C# extractor produces zero `references` edges — type annotations on parameters, return types, and fields are never extracted

[danielsogl]

Problem

For C# projects, extractTypeAnnotations runs (csharp is in TYPE_ANNOTATION_LANGUAGES) but emits zero references edges. This breaks codegraph_callers, codegraph_impact, and codegraph_context for any DTO, interface, record, or value type used as a parameter/return type — even though the same tools work correctly for TypeScript.

Reproduction

Indexed a .NET 10 + ASP.NET Core monorepo (533 C# files, 217 TS files):

SELECT n.language, e.kind, COUNT(*)
FROM edges e JOIN nodes n ON n.id = e.source
GROUP BY n.language, e.kind;

language	kind	count
csharp	calls	810
csharp	extends	70
csharp	implements	20
csharp	instantiates	169
csharp	imports	1338
csharp	contains	3296
csharp	references	0 ← bug
typescript	references	192

Concrete query failures on the same repo:

• codegraph_callers SessionInfoDto → 0 callers (grep finds 6 type usages in Produces<SessionInfoDto>, return signatures, etc.)
• codegraph_callees DataExporter (a class with 11 methods, each taking/returning multiple DTO types) → returns only using imports, no type references
• unresolved_refs table is empty for language='csharp' → references are never collected, not just never resolved

Root Cause (from dist/ inspection)

Two issues in extraction/languages/csharp.js and extraction/tree-sitter.js:

1. csharp.js is missing the returnField:

exports.csharpExtractor = {
    methodTypes: ['method_declaration', 'constructor_declaration'],
    paramsField: 'parameter_list',
    // returnField missing!  Falls back to 'return_type' in
    // tree-sitter.js, but C# uses field name 'type'.
};

Compare: java.js, dart.js, kotlin.js, pascal.js all set returnField: 'type'. TypeScript, Rust, Python, PHP use 'return_type'. C# has no setting at all, so extractTypeAnnotations never finds the return-type child.

2. extractTypeRefsFromSubtree only matches type_identifier:

if (node.type === 'type_identifier') {
    // emit 'references' edge
}

C# tree-sitter does not produce type_identifier nodes. C# type positions contain:

• identifier → simple class/interface names
• predefined_type → int, string, bool
• qualified_name → System.String, MyNs.Foo
• generic_name → List<T>, Task<Foo>
• array_type, nullable_type, pointer_type

So even if paramsField/returnField resolved correctly, the walker would still emit nothing because no type_identifier node exists.

Suggested Fix

a) Add returnField: 'type' to csharp.js (one line).

b) Extend extractTypeRefsFromSubtree with a language-aware node-type set, or make it a virtual method on the extractor so each language can override what counts as a "type reference leaf". For C#, recurse into type positions and capture identifier | predefined_type | qualified_name | generic_name.

Caveat: Naively matching identifier everywhere will produce false positives for parameter names. The walker should only recurse into the type field of each parameter, not the whole parameter_list. The cleanest implementation is a custom extractTypeRefs(node, source) method on the C# extractor that manually navigates parameter_list → parameter → type and method_declaration → type.

Impact

C# coverage is a major use case for the project (Java/Kotlin/Dart all handle type references; C# is the outlier). Without references edges, the headline feature "structural lookups grep can't do" — finding all consumers of a DTO or interface across a codebase — silently degrades to text search for ~half of typical backend stacks.

Happy to take this on as a PR if a maintainer can sanity-check the fix approach (custom extractor method vs language-aware walker extension).

Environment

• codegraph version: latest npm @colbymchenry/codegraph
• Node: lts via mise
• Project: .NET 10, mixed with Angular 21 TypeScript frontend
• DB backend: node:sqlite WAL