Struct SyntaxIrFile 

pub struct SyntaxIrFile {
    pub language: Language,
    pub frontend_version: &'static str,
    pub ir_schema_version: u32,
    pub tokens: Vec<Token>,
    pub signatures: Vec<(ByteRange, Signature)>,
    pub roots: Vec<IrNode>,
    pub diagnostics: Vec<Diagnostic>,
    pub error_ranges: Vec<ByteRange>,
    pub depth_truncated: bool,
    pub test_module: bool,
}

The Syntax IR of one source file.

Fields

language: Language

Language the file was parsed as.

frontend_version: &'static str

Version tag of the structural frontend that produced this IR; a fingerprint input.

ir_schema_version: u32

IR schema version this file conforms to.

tokens: Vec<Token>

Tokens in source order, comments and whitespace removed. Same representation as Fast mode, so token-level normalization is shared.

signatures: Vec<(ByteRange, Signature)>

Function and method signatures keyed by their exact source range.

This is a reporting and semantic-evidence side table only. It is not read by structural or Fast fingerprint construction, so changing a signature never changes an existing structural identity.

roots: Vec<IrNode>

Top-level IR nodes in source order.

diagnostics: Vec<Diagnostic>

Recoverable lexical problems, as in Fast mode.

error_ranges: Vec<ByteRange>

Source regions the parser marked as errors. Overlapping nodes are still emitted; these ranges only lower confidence downstream.

depth_truncated: bool

Whether the frontend stopped descending after reaching its structural depth budget.

This is deliberately separate from Self::error_ranges: malformed source is ordinary recovery, while a depth ceiling means the scan was intentionally incomplete and must be reported as such.

test_module: bool

Whether the file is the body of a module its tree declares test-only.

Not something a parse can answer: the declaration carrying the marker is in another file, so this is settled once the whole set is known and left here for the walk that reads a unit’s markers to start from. A frontend leaves it false. See declared_test_modules.

Implementations

impl SyntaxIrFile

pub fn signature_for_range(&self, range: ByteRange) -> Option<&Signature>

Look up a signature by exact source range.

pub fn walk(&self, visit: &mut impl FnMut(&IrNode))

Depth-first pre-order traversal over every node in the file.

pub fn node_count(&self) -> usize

Total number of nodes in the file.

pub fn unaccounted_tokens(&self) -> usize

Tokens the parser could not attach to any structure: those inside a Shape::Error node that none of its children recovered.

This is the honest measure of what a parse lost, and error_ranges is not. An error-tolerant parser recovering from one bad construct routinely wraps everything around it in a single error node: a header whose include guard encloses the file gets one error region covering every byte of it, with the whole file’s declarations intact inside. Measured over one project’s C++ sources, the error regions covered 13.3% of the bytes while the tokens that actually failed to parse were 1.82% — the difference is entirely code the parser did read, sitting inside a region it had to open.

Nesting is not double-counted: an error node inside another is covered by its parent’s children, so the parent contributes only the gaps around it and the child contributes its own.

Trait Implementations

impl Clone for SyntaxIrFile

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for SyntaxIrFile

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.