[Mikola Lysenko (mikolalysenko)]

LLM Description written by Claude Code:claude-opus-5-5

Fixes #1689

Problem

Tonight's nightly CI on main (run 40067511756, head 3db3cdf) fails on one leg only, test (windows-latest, 1) (job 120794033397). Every other job is green, including test (windows-latest, 2), so the fixes from #1686 still hold. The only failing test is:

patch::redirect::yarnrc::tests::global_folder_values_read_like_yarn
panicked at crates\socket-patch-core\src\patch\redirect\yarnrc_tests.rs:594:9:
assertion `left == right` failed: "global-folder=\\${HOME}/x\n"
  left: ["\\${HOME}/x"]
 right: ["/w/\\${HOME}/x"]

#1696 added this test yesterday. Windows test legs run only in the full-scope tier, so neither PR CI nor the merge queue saw it.

Root cause

The escaped-expression case resolves the verbatim value \${HOME}/x against the cwd /w (resolve_with_home → cwd.join(value)). On Unix that gives /w/\${HOME}/x. On Windows a leading \ is root-relative: Path::join keeps the cwd's drive (/w has none) and replaces the rest, giving \${HOME}/x. Yarn behaves the same way, since Node's path.resolve('C:\\w', '\\${HOME}/x') is C:\${HOME}\x. The product behavior is correct; the expectation only held on Unix. This is the same kind of bug as the path-form failures fixed in #1686.

Fix

Test-only change. The expected value is \${HOME}/x on Windows and /w/\${HOME}/x elsewhere, with a comment explaining why. No production code changes.

Evidence

• Windows, proven: the full-scope workflow_dispatch run 40070451767 on b3e8e4c is green: test (windows-latest, 1), test (windows-latest, 2) and ci-ok all pass. The same leg fails on main's nightly. The later push (50f5840) only rewords the comment.
• Lean PR CI on 50f5840: green (ci-ok, clippy, coverage, all e2e legs). Cursor Bugbot found no issues.
• Linux: cargo test -p socket-patch-core --lib global_folder_values_read_like_yarn passes. cargo fmt --all -- --check and cargo clippy --locked --workspace --all-features -- -D warnings are clean.

/code-review high findings

• Comment wording (it said the cwd's root is replaced, but Windows keeps the drive): fixed in 50f5840.
• Two findings were not taken: "use a platform-neutral input such as a\${HOME}/x instead of a cfg! branch", and "the Windows expectation comes from Path::join, not a yarn run". The current input is the case Fix yarn global folder moved by global-folder (#1629) #1696 checked against yarn global dir (escape at the start of the value), and its Windows result is now proven on a real Windows runner. Swapping inputs would change what the test covers and would need a second full-scope Windows run to prove. A real-drive Windows check of resolve_with_home would be a product-lane test, not a CI fix.

CI cost

No workflow change. A lean PR run is unchanged at about 28 Linux jobs, about 8 min wall-clock. This PR used one full-scope dispatch run because the expectation is Windows-specific.

https://claude.ai/code/session_01UCnLJqUYK9uEo1t22xjq2i

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

BugBot review

---

Generated by Claude Code

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 50f5840. Configure here.}

[Mikola Lysenko (mikolalysenko)]

Labeled Ready for review by the burn-down agent.

• Head: 50f5840145d80303a4944493f16dc9493df15762
• CI: 38/38 non-skipped checks green (50 skipped), no main-wide failures.
• Bugbot: reviewed this head, no findings.
• Mergeable against main, no CHANGELOG.md change.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Final review brief

What it does. Fixes the one failing Windows nightly leg (#1689). The global_folder_values_read_like_yarn test expected the escaped value \${HOME}/x to resolve to /w/\${HOME}/x on every OS; on Windows a leading \ is root-relative, so Path::join gives \${HOME}/x (same as Node's path.resolve). The expectation is now per-platform. Test-only; no production code changes.

Risk: low. One test file, 9 lines.

Look here

• yarnrc_tests.rs:592 – the cfg!(windows) expectation and the comment explaining it

Verified. Read the diff and the failing nightly log quoted in the description. The full-scope dispatch run 40070451767 on b3e8e4c is green on both Windows test legs (the later commit only rewords the comment). CI 38/38 non-skipped green on 50f5840; Cursor Bugbot: no issues; no review threads; CHANGELOG.md untouched.

Changes I made. None. Open questions. None.

Auto-merge is armed, so approving sends it straight to the merge queue.

---

Generated by Claude Code