Crate code2graph 

code2graph

Source files → structural facts. A purpose-neutral, language-agnostic code-graph extraction library: it turns source code into symbols, references, and cross-file edges as plain data — no storage, no scoring, no embeddings, no judgement. See README.md for the design boundary.

Pipeline

source ──[extract]──▶ FileFacts (symbols + references) ──[resolve]──▶ CodeGraph (symbols + edges)

use code2graph::{extract_path, resolve::{Resolver, SymbolTableResolver}};

let a = extract_path("src/util.rs", "pub fn helper() {}").unwrap();
let b = extract_path("src/main.rs", "pub fn run() { helper() }").unwrap();
let graph = SymbolTableResolver.resolve(&[a, b]);
assert_eq!(graph.edges.len(), 1); // run --calls--> helper

Design

• Identity (symbol) is SCIP-aligned: a symbol is a descriptor path rendering to a stable, human-readable string, so cross-file matching is string equality.
• Resolution (resolve) is a tier seam: the fast recall-first SymbolTableResolver (name matching, all languages, NameOnly edges) and the precise scope-aware ScopeGraphResolver (lexical-scope + import + qualified-path resolution, Scoped/Exact edges, Rust/Python/TypeScript) emit the same schema, tagging every edge with a graph::Confidence and a graph::Provenance (which analysis derived it, orthogonal to confidence). A consumer picks the tier; the output shape is identical.
• Cross-language bridges (FfiBridgeResolver) link call sites to FFI exports (Rust #[no_mangle] → C, today) across a language boundary, deterministically and with honest confidence — composable on top of any tier.
• Incremental maintenance (IncrementalGraph) keeps a resolved graph current as files change: each file is resolved in isolation and cross-file edges are stitched on demand, so re-extracting one file rebuilds only that file’s subgraph — never the whole workspace.
• No storage, no source bodies — graph::Symbols carry a byte span; consumers slice what they need.

Coverage

All 23 languages (lang::Language) are implemented end-to-end, each behind the extract::Extractor trait.

Re-exports

pub use error::CodegraphError;

pub use error::Result;

pub use extract::Extractor;

pub use extract::extract_file;

pub use extract::extract_path;

pub use graph::Binding;

pub use graph::BindingKind;

pub use graph::BindingTarget;

pub use graph::ByteSpan;

pub use graph::CodeGraph;

pub use graph::Confidence;

pub use graph::Edge;

pub use graph::EntryPoint;

pub use graph::FfiAbi;

pub use graph::FfiExport;

pub use graph::FileFacts;

pub use graph::Occurrence;

pub use graph::Provenance;

pub use graph::RefRole;

pub use graph::Reference;

pub use graph::Scope;

pub use graph::ScopeId;

pub use graph::ScopeKind;

pub use graph::Symbol;

pub use graph::SymbolKind;

pub use graph::TypeRefContext;

pub use graph::Visibility;

pub use lang::Language;

pub use resolve::FfiBridgeResolver;

pub use resolve::FileSubgraph;

pub use resolve::IncrementalGraph;

pub use resolve::LayeredResolver;

pub use resolve::Resolver;

pub use resolve::ScopeGraphResolver;

pub use resolve::SymbolTableResolver;

pub use symbol::Descriptor;

pub use symbol::Package;

pub use symbol::SymbolId;

Modules

error

CodegraphError — errors surfaced by extraction and resolution.

extract

Extraction: one tree-sitter pass per language → neutral FileFacts.

grammar

Grammar chokepoint — the sole importer of every tree_sitter_* grammar crate.

graph

Neutral graph data model — the facts code2graph produces.

lang

The set of languages code2graph can parse, plus extension dispatch.

package

Optional package enrichment: stamp Package identity onto extracted facts.

resolve

Resolution: link references to definitions, producing cross-file edges.

symbol

SCIP-aligned symbol identity: descriptors and SymbolId.