fix(rust): a variant path references its enum (#2328)

[danusha2345]

Fixes #2328. Thanks @mg-mg-mg for the report and the repro script.

Problem

A Rust enum used only through its variants records no reference to the enum. This covers mode::Mode::A in an expression and Mode::B => …, Mode::C(x) or Mode::D { .. } in a pattern. A type annotation (fn takes(_m: Mode)) does record one. So the functions that build or match on variants without naming the type in a signature were missing from the enum's callers and impact.

Fix

Extraction (wasm extractor and native kernel together). Rust joins the static-member pass that the other capitalized-type languages already run (Enum.value, Type.CONST). A path whose member is capitalized, meaning a variant or an associated const, emits a references ref to its receiver. The receiver is the segment before the member, and the ref sits where it is written. This is the same bare-name ref a type annotation produces, so a mode:: / other:: prefix scopes it the way it scopes fn f(_: other::Mode).

• Value reads (Mode::A, mode::Mode::A.flip(), .map(Mode::C)), match arms, if let / | patterns, tuple-variant patterns and struct-variant patterns (Mode::D { x }).
• Self::A inside an impl names the impl's type.
• These emit nothing:
  • A lowercase receiver, which is a module (util::take(3), u8::MAX), or a lowercase member, which is a function (Foo::new as a value).
  • A call's callee (Mode::C(1), Foo::new()) and a struct literal's name (Mode::D { x: 1 }). Both are already linked to their member by calls / instantiates, as in the other languages. So Type::new() still does not reference Type.
  • The prefix of a longer path, and use trees.

Resolution. In isRustNameInScope, a type read through one of its items is a type that declares that item. That means an enum with that variant, or a type whose file defines that associated const. Without this, a bare name shared across crates landed on a same-named struct that lacks the member. On ripgrep, ignore's Match::None and Error::Partial went to the matcher's struct Match and globset's struct Error. A loose path match did the same on tokio, where scheduler::Handle::MultiThread went to current_thread's struct Handle. With the check, the right enum is the only candidate left, or nothing links.

EXTRACTION_VERSION 27 → 28.

Tests

• New __tests__/rust-enum-variant-reference.test.ts builds real temp Cargo projects in SQLite with no mocks:
  • the issue's repro;
  • a bare Mode::A after use, with the edge's line and column;
  • tuple and struct variant patterns, and a variant passed as a value;
  • Self::A in the enum's file and in an impl in another file;
  • picking the type that declares the variant over a same-named struct, and an external termcolor::ColorChoice::AlwaysAnsi that does not reach the file's own ColorChoice;
  • an associated const (Limits::MAX);
  • negative controls: util::take(3), mode::Mode::new(), and other::Mode::A against mode::Mode;
  • two cases that are unchanged on purpose: constructing Mode::C(3) / Mode::D { x } stays linked to the variant, and a variant imported with use Mode::* and used bare gets no edge.
• Results: 7 of 9 fail on main in both modes, and all 9 pass with the kernel (CODEGRAPH_KERNEL_EXPECT=1) and with wasm (CODEGRAPH_KERNEL=0). With main's kernel binary under this branch's TypeScript, the native run fails the extraction cases again, so the kernel half is exercised.
• kernel-rustlang-parity has a new case covering every shape above, plus a non-ASCII column and CRLF.
• All kernel-*-parity, rust-*, resolution, extraction and the other Rust-touching suites pass in both modes.

A/B (origin/main → this branch)

repo	nodes	edges	new references → enum	→ struct (assoc const)	removed
ripgrep @3fce3b5	4141 =	+825	822	3	0
tokio @5d5cd8b	14168 =	+887	665	222 (Interest::READABLE, Ready::*, Semaphore::MAX_PERMITS)	0
two private Rust GUI/CLI crates	=	+505 / +122	504 / 122	1 / 0	0

• Every new enum edge targets an enum that declares the variant written. The exceptions are 5 edges in the private crates, which go through an enum's own associated const (Column::ALL). Every struct edge targets a struct whose file defines the const.
• New distinct function→enum pairs: ripgrep +220, tokio +254.
• I spot-checked 10 edges by hand and all were correct.
• One wrong edge remains, on tokio. oneshot::error::TryRecvError::Empty goes to mpsc's TryRecvError, which also has Empty. The path-scope check accepts any file named error.rs for the error:: segment, and type annotations already have the same issue.
• Index time is within noise (ripgrep 1.8–1.9 s → 1.9–2.2 s, tokio 7.0–7.2 s → 7.1–7.2 s).

Not in this PR

• A variant brought in by use Mode::* or use Mode::A and used bare (A =>). There is no path to read the enum from, and covering it would mean a ref for every capitalized identifier (None, Some, …).
• Paths inside macro token trees (matches!(m, Mode::A)). tree-sitter does not parse them, and calls there are not extracted either.
• A -> Self / &Self type annotation still records no reference.

🤖 Generated with Claude Code

[colbymchenry]

Thanks @danusha2345! This landed in #2354, which keeps your extraction code in both the TypeScript extractor and the native kernel, and most of your tests. Three changes on top. Associated consts like Limits::MAX no longer reference the type: impl consts are file-level variables in the graph, so a same-named constant can't be tied to that type. Both walkers now pass each node's parent down instead of looking it up, which brings kernel extraction on tokio back from about 3.9 s to 2.1 s. And the EXTRACTION_VERSION bump was dropped because it was already bumped this release. You're credited as a co-author on the commit and in the changelog. Closing this one in favor of #2354.