#1790's *.mjs text eol=lf does not repair existing working trees — 3 suites stay dead after merge #1793

Found while building .squad/e2e/windows-test-baseline.md (EECOM, baseline commit 369bba8f).

Summary

#1790 adds *.mjs text eol=lf to .gitattributes, which fixes the #1788 dead-gate class on a fresh checkout. It does not fix a working tree that already exists — and every Windows developer running the E2E series has one.

Merging #1790 and pulling will leave check-changeset-drift.test.ts, promote-insider-tag.test.ts, and patch-esm-imports.test.ts still red, which will read as "the fix didn't work."

Mechanism

.gitattributes governs checkout, not files already on disk. Git only rewrites a working file when the merge changes its index content.

• core.autocrlf = true on Windows.
• These three .mjs are already stored LF in the index, so fix(ci): force LF on *.mjs so shebanged scripts stay loadable by vitest #1790 does not renormalize them (it renormalizes only the 9 .mjs that were CRLF in the index).
• Index unchanged ⇒ working file untouched ⇒ stays CRLF on disk ⇒ Vite's shebang stripping still leaves a bare # ⇒ suite loads zero tests.

Controlled proof

Applied #1790's .gitattributes line locally without a re-checkout:

$ git ls-files --eol scripts/promote-insider-tag.mjs packages/squad-cli/scripts/patch-esm-imports.mjs
i/lf    w/crlf  attr/text eol=lf      packages/squad-cli/scripts/patch-esm-imports.mjs
i/lf    w/crlf  attr/text eol=lf      scripts/promote-insider-tag.mjs

$ npx vitest run test/promote-insider-tag.test.ts
 Test Files  1 failed (1)
      Tests  no tests

Attribute applied. Disk still CRLF. Still zero tests.

Then forced a re-checkout of the same files:

$ Remove-Item <the .mjs files> -Force ; git checkout -- <the .mjs files>
$ git ls-files --eol ...
i/lf    w/lf    attr/text eol=lf      packages/squad-cli/scripts/patch-esm-imports.mjs
i/lf    w/lf    attr/text eol=lf      scripts/promote-insider-tag.mjs

$ npx vitest run test/promote-insider-tag.test.ts test/cli/patch-esm-imports.test.ts
 Test Files  2 passed (2)
      Tests  21 passed (21)

Suggested fix

This is a rollout gap, not a code bug in #1790. Options, cheapest first:

1. Document it in fix(ci): force LF on *.mjs so shebanged scripts stay loadable by vitest #1790's PR body and in the release/upgrade notes: after merging, run git rm --cached -r . ; git reset --hard (or targeted delete + git checkout --) and verify with git ls-files --eol '*.mjs' → expect w/lf.
2. Have squad doctor check git ls-files --eol '*.mjs' for any w/crlf and emit a fix-it hint. This makes the condition self-diagnosing instead of presenting as mystery test failures.

Success criteria

• A developer with a pre-fix(ci): force LF on *.mjs so shebanged scripts stay loadable by vitest #1790 Windows checkout can follow one documented command and get git ls-files --eol '*.mjs' reporting w/lf for every entry.
• After that command, check-changeset-drift, promote-insider-tag, and patch-esm-imports all run their tests (8 + 15 + 6 = 29 tests, none reporting no tests).
• Ideally: squad doctor flags any w/crlf .mjs so this never has to be remembered.

Related

• check-changeset-drift.test.ts silently runs zero tests on Windows (shebang + CRLF) #1788 / fix(ci): force LF on *.mjs so shebanged scripts stay loadable by vitest #1790 — the underlying CRLF shebang defect and its fix. This issue is the residual, not a duplicate.
• .squad/e2e/windows-test-baseline.md §4 for the full measurement.