new(),
 
        }
 
    }
 

	
 
    /// Creates a new channel. Can only fail if the application interface is
 
    /// currently in sync mode.
 
    pub fn create_channel(&mut self) -> Result<Channel, ChannelCreationError> {
 
        if self.is_in_sync {
 
            return Err(ChannelCreationError::InSync);
 
        }
 

	
 
        let (getter_port, putter_port) = self.runtime.create_channel(self.connector_id);
 
        debug_assert_eq!(getter_port.kind, PortKind::Getter);
 
        let getter_id = getter_port.self_id;
 
        let putter_id = putter_port.self_id;
 

	
 
        {
 
            let mut lock = self.job_queue.lock().unwrap();
 
            lock.push_back(ApplicationJob::NewChannel((getter_port, putter_port)));
 
        }
 

	
 
        // Add to owned ports for error checking while creating a connector
 
        self.owned_ports.reserve(2);
 
        self.owned_ports.push((PortKind::Putter, putter_id));
 
        self.owned_ports.push((PortKind::Getter, getter_id));
 

	
 
        return Ok(Channel{ putter_id, getter_id });
 
    }
 

	
 
    /// Creates a new connector. Note that it is not scheduled immediately, but
 
    /// depends on the `ApplicationConnector` to run, followed by the created
 
    /// connector being scheduled.
 
    pub fn create_connector(&mut self, module: &str, routine: &str, arguments: ValueGroup) -> Result<(), ComponentCreationError> {
 
        if self.is_in_sync {
 
            return Err(ComponentCreationError::InSync);
 
        }
 

	
 
        // Retrieve ports and make sure that we own the ones that are currently
 
        // specified. This is also checked by the scheduler, but that is done
 
        // asynchronously.
 
        let mut initial_ports = Vec::new();
 
        find_ports_in_value_group(&arguments, &mut initial_ports);
 
        for initial_port in &initial_ports {
 
            if !self.owned_ports.iter().any(|(_, v)| v == initial_port) {
 
                return Err(ComponentCreationError::UnownedPort);
 
            }
 
        }
 

	
 
        // We own all ports, so remove them on this side
 
        for initial_port in &initial_ports {
 
            let position = self.owned_ports.iter().position(|(_, v)| v == initial_port).unwrap();
 
            self.owned_ports.remove(position);
 
        }
 

	
 
        let prompt = self.runtime.protocol_description.new_component_v2(module.as_bytes(), routine.as_bytes(), arguments)?;
 
        let connector = ConnectorPDL::new(prompt);
 

	
 
        // Put on job queue
 
        {
 
            let mut queue = self.job_queue.lock().unwrap();
 
            queue.push_back(ApplicationJob::NewConnector(connector, initial_ports));
 
        }
 

	
 
        self.wake_up_connector_with_ping();
 

	
 
        return Ok(());
 
    }
 

	
 
    /// Queues up a description of a synchronous round to run. Will not actually
 
    /// run the synchronous behaviour in blocking fashion. The results *must* be
 
    /// retrieved using `try_wait` or `wait` for the interface to be considered
 
    /// in non-sync mode.
 
    // TODO: Maybe change API in the future. For now it does the job
 
    pub fn perform_sync_round(&mut self, actions: Vec<ApplicationSyncAction>) -> Result<(), ApplicationStartSyncError> {
 
        if self.is_in_sync {
 
            return Err(ApplicationStartSyncError::AlreadyInSync);
 
        }
 

	
 
        // Check the action ports for consistency
 
        for action in &actions {
 
            let (port_id, expected_kind) = match action {
 
                ApplicationSyncAction::Put(port_id, _) => (*port_id, PortKind::Putter),
 
                ApplicationSyncAction::Get(port_id) => (*port_id, PortKind::Getter),
 
            };
 

	
 
            match self.find_port_by_id(port_id) {
 
                Some(port_kind) => {
 
                    if port_kind != expected_kind {
 
                        return Err(ApplicationStartSyncError::IncorrectPortKind)
 
                    }
 
                },
 
                None => {
 
                    return Err(ApplicationStartSyncError::UnownedPort);
 
                }
 
            }
 
        }
 

	
 
        // Everything is consistent, go into sync mode and send the actions off
 
        // to the component that will actually perform the sync round
 
        self.is_in_sync = true;
 
        {
 
            let (is_done, _) = &*self.sync_done;
 
            let mut lock = is_done.lock().unwrap();
 
            *lock = None;
 
        }
 

	
 
        {
 
            let mut lock = self.job_queue.lock().unwrap();
 
            lock.push_back(ApplicationJob::SyncRound(actions));
 
        }
 

	
 
        self.wake_up_connector_with_ping();
 
        return Ok(())
 
    }
 

	
 
    /// Wait until the next sync-round is finished, returning the received
 
    /// messages in order of `get` calls.
 
    pub fn wait(&mut self) -> Result<Vec<ValueGroup>, ApplicationEndSyncError> {
 
        if !self.is_in_sync {
 
            return Err(ApplicationEndSyncError::NotInSync);
 
        }
 

	
 
        let (is_done, condition) = &*self.sync_done;
 
        let mut lock = is_done.lock().unwrap();
 
        lock = condition.wait_while(lock, |v| v.is_none()).unwrap(); // wait while not done
 

	
 
        self.is_in_sync = false;
 
        let result = lock.take().unwrap();
 
        if result.success {
 
            return Ok(result.inbox);
 
        } else {
 
            return Err(ApplicationEndSyncError::Failure);
 
        }
 
    }
 

	
 
    /// Called by runtime to set associated connector's ID.
 
    pub(crate) fn set_connector_id(&mut self, id: ConnectorId) {
 
        self.connector_id = id;
 
    }
 

	
 
    fn wake_up_connector_with_ping(&self) {
 
        let message = ControlMessage {
 
            id: 0,
 
            sending_component_id: self.connector_id,
 
            content: ControlContent::Ping,
 
        };
 
        self.runtime.send_message_maybe_destroyed(self.connector_id, Message::Control(message));
 
    }
 

	
 
    fn find_port_by_id(&self, port_id: PortIdLocal) -> Option<PortKind> {
 
        return self.owned_ports.iter()
 
            .find(|(_, owned_id)| *owned_id == port_id)
 
            .map(|(port_kind, _)| *port_kind);
 
    }
 
}
 

	
 
impl Drop for ApplicationInterface {
 
    fn drop(&mut self) {
 
        {
 
            let mut lock = self.job_queue.lock().unwrap();
 
            lock.push_back(ApplicationJob::Shutdown);
 
        }
 

	
 
        self.wake_up_connector_with_ping();
 
        self.runtime.decrement_active_interfaces();
 
    }
 
}
 
\ No newline at end of file
src/runtime2/port.rs
Show inline comments
 
use super::ConnectorId;
 

	
 
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
 
pub struct PortIdLocal {
 
    pub index: u32,
 
}
 

	
 
impl PortIdLocal {
 
    pub fn new(id: u32) -> Self {
 
        Self{ index: id }
 
    }
 

	
 
    // TODO: Unsure about this, maybe remove, then also remove all struct
 
    //  instances where I call this
 
    pub fn new_invalid() -> Self {
 
        Self{ index: u32::MAX }
 
    }
 

	
 
    pub fn is_valid(&self) -> bool {
 
        return self.index != u32::MAX;
 
    }
 
}
 

	
 
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
 
pub struct ChannelId {
 
    pub index: u32,
 
}
 

	
 
impl ChannelId {
 
    pub fn new(id: u32) -> Self {
 
        return Self{ index: id };
 
    }
 
}
 

	
 
#[derive(Debug, Clone, Copy, Eq, PartialEq)]
 
pub enum PortKind {
 
    Putter,
 
    Getter,
 
}
 

	
 
#[derive(Debug, Clone, Copy, Eq, PartialEq)]
 
pub enum PortState {
 
    Open,
 
    Closed,
 
}
 

	
 
/// Represents a port inside of the runtime. This is generally the local view of
 
/// a connector on its port, which may not be consistent with the rest of the
 
/// global system (e.g. its peer was moved to a new connector, or the peer might
 
/// have died in the meantime, so it is no longer usable).
 
#[derive(Clone)]
 
pub struct Port {
 
    pub self_id: PortIdLocal,
 
    pub peer_id: PortIdLocal,
 
    pub channel_id: ChannelId,
 
    pub kind: PortKind,
 
    pub state: PortState,
 
    pub peer_connector: ConnectorId, // might be temporarily inconsistent while peer port is sent around in non-sync phase
 
}
 

	
 
// TODO: Turn port ID into its own type
 
pub struct Channel {
 
    pub putter_id: PortIdLocal, // can put on it, so from the connector's point of view, this is an output
 
    pub getter_id: PortIdLocal, // vice versa: can get on it, so an input for the connector
 
}
 
\ No newline at end of file
src/runtime2/scheduler.rs
Show inline comments
 
use std::collections::VecDeque;
 
use std::sync::Arc;
 
use std::sync::atomic::Ordering;
 

	
 
use crate::protocol::eval::EvalError;
 
use crate::runtime2::port::ChannelId;
 

	
 
use super::{ScheduledConnector, RuntimeInner, ConnectorId, ConnectorKey};
 
use super::port::{Port, PortState, PortIdLocal};
 
use super::native::Connector;
 
use super::branch::{BranchId};
 
use super::connector::{ConnectorPDL, ConnectorScheduling};
 
use super::inbox::{
 
    Message, DataMessage,
 
    ControlMessage, ControlContent,
 
    SyncControlMessage, SyncControlContent,
 
};
 

	
 
// Because it contains pointers we're going to do a copy by value on this one
 
#[derive(Clone, Copy)]
 
pub(crate) struct SchedulerCtx<'a> {
 
    pub(crate) runtime: &'a RuntimeInner
 
}
 

	
 
pub(crate) struct Scheduler {
 
    runtime: Arc<RuntimeInner>,
 
    scheduler_id: u32,
 
}
 

	
 
impl Scheduler {
 
    pub fn new(runtime: Arc<RuntimeInner>, scheduler_id: u32) -> Self {
 
        return Self{ runtime, scheduler_id };
 
    }
 

	
 
    pub fn run(&mut self) {
 
        // Setup global storage and workspaces that are reused for every
 
        // connector that we run
 
        'thread_loop: loop {
 
            // Retrieve a unit of work
 
            self.debug("Waiting for work");
 
            let connector_key = self.runtime.wait_for_work();
 
            if connector_key.is_none() {
 
                // We should exit
 
                self.debug(" ... No more work, quitting");
 
                break 'thread_loop;
 
            }
 

	
 
            // We have something to do
 
            let connector_key = connector_key.unwrap();
 
            let connector_id = connector_key.downcast();
 
            self.debug_conn(connector_id, &format!(" ... Got work, running {}", connector_key.index));
 

	
 
            let scheduled = self.runtime.get_component_private(&connector_key);
 

	
 
            // Keep running until we should no longer immediately schedule the
 
            // connector.
 
            let mut cur_schedule = ConnectorScheduling::Immediate;
 
            while let ConnectorScheduling::Immediate = cur_schedule {
 
                self.handle_inbox_messages(scheduled);
 

	
 
                // Run the main behaviour of the connector, depending on its
 
                // current state.
 
                if scheduled.shutting_down {
 
                    // Nothing to do. But we're stil waiting for all our pending
 
                    // control messages to be answered.
 
                    self.debug_conn(connector_id, &format!("Shutting down, {} Acks remaining", scheduled.router.num_pending_acks()));
 
                    if scheduled.router.num_pending_acks() == 0 {
 
                        // We're actually done, we can safely destroy the
 
                        // currently running connector
 
                        self.runtime.initiate_component_destruction(connector_key);
 
                        continue 'thread_loop;
 
                    } else {
 
                        cur_schedule = ConnectorScheduling::NotNow;
 
                    }
 
                } else {
 
                    self.debug_conn(connector_id, "Running ...");
 
                    let scheduler_ctx = SchedulerCtx{ runtime: &*self.runtime };
 
                    let new_schedule = scheduled.connector.run(scheduler_ctx, &mut scheduled.ctx);
 
                    self.debug_conn(connector_id, &format!("Finished running (new scheduling is {:?})", new_schedule));
 

	
 
                    // Handle all of the output from the current run: messages to
 
                    // send and connectors to instantiate.
 
                    self.handle_changes_in_context(scheduled);
 

	
 
                    cur_schedule = new_schedule;
 
                }
 
            }
 

	
 
            // If here then the connector does not require immediate execution.
 
            // So enqueue it if requested, and otherwise put it in a sleeping
 
            // state.
 
            match cur_schedule {
 
                ConnectorScheduling::Immediate => unreachable!(),
 
                ConnectorScheduling::Later => {
 
                    // Simply queue it again later
 
                    self.runtime.push_work(connector_key);
 
                },
 
                ConnectorScheduling::NotNow => {
 
                    // Need to sleep, note that we are the only ones which are
 
                    // allows to set the sleeping state to `true`, and since
 
                    // we're running it must currently be `false`.
 
                    self.try_go_to_sleep(connector_key, scheduled);
 
                },
 
                ConnectorScheduling::Exit => {
 
                    // Prepare for exit. Set the shutdown flag and broadcast
 
                    // messages to notify peers of closing channels
 
                    scheduled.shutting_down = true;
 
                    for port in &scheduled.ctx.ports {
 
                        if port.state != PortState::Closed {
 
                            let message = scheduled.router.prepare_closing_channel(
 
                                port.self_id, port.peer_id,
 
                                connector_id
 
                            );
 
                            self.debug_conn(connector_id, &format!("Sending message to {:?} [ exit ] \n --- {:?}", port.peer_connector, message));
 
                            self.runtime.send_message_assumed_alive(port.peer_connector, Message::Control(message));
 
                        }
 
                    }
 

	
 
                    // Any messages still in the public inbox should be handled
 
                    scheduled.ctx.inbox.clear_read_messages();
 
                    while let Some(ticket) = scheduled.ctx.get_next_message_ticket_even_if_not_in_sync() {
 
                        let message = scheduled.ctx.take_message_using_ticket(ticket);
 
                        self.handle_message_while_shutting_down(message, scheduled);
 
                    }
 

	
 
                    if scheduled.router.num_pending_acks() == 0 {
 
                        // All ports (if any) already closed
 
                        self.runtime.initiate_component_destruction(connector_key);
 
                        continue 'thread_loop;
 
                    }
 

	
 
                    self.try_go_to_sleep(connector_key, scheduled);
 
                },
 
            }
 
        }
 
    }
 

	
 
    /// Receiving messages from the public inbox and handling them or storing
 
    /// them in the component's private inbox
 
    fn handle_inbox_messages(&mut self, scheduled: &mut ScheduledConnector) {
 
        let connector_id = scheduled.ctx.id;
 

	
 
        while let Some(message) = scheduled.public.inbox.take_message() {
 
            // Check if the message has to be rerouted because we have moved the
 
            // target port to another component.
 
            self.debug_conn(connector_id, &format!("Handling message\n --- {:#?}", message));
 
            if let Some(target_port) = message.target_port() {
 
                if let Some(other_component_id) = scheduled.router.should_reroute(target_port) {
 
                    self.debug_conn(connector_id, " ... Rerouting the message");
 

	
 
                    // We insert directly into the private inbox. Since we have
 
                    // a reroute entry the component can not yet be running.
 
                    if let Message::Control(_) = &message {
 
                        self.runtime.send_message_assumed_alive(other_component_id, message);
 
                    } else {
 
                        let key = unsafe { ConnectorKey::from_id(other_component_id) };
 
                        let component = self.runtime.get_component_private(&key);
 
                        component.ctx.inbox.insert_new(message);
 
                    }
 

	
 
                    continue;
 
                }
 

	
 
                match scheduled.ctx.get_port_by_id(target_port) {
 
                    Some(port_info) => {
 
                        if port_info.state == PortState::Closed {
 
                            // We're no longer supposed to receive messages
 
                            // (rerouted message arrived much later!)
 
                            continue
 
                        }
 
                    },
 
                    None => {
 
                        // Apparently we no longer have a handle to the port
 
                        continue;
 
                    }
 
                }
 
            }
 

	
 
            // If here, then we should handle the message
 
            self.debug_conn(connector_id, " ... Handling the message");
 
            if let Message::Control(message) = &message {
 
                match message.content {
 
                    ControlContent::PortPeerChanged(port_id, new_target_connector_id) => {
 
                        // Need to change port target
 
                        let port = scheduled.ctx.get_port_mut_by_id(port_id).unwrap();
 
                        port.peer_connector = new_target_connector_id;
 

	
 
                        // Note: for simplicity we program the scheduler to always finish
 
                        // running a connector with an empty outbox. If this ever changes
 
                        // then accepting the "port peer changed" message implies we need
 
                        // to change the recipient of the message in the outbox.
 
                        debug_assert!(scheduled.ctx.outbox.is_empty());
 

	
 
                        // And respond with an Ack
 
                        let ack_message = Message::Control(ControlMessage {
 
                            id: message.id,
 
                            sending_component_id: connector_id,
 
                            content: ControlContent::Ack,
 
                        });
 
                        self.debug_conn(connector_id, &format!("Sending message to {:?} [pp ack]\n --- {:?}", message.sending_component_id, ack_message));
 
                        self.runtime.send_message_assumed_alive(message.sending_component_id, ack_message);
 
                    },
 
                    ControlContent::CloseChannel(port_id) => {
 
                        // Mark the port as being closed
 
                        let port = scheduled.ctx.get_port_mut_by_id(port_id).unwrap();
 
                        port.state = PortState::Closed;
 

	
 
                        // Send an Ack
 
                        let ack_message = Message::Control(ControlMessage {
 
                            id: message.id,
 
                            sending_component_id: connector_id,
 
                            content: ControlContent::Ack,
 
                        });
 
                        self.debug_conn(connector_id, &format!("Sending message to {:?} [cc ack] \n --- {:?}", message.sending_component_id, ack_message));
 
                        self.runtime.send_message_assumed_alive(message.sending_component_id, ack_message);
 
                    },
 
                    ControlContent::Ack => {
 
                        if let Some(component_key) = scheduled.router.handle_ack(message.id) {
 
                            self.runtime.push_work(component_key);
 
                        };
 
                    },
 
                    ControlContent::Ping => {},
 
                }
 
            } else {
 
                // Not a control message
 
                if scheduled.shutting_down {
 
                    // Since we're shutting down, we just want to respond with a
 
                    // message saying the message did not arrive.
 
                    debug_assert!(scheduled.ctx.inbox.get_next_message_ticket().is_none()); // public inbox should be completely cleared
 
                    self.handle_message_while_shutting_down(message, scheduled);
 
                } else {
 
                    scheduled.ctx.inbox.insert_new(message);
 
                }
 
            }
 
        }
 
    }
 

	
 
    fn handle_message_while_shutting_down(&mut self, message: Message, scheduled: &mut ScheduledConnector) {
 
        let target_port_and_round_number = match message {
 
            Message::Data(msg) => Some((msg.data_header.target_port, msg.sync_header.sync_round)),
 
            Message::SyncComp(_) => None,
 
            Message::SyncPort(msg) => Some((msg.target_port, msg.sync_header.sync_round)),
 
            Message::SyncControl(_) => None,
 
            Message::Control(_) => None,
 
        };
 

	
 
        if let Some((target_port, sync_round)) = target_port_and_round_number {
 
            // This message is aimed at a port, but we're shutting down, so
 
            // notify the peer that its was not received properly.
 
            // (also: since we're shutting down, we're not in sync mode and
 
            // the context contains the definitive set of owned ports)
 
            let port = scheduled.ctx.get_port_by_id(target_port).unwrap();
 
            if port.state == PortState::Open {
 
                let message = SyncControlMessage {
 
                    in_response_to_sync_round: sync_round,
 
                    target_component_id: port.peer_connector,
 
                    content: SyncControlContent::ChannelIsClosed(port.peer_id),
 
                };
 
                self.debug_conn(scheduled.ctx.id, &format!("Sending message to {:?} [shutdown]\n --- {:?}", port.peer_connector, message));
 
                self.runtime.send_message_assumed_alive(port.peer_connector, Message::SyncControl(message));
 
            }
 
        }
 
    }
 

	
 
    /// Handles changes to the context that were made by the component. This is
 
    /// the way (due to Rust's borrowing rules) that we bubble up changes in the
 
    /// component's state that the scheduler needs to know about (e.g. a message
 
    /// that the component wants to send, a port that has been added).
 
    fn handle_changes_in_context(&mut self, scheduled: &mut ScheduledConnector) {
 
        let connector_id = scheduled.ctx.id;
 

	
 
        // Handling any messages that were sent
 
        while let Some(message) = scheduled.ctx.outbox.pop_front() {
 
            let (target_component_id, over_port) = match &message {
 
                Message::Data(content) => {
 
                    // Data messages are always sent to a particular port, and
 
                    // may end up being rerouted.
 
                    let port_desc = scheduled.ctx.get_port_by_id(content.data_header.sending_port).unwrap();
 
                    debug_assert_eq!(port_desc.peer_id, content.data_header.target_port);
 

	
 
                    if port_desc.state == PortState::Closed {
 
                        todo!("handle sending over a closed port")
 
                    }
 
                    debug_assert_eq!(port_desc.state, PortState::Open); // checked when adding to context
 

	
 
                    (port_desc.peer_connector, true)
 
                },
 
                Message::SyncComp(content) => {
 
                    // Sync messages are always sent to a particular component,
 
                    // the sender must make sure it actually wants to send to
 
                    // the specified component (and is not using an inconsistent
 
                    // component ID associated with a port).
 
                    (content.target_component_id, false)
 
                },
 
                Message::SyncPort(content) => {
 
                    let port_desc = scheduled.ctx.get_port_by_id(content.source_port).unwrap();
 
                    debug_assert_eq!(port_desc.peer_id, content.target_port);
 
                    if port_desc.state == PortState::Closed {
 
                        todo!("handle sending over a closed port")
 
                    }
 
                    debug_assert_eq!(port_desc.state, PortState::Open); // checked when adding to context
 

	
 
                    (port_desc.peer_connector, true)
 
                },
 
                Message::SyncControl(_) => unreachable!("component sending 'SyncControl' messages directly"),
 
                Message::Control(_) => unreachable!("component sending 'Control' messages directly"),
 
            };
 

	
 
            self.debug_conn(connector_id, &format!("Sending message to {:?} [outbox, over port: {}] \n --- {:#?}", target_component_id, over_port, message));
 
            if over_port {
 
                self.runtime.send_message_assumed_alive(target_component_id, message);
 
            } else {
 
                self.runtime.send_message_maybe_destroyed(target_component_id, message);
 
            }
 
        }
 

	
 
        while let Some(state_change) = scheduled.ctx.state_changes.pop_front() {
 
            match state_change {
 
                ComponentStateChange::CreatedComponent(component, initial_ports) => {
 
                    // Creating a new component. Need to relinquish control of
 
                    // the ports.
 
                    let new_component_key = self.runtime.create_pdl_component(component, false);
 
                    let new_connector = self.runtime.get_component_private(&new_component_key);
 

	
 
                    // First pass: transfer ports and the associated messages,
 
                    // also count the number of ports that have peers
 
                    let mut num_peers = 0;
 
                    for port_id in initial_ports {
 
                        // Transfer messages associated with the transferred port
 
                        scheduled.ctx.inbox.transfer_messages_for_port(port_id, &mut new_connector.ctx.inbox);
 

	
 
                        // Transfer the port itself
 
                        let port_index = scheduled.ctx.ports.iter()
 
                            .position(|v| v.self_id == port_id)
 
                            .unwrap();
 
                        let port = scheduled.ctx.ports.remove(port_index);
 
                        new_connector.ctx.ports.push(port.clone());
 

	
 
                        if port.state == PortState::Open {
 
                            num_peers += 1;
 
                        }
 
                    }
 

	
 
                    if num_peers == 0 {
 
                        // No peers to notify, so just schedule the component
 
                        self.runtime.push_work(new_component_key);
 
                    } else {
 
                        // Some peers to notify
 
                        let new_component_id = new_component_key.downcast();
 
                        let control_id = scheduled.router.prepare_new_component(new_component_key);
 
                        for port in new_connector.ctx.ports.iter() {
 
                            if port.state == PortState::Closed {
 
                                continue;
 
                            }
 

	
 
                            let control_message = scheduled.router.prepare_changed_port_peer(
 
                                control_id, scheduled.ctx.id,
 
                                port.peer_connector, port.peer_id,
 
                                new_component_id, port.self_id
 
                            );
 
                            self.debug_conn(connector_id, &format!("Sending message to {:?} [newcom]\n --- {:#?}", port.peer_connector, control_message));
 
                            self.runtime.send_message_assumed_alive(port.peer_connector, Message::Control(control_message));
 
                        }
 
                    }
 
                },
 
                ComponentStateChange::CreatedPort(port) => {
 
                    scheduled.ctx.ports.push(port);
 
                },
 
                ComponentStateChange::ChangedPort(port_change) => {
 
                    if port_change.is_acquired {
 
                        scheduled.ctx.ports.push(port_change.port);
 
                    } else {
 
                        let index = scheduled.ctx.ports
 
                            .iter()
 
                            .position(|v| v.self_id == port_change.port.self_id)
 
                            .unwrap();
 
                        scheduled.ctx.ports.remove(index);
 
                    }
 
                }
 
            }
 
        }
 

	
 
        // Finally, check if we just entered or just left a sync region
 
        if scheduled.ctx.changed_in_sync {
 
            if scheduled.ctx.is_in_sync {
 
                // Just entered sync region
 
            } else {
 
                // Just left sync region. So prepare inbox for the next sync
 
                // round
 
                scheduled.ctx.inbox.clear_read_messages();
 
            }
 

	
 
            scheduled.ctx.changed_in_sync = false; // reset flag
 
        }
 
    }
 

	
 
    fn try_go_to_sleep(&self, connector_key: ConnectorKey, connector: &mut ScheduledConnector) {
 
        debug_assert_eq!(connector_key.index, connector.ctx.id.index);
 
        debug_assert_eq!(connector.public.sleeping.load(Ordering::Acquire), false);
 

	
 
        // This is the running connector, and only the running connector may
 
        // decide it wants to sleep again.
 
        connector.public.sleeping.store(true, Ordering::Release);
 

	
 
        // But due to reordering we might have received messages from peers who
 
        // did not consider us sleeping. If so, then we wake ourselves again.
 
        if !connector.public.inbox.is_empty() {
 
            // Try to wake ourselves up (needed because someone might be trying
 
            // the exact same atomic compare-and-swap at this point in time)
 
            let should_wake_up_again = connector.public.sleeping
 
                .compare_exchange(true, false, Ordering::SeqCst, Ordering::Acquire)
 
                .is_ok();
 

	
 
            if should_wake_up_again {
 
                self.runtime.push_work(connector_key)
 
            }
 
        }
 
    }
 

	
 
    // TODO: Remove, this is debugging stuff
 
    fn debug(&self, message: &str) {
 
        println!("DEBUG [thrd:{:02} conn:  ]: {}", self.scheduler_id, message);
 
    }
 

	
 
    fn debug_conn(&self, conn: ConnectorId, message: &str) {
 
        println!("DEBUG [thrd:{:02} conn:{:02}]: {}", self.scheduler_id, conn.index, message);
 
    }
 
}
 

	
 
// -----------------------------------------------------------------------------
 
// ComponentCtx
 
// -----------------------------------------------------------------------------
 

	
 
enum ComponentStateChange {
 
    CreatedComponent(ConnectorPDL, Vec<PortIdLocal>),
 
    CreatedPort(Port),
 
    ChangedPort(ComponentPortChange),
 
}
 

	
 
#[derive(Clone)]
 
pub(crate) struct ComponentPortChange {
 
    pub is_acquired: bool, // otherwise: released
 
    pub port: Port,
 
}
 

	
 
/// The component context (better name may be invented). This was created
 
/// because part of the component's state is managed by the scheduler, and part
 
/// of it by the component itself. When the component starts a sync block or
 
/// exits a sync block the partially managed state by both component and
 
/// scheduler need to be exchanged.
 
pub(crate) struct ComponentCtx {
 
    // Mostly managed by the scheduler
 
    pub(crate) id: ConnectorId,
 
    ports: Vec<Port>,
 
    inbox: Inbox,
 
    // Submitted by the component
 
    is_in_sync: bool,
 
    changed_in_sync: bool,
 
    outbox: VecDeque<Message>,
 
    state_changes: VecDeque<ComponentStateChange>,
 

	
 
    // Workspaces that may be used by components to (generally) prevent
 
    // allocations. Be a good scout and leave it empty after you've used it.
 
    // TODO: Move to scheduler ctx, this is the wrong place
 
    pub workspace_ports: Vec<PortIdLocal>,
 
    pub workspace_branches: Vec<BranchId>,
 
}
 

	
 
impl ComponentCtx {
 
    pub(crate) fn new_empty() -> Self {
 
        return Self{
 
            id: ConnectorId::new_invalid(),
 
            ports: Vec::new(),
 
            inbox: Inbox::new(),
 
            is_in_sync: false,
 
            changed_in_sync: false,
 
            outbox: VecDeque::new(),
 
            state_changes: VecDeque::new(),
 
            workspace_ports: Vec::new(),
 
            workspace_branches: Vec::new(),
 
        };
 
    }
 

	
 
    /// Notify the runtime that the component has created a new component. May
 
    /// only be called outside of a sync block.
 
    pub(crate) fn push_component(&mut self, component: ConnectorPDL, initial_ports: Vec<PortIdLocal>) {
 
        debug_assert!(!self.is_in_sync);
 
        self.state_changes.push_back(ComponentStateChange::CreatedComponent(component, initial_ports));
 
    }
 

	
 
    /// Notify the runtime that the component has created a new port. May only
 
    /// be called outside of a sync block (for ports received during a sync
 
    /// block, pass them when calling `notify_sync_end`).
 
    pub(crate) fn push_port(&mut self, port: Port) {
 
        debug_assert!(!self.is_in_sync);
 
        self.state_changes.push_back(ComponentStateChange::CreatedPort(port))
 
    }
 

	
 
    /// Notify the runtime of an error. Note that this will not perform any
 
    /// special action beyond printing the error. The component is responsible
 
    /// for waiting until it is appropriate to shut down (i.e. being outside
 
    /// of a sync region) and returning the `Exit` scheduling code.
 
    pub(crate) fn push_error(&mut self, error: EvalError) {
 
        println!("ERROR: Component ({}) encountered a critical error:\n{}", self.id.index, error);
 
    }
 

	
 
    #[inline]
 
    pub(crate) fn get_ports(&self) -> &[Port] {
 
        return self.ports.as_slice();
 
    }
 

	
 
    pub(crate) fn get_port_by_id(&self, id: PortIdLocal) -> Option<&Port> {
 
        return self.ports.iter().find(|v| v.self_id == id);
 
    }
 

	
 
    pub(crate) fn get_port_by_channel_id(&self, id: ChannelId) -> Option<&Port> {
 
        return self.ports.iter().find(|v| v.channel_id == id);
 
    }
 

	
 
    fn get_port_mut_by_id(&mut self, id: PortIdLocal) -> Option<&mut Port> {
 
        return self.ports.iter_mut().find(|v| v.self_id == id);
 
    }
 

	
 
    /// Notify that component will enter a sync block. Note that after calling
 
    /// this function you must allow the scheduler to pick up the changes in the
 
    /// context by exiting your code-executing loop, and to continue executing
 
    /// code the next time the scheduler picks up the component.
 
    pub(crate) fn notify_sync_start(&mut self) {
 
        debug_assert!(!self.is_in_sync);
 

	
 
        self.is_in_sync = true;
 
        self.changed_in_sync = true;
 
    }
 

	
 
    #[inline]
 
    pub(crate) fn is_in_sync(&self) -> bool {
 
        return self.is_in_sync;
 
    }
 

	
 
    /// Submit a message for the scheduler to send to the appropriate receiver.
 
    /// May only be called inside of a sync block.
 
    pub(crate) fn submit_message(&mut self, contents: Message) -> Result<(), ()> {
 
        debug_assert!(self.is_in_sync);
 
        if let Some(port_id) = contents.source_port() {
 
            let port_info = self.get_port_by_id(port_id);
 
            let is_valid = match port_info {
 
                Some(port_info) => {
 
                    port_info.state == PortState::Open
 
                },
 
                None => false,
 
            };
 
            if !is_valid {
 
                // We don't own the port
 
                println!(" ****** DEBUG ****** : Sending through closed port!!! {}", port_id.index);
 
                return Err(());
 
            }
 
        }
 

	
 
        self.outbox.push_back(contents);
 
        return Ok(());
 
    }
 

	
 
    /// Notify that component just finished a sync block. Like
 
    /// `notify_sync_start`: drop out of the `Component::Run` function.
 
    pub(crate) fn notify_sync_end(&mut self, changed_ports: &[ComponentPortChange]) {
 
        debug_assert!(self.is_in_sync);
 

	
 
        self.is_in_sync = false;
 
        self.changed_in_sync = true;
 

	
 
        self.state_changes.reserve(changed_ports.len());
 
        for changed_port in changed_ports {
 
            self.state_changes.push_back(ComponentStateChange::ChangedPort(changed_port.clone()));
 
        }
 
    }
 

	
 
    /// Retrieves messages matching a particular port and branch id. But only
 
    /// those messages that have been previously received with
 
    /// `read_next_message`.
 
    pub(crate) fn get_read_data_messages(&self, match_port_id: PortIdLocal) -> MessagesIter {
 
        return self.inbox.get_read_data_messages(match_port_id);
 
    }
 

	
 
    pub(crate) fn get_next_message_ticket(&mut self) -> Option<MessageTicket> {
 
        if !self.is_in_sync { return None; }
 
        return self.inbox.get_next_message_ticket();
 
    }
 

	
 
    #[inline]
 
    pub(crate) fn get_next_message_ticket_even_if_not_in_sync(&mut self) -> Option<MessageTicket> {
 
        return self.inbox.get_next_message_ticket();
 
    }
 

	
 
    #[inline]
 
    pub(crate) fn read_message_using_ticket(&self, ticket: MessageTicket) -> &Message {
 
        return self.inbox.read_message_using_ticket(ticket);
 
    }
 

	
 
    #[inline]
 
    pub(crate) fn take_message_using_ticket(&mut self, ticket: MessageTicket) -> Message {
 
        return self.inbox.take_message_using_ticket(ticket)
 
    }
 

	
 
    /// Puts back a message back into the inbox. The reason being that the
 
    /// message is actually part of the next sync round. This will
 
    pub(crate) fn put_back_message(&mut self, message: Message) {
 
        self.inbox.put_back_message(message);
 
    }
 
}
 

	
 
pub(crate) struct MessagesIter<'a> {
 
    messages: &'a [Message],
 
    next_index: usize,
 
    max_index: usize,
 
    match_port_id: PortIdLocal,
 
}
 

	
 
impl<'a> Iterator for MessagesIter<'a> {
 
    type Item = &'a DataMessage;
 

	
 
    fn next(&mut self) -> Option<Self::Item> {
 
        // Loop until match is found or at end of messages
 
        while self.next_index < self.max_index {
 
            let message = &self.messages[self.next_index];
 
            if let Message::Data(message) = &message {
 
                if message.data_header.target_port == self.match_port_id {
 
                    // Found a match
 
                    self.next_index += 1;
 
                    return Some(message);
 
                }
 
            } else {
 
                // Unreachable because:
 
                //  1. We only iterate over messages that were previously retrieved by `read_next_message`.
 
                //  2. Inbox does not contain control/ping messages.
 
                //  3. If `read_next_message` encounters anything else than a data message, it is removed from the inbox.
 
                unreachable!();
 
            }
 

	
 
            self.next_index += 1;
 
        }
 

	
 
        // No more messages
 
        return None;
 
    }
 
}
 

	
 
// -----------------------------------------------------------------------------
 
// Private Inbox
 
// -----------------------------------------------------------------------------
 

	
 
/// A structure that contains inbox messages. Some messages are left inside and
 
/// continuously re-read. Others are taken out, but may potentially be put back
 
/// for later reading. Later reading in this case implies that they are put back
 
/// for reading in the next sync round.
 
/// TODO: Again, lazy concurrency, see git history for other implementation
 
struct Inbox {
 
    messages: Vec<Message>,
 
    delayed: Vec<Message>,
 
    next_read_idx: u32,
 
    generation: u32,
 
}
 

	
 
#[derive(Clone, Copy)]
 
pub(crate) struct MessageTicket {
 
    index: u32,
 
    generation: u32,
 
}
 

	
 
impl Inbox {
 
    fn new() -> Self {
 
        return Inbox {
 
            messages: Vec::new(),
 
            delayed: Vec::new(),
 
            next_read_idx: 0,
 
            generation: 0,
 
        }
 
    }
 

	
 
    fn insert_new(&mut self, message: Message) {
 
        assert!(self.messages.len() < u32::MAX as usize); // TODO: @Size
 
        self.messages.push(message);
 
    }
 

	
 
    fn get_next_message_ticket(&mut self) -> Option<MessageTicket> {
 
        if self.next_read_idx as usize >= self.messages.len() { return None };
 
        let idx = self.next_read_idx;
 
        self.generation += 1;
 
        self.next_read_idx += 1;
 
        return Some(MessageTicket{ index: idx, generation: self.generation });
 
    }
 

	
 
    fn read_message_using_ticket(&self, ticket: MessageTicket) -> &Message {
 
        debug_assert_eq!(self.generation, ticket.generation);
 
        return &self.messages[ticket.index as usize];
 
    }
 

	
 
    fn take_message_using_ticket(&mut self, ticket: MessageTicket) -> Message {
 
        debug_assert_eq!(self.generation, ticket.generation);
 
        debug_assert!(ticket.index < self.next_read_idx);
 
        self.next_read_idx -= 1;
 
        return self.messages.remove(ticket.index as usize);
 
    }
 

	
 
    fn put_back_message(&mut self, message: Message) {
 
        // We have space in front of the array because we've taken out a message
 
        // before.
 
        self.delayed.push(message);
 
    }
 

	
 
    fn get_read_data_messages(&self, match_port_id: PortIdLocal) -> MessagesIter {
 
        return MessagesIter{
 
            messages: self.messages.as_slice(),
 
            next_index: 0,
 
            max_index: self.next_read_idx as usize,
 
            match_port_id
 
        };
 
    }
 

	
 
    fn clear_read_messages(&mut self) {
 
        self.messages.drain(0..self.next_read_idx as usize);
 
        for (idx, v) in self.delayed.drain(..).enumerate() {
 
            self.messages.insert(idx, v);
 
        }
 
        self.next_read_idx = 0;
 
    }
 

	
 
    fn transfer_messages_for_port(&mut self, port: PortIdLocal, new_inbox: &mut Inbox) {
 
        debug_assert!(self.delayed.is_empty());
 
        let mut idx = 0;
 
        while idx < self.messages.len() {
 
            let msg = &self.messages[idx];
 
            if let Some(target) = msg.target_port() {
 
                if target == port {
 
                    new_inbox.messages.push(self.messages.remove(idx));
 
                    continue;
 
                }
 
            }
 

	
 
            idx += 1;
 
        }
 
    }
 
}
 

	
 
// -----------------------------------------------------------------------------
 
// Control messages
 
// -----------------------------------------------------------------------------
 

	
 
struct ControlEntry {
 
    id: u32,
 
    variant: ControlVariant,
 
}
 

	
 
enum ControlVariant {
 
    NewComponent(ControlNewComponent),
 
    ChangedPort(ControlChangedPort),
 
    ClosedChannel(ControlClosedChannel),
 
}
 

	
 
impl ControlVariant {
 
    fn as_new_component_mut(&mut self) -> &mut ControlNewComponent {
 
        match self {
 
            ControlVariant::NewComponent(v) => v,
 
            _ => unreachable!(),
 
        }
 
    }
 
}
 

	
 
/// Entry for a new component waiting for execution after all of its peers have
 
/// confirmed the `ControlChangedPort` messages.
 
struct ControlNewComponent {
 
    num_acks_pending: u32,          // if it hits 0, we schedule the component
 
    component_key: ConnectorKey,    // this is the component we schedule
 
}
 

	
 
struct ControlChangedPort {
 
    reroute_if_sent_to_this_port: PortIdLocal, // if sent to this port, then reroute
 
    source_connector: ConnectorId,             // connector we expect messages from
 
    target_connector: ConnectorId,             // connector we need to reroute to
 
    new_component_entry_id: u32,               // if Ack'd, we reduce the counter on this `ControlNewComponent` entry
 
}
 

	
 
struct ControlClosedChannel {
 
    source_port: PortIdLocal,
 
    target_port: PortIdLocal,
 
}
 

	
 
pub(crate) struct ControlMessageHandler {
 
    id_counter: u32,
 
    active: Vec<ControlEntry>,
 
}
 

	
 
impl ControlMessageHandler {
 
    pub fn new() -> Self {
 
        ControlMessageHandler {
 
            id_counter: 0,
 
            active: Vec::new(),
 
        }
 
    }
 

	
 
    /// Prepares a message indicating that a channel has closed, we keep a local
 
    /// entry to match against the (hopefully) returned `Ack` message.
 
    pub fn prepare_closing_channel(
 
        &mut self, self_port_id: PortIdLocal, peer_port_id: PortIdLocal,
 
        self_connector_id: ConnectorId
0 comments (0 inline, 0 general)