[calvin-archastro]

Review on ArchCode

Epic: Aster bug bash fixes
Task: JS --related follows export { a as b } from and namespace re-exports to their importers

Problem and author intent

--related missed JS/TS importers that reached a changed export only through a renaming source re-export or a re-exported namespace. Reported against aster 0.30.1 (GitHub issue 240): with export { orig as renamed } from "./orig.js", import * as o from "./orig.js"; export { o }, and a direct importer, editing orig planned only test/direct.test.js. Without that direct test, the rename barrel was selected and the namespace barrel still was not. export { default as X } from failed the same way. Same-name export { orig } from, export * as oo from, and import { orig }; export { orig as renamed2 } already worked.

Before, Walker::export treated an export with a module source as a unit edge only (reexports) and dropped specifier names and aliases. Symbol matching is by imported name, so a test of renamed or o does not match a change to orig. The unit edge is not enough once a named direct importer exists, because that suppresses the users(unit) fallback. Local export { a as b } already created a Const definition of b whose refs held a. A bare local export { o } created nothing, and the namespace import alone was not a unit edge, so importers of that binding never matched.

Intent is to add the missing name edges without changing same-name or star re-exports, and without touching dynamic imports.

What changed

Source export { name as alias } from spec, including default as X, now also records a Const definition of the alias. refs holds the source name (orig or default) and uses holds the module spec. reexports is unchanged. A same-name source re-export still adds no alias definition.

Local export { … } clauses are walked through a shared export_specifiers helper. Each specifier is stored as an Exported (local, exported, span) and resolved after imports are bound. If that local name is a namespace import (imported is none), the file gains a reexports edge for the namespace spec. An existing alias definition gets uses; a bare export { o } (which previously created no definition) gets a Const definition of the name importers use, with refs containing that name and uses containing the spec. Export-before-import is covered because resolution happens after binding. export * as stays reexport-only.

Scope

• src/related/js.rs — export fact extraction and binding resolution
• tests/related.rs — selection regression
• Diff evidence for these paths is complete in this payload: 10 hunks, +296/−18, 0 omitted hunks, paths, or patch bytes
• Dynamic import handling is unchanged

Risk assessment

Medium. Selection widens only for importers of renaming re-exports and of re-exported namespaces. Namespace matching is an accepted over-approximation: any changed export of the namespace module selects importers of the namespace binding, even if the test never names that export. Barrel files are common, so affected-test sets can grow; this change does not include a large barrel fixture. Same-name and export * edges are intentionally unchanged.

User impact

aster --related on JS/TS projects selects tests that import through export { a as b } from, export { default as X } from, and import * as o; export { o } (including a rename of that namespace). Direct importers are still selected. Tests of unrelated modules are not selected by these edges.

Testing

The diff adds, but does not prove execution of:

• source_reexports_that_rename_are_definitions_of_the_alias and a_reexported_namespace_is_a_definition_of_that_binding in src/related/js.rs (alias defs, default as X, same-name and star re-exports stay definition-free, namespace re-export including export-before-import)
• js_renamed_and_namespace_reexports_reach_their_importers in tests/related.rs (issue layout with direct/rename/namespace tests, the no-direct variant, default as X, and an unrelated export selecting the namespace importer but not another module)

Pipeline checks format, lint, test, fix, reformat, verify-lint, verify-test, and commit are reported succeeded with exit 0. Check excerpts are truncated and do not name these functions, so that success is not evidence the new tests ran. Author-reported cargo test --lib related::, cargo test --test related, clippy, and fmt are advisory only and are not confirmed here.

Follow-ups and known issues

• Dynamic import selection (issue 239) remains out of scope.
• Namespace over-selection of unnamed exports is accepted, not fixed.
• No measurement here of selection size on a large barrel.

[calvin-archastro]

ArchDev is watching this pull request. State: watching (cycle 2 of 2 for head 3a9aa51).

when	trigger	decision	result
2026-11-01T16:42:50.973Z	merge conflict	enqueue job_5110faeec72161ac68c47096	failed Pipeline step finalize-conflict failed: command exited with code 1
2026-11-01T17:01:23.574Z	verify head	enqueue job_9e8b199d64a671cb2bbdd8e2	no change

Paused by the no-autofix label or a draft; never resolves reviewer threads. Jobs: job_5110faeec72161ac68c47096, job_9e8b199d64a671cb2bbdd8e2.