[colbymchenry]

Summary

A Rust path into a module, like crate::util::take(1), super::util::take(1), util::take(1) or use crate::util::take;, now links to the item that module declares. Before, it went to the first node of that name in the module's file, methods included, so an impl Buf { fn take } written above pub fn take caught every such call.

Root cause

resolveRustPathReference mapped the path's module to a file and took getNodesInFile(file).find(n => n.name === leaf && …), with method among the accepted kinds. The first same-named node won: a method of an impl above the free function, a function written inside another function, or an item of a mod tests { … } above the module's own. A path into the module's own file was refused outright, and a path whose head a use binds from the standard library (fs::read under use std::fs;) went to a project module of the same name.

Fix

The path's item is the module's own: never a method, an impl's or trait's associated item, or an item of a function body; an item written at the file's top level comes before one in a module the file declares inline, which the path reaches only through the module's re-export (#[cfg(unix)] mod imp { … } + pub use imp::*;). A head a use binds from the standard library leaves the path to it, and a path into the module's own file resolves.

From the contributor, and what we changed

From @danusha2345's PR #2308: the new candidate selection in resolveRustPathReference (no methods, lexically reachable, module-level first, inline-module fallback, same-file paths, the std-head rule) and __tests__/rust-module-path-call.test.ts.

What we changed:

• Their module-level and inline-module checks, and the std-head rule, read a second Rust use parser the PR adds (rust-use-bindings.ts). Here they use what main already has: a small rustItemPlacement helper over main's rustUsesOf/rustBlockAt (the same test rustDeclaredItems makes), and main's isBoundToOutOfRepoImport for the std-bound head (the PR's rule, minus the source positions).
• Left out: the bare-call bindings (as aliases, block-local use, shadowing by locals, enum variants through a use), the new parser and its unit test, and the file-walker changes that depend on it. main's use walk (rustUsesOf, rustWalkModules, rustModuleItems, fix(rust): a name a use brings in is the item of the module its path walks #2702/fix(rust): a name a glob use brings in is an item the globbed module passes on #2714) and scope reader (rust-scopes.ts, fix(rust): a call on a local reaches the method of the type it is declared with #2715) already derive all of that; folding those cases into them is proposed to the maintainer separately.
• Added three cases to their test file: a namesake in an inline mod tests before the module's item, an inline module's item reached through its re-export from its own file, and fs::read under use std::fs; beside a project src/fs/mod.rs.
• The CHANGELOG bullet is rewritten for this scope.

Validation

• __tests__/rust-module-path-call.test.ts: 8/8. On main, 7 fail: method Buf::take, Deep::get, Only::grab, no imports edge for use crate::util::take;, nothing for the inline-module namesake and the re-export, and fs::read → src/fs/mod.rs::read.

• A/B, main vs this branch, on 6 Rust repos (both arms on main 7cceba5, same kernel; after rebasing onto 2577601, express and flask were rechecked byte-identical and gleam's delta is unchanged):

  repo	changed sites	what moved
  gleam	152	133 calls like fs::write(…), crate::fs::mkdir(…) move from impl … for ProjectIO's methods to the free functions in compiler-cli/src/fs.rs (plus 2 erlang::module from Generator::module); 4 type_::list(…) → Type::list edges are gone (the real one is type_/prelude.rs::list, through pub use prelude::*); 15 uses of a file's own items inside its test modules link
  dioxus	8	7 links (same-file paths and uses), 1 more target on a shared column
  tokio	5	crate::io::uring::open::open(…) moves from Op::open to the free open; 4 links (crate::sync::watch::channel from watch.rs itself, test-module uses)
  ripgrep	1	use crate::{…, dir::IgnoreBuilder, …} inside dir.rs links
  clap	0
  serde	0

  Every changed site was inspected; all 162 are right. The 135 retargets were each checked to be a module-path call at the site. express and flask are byte-identical.

• Kernel parity: not affected (resolution only; no extraction change).

• Cost (medians of 3 interleaved index runs, both arms on main 7cceba5):

  repo	CPU main → branch	peak RSS main → branch
  tokio	36.6 s → 36.0 s	980 → 986 MB
  gleam	34.3 s → 35.2 s	1003 → 1012 MB

  Both differences are within the runs' spread. In a CPU profile of tokio the changed functions take under 0.1 s. The standard-library check reads the file's uses, so it is asked only when the path would otherwise link a project item; asked for every path it cost 0.3 s on tokio.

• Full suite on the exact commit (Windows, kernel built, CODEGRAPH_KERNEL_EXPECT=1, --maxWorkers=4): 776 files, 9,175 passed, 0 failed, 85 skipped.

Lands #2308 by @danusha2345.

🤖 Generated with Claude Code