fix(extract-rust): inline test fns (#[test]/#[tokio::test]) are not marked is_test — leak into graph/agent context #855

Summary

Inline Rust test functions — #[test], #[tokio::test], #[async_std::test], #[actix_rt::test], and #[cfg(test)] mod tests { … } bodies — that live inside a regular source file (e.g. src/lib.rs) are not detected as tests. They are indexed as ordinary Function nodes with is_test = false, so they are not filtered by the is_test != 1 clause in store.c and leak into graph queries / agent context, polluting it with test code.

Root cause

Test detection is file-path-based only:

• cbm_is_test_file(rel_path, lang) (internal/cbm/helpers.c:292) marks a file/module as a test only by path. For Rust it is return has_suffix(base, "_test.rs") || has_prefix(base, "test_"); (with the code comment "Rust tests are typically mod tests inside the file, but test files too"). A file like src/lib.rs is therefore never a test file, regardless of its contents.
• CBMDefinition.is_test (internal/cbm/cbm.h:213) is the per-function flag. It is read by src/pipeline/pass_definitions.c and src/pipeline/pass_parallel.c (emitted into node properties) but is never assigned in the extraction layer — the only is_test write is mod.is_test = ctx->result->is_test_file; (internal/cbm/extract_defs.c:6316), which is module/file-level. So def->is_test stays at its zero-initialized default (false).

Because functions are never marked, the store.c filter (json_extract(properties, '$.is_test') != 1) only ever drops whole test-files, never inline test functions.

Repro

Index any repo containing, in a non-test-named file:

// src/lib.rs
pub fn real_fn() {}

#[cfg(test)]
mod tests {
    #[tokio::test]
    async fn something_works() {}
}

cbm cli --json search_graph '{"project":"…","name_pattern":"something_works"}' returns something_works as a Function with is_test absent/false — i.e. it is not filtered and appears alongside real_fn in graph context.

Proposed fix (Rust-scoped)

In the Rust function-extraction path in internal/cbm/extract_defs.c, scan each function's attribute_item / meta_item children for test macros and set def->is_test = true. Initial attribute set:

• bare #[test]
• #[tokio::test] (the case that surfaced this — async tests)
• #[async_std::test]
• #[actix_rt::test]
• #[test_case::case] (optionally)

Plus a regression test in tests/test_extraction.c asserting a Rust snippet with #[tokio::test] async fn … yields def->is_test == true. make -f Makefile.cbm test + scripts/lint.sh clean.

Scope

Rust only in the follow-up PR. Rust is a high-value first target because inline #[cfg(test)] mod tests { … } is idiomatic, so these unmarked test functions leak into the graph for any typical Rust crate — not an edge case. Go / Python / JS / Java / C# can follow in separate PRs per the one-issue-per-PR guidance.

Ask

Status: PR #857 implements this (Rust-scoped, with a regression test; scripts/test.sh green, red→green verified). Feedback on the attribute set and the extraction-hook point welcome there or here — happy to adjust. (Filed for alignment first; went ahead since it's a focused correctness fix.)

Citations verified against main at 37c00d1.