Rust: a grouped use a::{B, C} becomes one import with local name {B, C}, so calls through it never use the import map #2922

Version: codebase-memory-mcp v0.12.1 (pkg/go/v0.12.1, a7158ac). main at 612eee2 has the
same parser (internal/cbm/extract_imports.c:1900, identical body) and the same lookup
(src/pipeline/registry.c:1859, called at :2321).

What I did

I indexed jdx/mise v2026.11.2 and compared its CALLS edges with a rust-analyzer SCIP index of
the same checkout. src/cli/run.rs:27 imports the handler in a group:

use crate::task::task_output_handler::{OutputHandler, SummaryOutcome};

and src/cli/run.rs:1808 calls OutputHandler::new(output_config). SCIP links the call to the
one OutputHandler.new in the project; cbm has no edge. Of mise's 5,172 Rust IMPORTS edges,
1,912 have a local_name that starts with {.

To isolate it I wrote a probe crate (mode full, cbm 0.12.1):

// src/h.rs
pub struct Handler { pub n: u32 }
impl Handler { pub fn new() -> Self { Handler { n: 1 } } }
pub struct Other;

// src/a.rs
use crate::h::{Handler, Other};
pub fn grouped() -> u32 { let _o = Other; Handler::new().n }

// src/b.rs
use crate::h::Handler;
pub fn single() -> u32 { Handler::new().n }

I indexed it twice: with src/many.rs empty, and with many.rs holding 300
pub struct S{i}; impl S{i} { pub fn new() -> Self { S{i} } }. The second run pushes the bare
name new past the registry's 256-candidate cap, so the name lookup can't cover for the import map.

What happened

a.rs also gets only one IMPORTS edge, to Handler. The import of Other is lost.

Why

parse_rust_imports (internal/cbm/extract_imports.c:1099-1127) makes one CBMImport per
top-level use_declaration. module_path is the whole text after use , and
local_name = path_last(full). path_last (:99-113) cuts after the last /, ., : or \,
so for crate::h::{Handler, Other} the local name is {Handler, Other}.

The IMPORTS edge stores that local name (src/pipeline/pass_parallel.c:1862-1866). The call-time
import map is built from those edges (src/pipeline/pass_lsp_cross.c:1007, read at :1054-1066).
resolve_import_map (src/pipeline/registry.c:1350) looks the callee's first segment up by exact
key (:1360-1368). Handler never equals {Handler, Other}, so the lookup is empty (:1370-1371)
and resolution falls through to the name lookup.

Proposal

• Walk the use_list and scoped_use_list children and push one import per member:
  • each with its own local name: the member's last segment, or its alias for as;
  • each with its full path, the group's prefix joined to the member;
  • self in a group (use a::b::{self, C}) binds b.
• Do the same for nested groups (use a::{b::{C, D}, E}).
• Add an extractor test: use crate::h::{Handler, Other as O}; gives two imports, Handler →
  crate::h::Handler and O → crate::h::Other.