Both found during review of #1. Neither is a bug in shipped code — both are tests that would stay green through a real regression.
1. tests/proxy-accumulation.test.ts — "does not rewrite when a later importer adds nothing"
It asserts the emitted body is byte-identical after a second ensureProxy call that adds no names. But a redundant re-emit produces byte-identical content too, so the assertion cannot distinguish "skipped the write" from "wrote the same bytes again". A reviewer confirmed this empirically: disabling the entire if (!firstTouch && !grew && exists) return guard leaves the test passing.
To actually test the skip, count writes — e.g. a FilesApi wrapper that tallies write calls for that path.
2. tests/deps-proxy-server.test.ts — "class-as-adapter-key: two importers share one root proxy…"
The name claims it proves shape merging, but both importers use an identical import shape (import { K }), so it would pass under first-write-wins as well. It really only proves the shared root path exists.
The merging property IS genuinely covered — by its sibling React test, where export default __m comes only from a.ts and export const useState only from b.ts, so either write order without accumulation drops one. So this is a naming/coverage-clarity fix, not a hole.
🤖 Filed with Claude Code
Both found during review of #1. Neither is a bug in shipped code — both are tests that would stay green through a real regression.
1.
tests/proxy-accumulation.test.ts— "does not rewrite when a later importer adds nothing"It asserts the emitted body is byte-identical after a second
ensureProxycall that adds no names. But a redundant re-emit produces byte-identical content too, so the assertion cannot distinguish "skipped the write" from "wrote the same bytes again". A reviewer confirmed this empirically: disabling the entireif (!firstTouch && !grew && exists) returnguard leaves the test passing.To actually test the skip, count writes — e.g. a
FilesApiwrapper that tallieswritecalls for that path.2.
tests/deps-proxy-server.test.ts— "class-as-adapter-key: two importers share one root proxy…"The name claims it proves shape merging, but both importers use an identical import shape (
import { K }), so it would pass under first-write-wins as well. It really only proves the shared root path exists.The merging property IS genuinely covered — by its sibling React test, where
export default __mcomes only froma.tsandexport const useStateonly fromb.ts, so either write order without accumulation drops one. So this is a naming/coverage-clarity fix, not a hole.🤖 Filed with Claude Code