fix: add receiver field to call sites to eliminate false positive edges

[carlos-alm]

Summary

• Receiver-aware call sites: Add optional receiver field to call site extraction across all 11 language extractors (WASM + Rust native) to distinguish obj.method() from standalone() calls. Skip global fallback for method calls with a receiver (except this/self/super)
• Scoped fallback: standalone calls no longer resolve globally across the entire codebase — only same-directory (confidence 0.7) and parent-directory (0.5) matches are kept, dropping random cross-codebase matches (0.3)
• Edge deduplication: tracks seen caller→target pairs per file, preventing duplicate edges when a function is called multiple times
• Built-in receiver skip: fast-path skip for calls on known runtime globals (console, Math, JSON, Object, Array, Promise, process, Buffer, etc.) that never resolve to user-defined symbols
• MCP schema fix, benchmark automation, lint/format cleanup

Results (dogfooding on self)

Metric	Before	After receiver fix	After scoped+dedup+builtin
Total edges	~1742	~1321 (−24%)	~615 (−57% total)
False positives	~52%	reduced	near-zero

Test plan

• All 425 tests pass
• npm run lint — clean
• node src/cli.js build . — clean rebuild, 615 edges
• node src/cli.js fn buildGraph — 11 correct call dependencies
• node src/cli.js fn openDb — 3 correct cross-file callers
• node scripts/benchmark.js — WASM 690 / Native 716 edges

[greptile-apps]

Greptile Summary

This PR eliminates false positive call graph edges by adding a receiver field to distinguish method calls (obj.method()) from standalone calls (method()). The change propagates across all 11 language extractors in both WASM (JS) and native (Rust) implementations, with updates to the graph resolution logic to skip global fallback for method calls with explicit receivers (except this/self/super). Testing confirms a 24% reduction in graph edges on self-analysis (1742→1321), with all removed edges being false positives.

Key changes:

• Added optional receiver field to Call struct in types.rs and all extractor implementations
• Updated builder.js resolution logic to skip global fallback for method calls with receivers
• Comprehensive test coverage including edge cases (.call()/.apply()/.bind() unwrapping, chained calls)
• Parity test updated to validate receiver field across engines
• Deleted stale documentation file

Implementation quality:

• Consistent pattern across all 11 languages (JS, TS, Python, Go, Rust, Java, C#, PHP, Ruby, HCL)
• Both WASM and native extractors updated for engine parity
• Well-tested with specific receiver extraction scenarios
• Clean, minimal changes to core resolution logic

Confidence Score: 5/5

• Safe to merge - well-architected change with comprehensive test coverage and proven impact on real codebase
• The implementation is consistent across all 11 language extractors (both WASM and native), follows established patterns, includes comprehensive test coverage, and has been validated through dogfooding on the codebase itself. The 24% reduction in false positive edges demonstrates real value. The change is purely additive (optional field) with backward compatibility maintained.
• No files require special attention

Important Files Changed

Filename	Overview
crates/codegraph-core/src/types.rs	Added optional receiver field to Call struct - straightforward and safe type extension
src/builder.js	Updated global fallback logic to skip method calls with receivers (except this/self/super), eliminating false positive edges
src/extractors/javascript.js	Added extractReceiverName helper and receiver extraction for all call types including member expressions and subscript expressions
crates/codegraph-core/src/extractors/javascript.rs	Added receiver extraction to Rust native JavaScript extractor for parity with WASM implementation
src/parser.js	Added receiver field to native symbol normalization
tests/parsers/javascript.test.js	Added comprehensive test coverage for receiver extraction including edge cases like .call()/.apply()/.bind() unwrapping
tests/engines/parity.test.js	Updated parity test to compare receiver field between WASM and native engines

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Source Code] --> B[Tree-sitter Parse]
    B --> C{Call Type?}
    
    C -->|Standalone call<br/>fn| D[Extract: name=fn<br/>receiver=undefined]
    C -->|Method call<br/>obj.method| E[Extract: name=method<br/>receiver=obj]
    C -->|this/self call<br/>this.foo| F[Extract: name=foo<br/>receiver=this]
    
    D --> G[builder.js Resolution]
    E --> G
    F --> G
    
    G --> H{Has receiver?}
    H -->|No receiver| I[Try global fallback]
    H -->|Receiver = this/self/super| I
    H -->|Receiver = other| J[Skip global fallback<br/>Prevents false positives]
    
    I --> K[nodesByName.get]
    J --> L[Only use method hierarchy]
    
    K --> M[Create Edge]
    L --> M
    
    style J fill:#90EE90
    style K fill:#FFB6C6
    style L fill:#90EE90

Loading

_{Last reviewed commit: b08c2b2}

[greptile-apps]

_{22 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}