[Mikola Lysenko (mikolalysenko)]

Problem

The nightly CI run on main is red. Both Windows test legs fail on every nightly since #1646 and #1658 merged (run 39907645034, head aa69125). The Windows test legs run only in the nightly (vars.CI_SCOPE == 'full' || schedule || workflow_dispatch), so neither PR CI nor the merge queue caught it. All other Windows coverage stays green.

job	failing target	test
test (windows-latest, 1)	socket-patch-core --lib	hosted::governing_root::tests::vlt_project_root_governs_unlisted_directories
test (windows-latest, 1)	socket-patch-cli --test user_patches	(log cut off by the API; same pattern, see below)
test (windows-latest, 2)	socket-patch-cli --test get	get_failure_paths::engine_nested_apply_preserves_conflicts_introduced_during_fetch

Root cause

In each failing test, the expected path is spelled differently from the one the code builds, though both point at the same file:

• governing_root: the refusal names the canonical root, \\?\C:\Users\runneradmin\...\vlt-lock.json. The test looks for tmp.path().join(...), and on the runners that is the 8.3 short form C:\Users\RUNNER~1\.... The panic shows both forms side by side. The sibling test vlt_project_root_overrides_the_listing_root already compares against canonicalize(tmp.path()).
• get_failure_paths (line 1660): the expected path is tmp.join("patches/x+1.0.0.patch"), which gives a mixed ...\patches/x... path on Windows. User-patch discovery joins patches and then the directory entry, so the reason contains ...\patches\x....
• user_patches: reason.contains("patches/example+1.0.0.patch") and stdout.contains("patches/example+1.0.0.patch") on JSON. On Windows the reason holds patches\example..., which JSON escapes as \\. The log of the user_patches target falls outside the roughly 5,000 lines the job-log API returns, so this cause is inferred from the code and the other two failures, not quoted from the log.

Fix

These are test-only changes. Each assertion still checks the same path:

• governing_root: compare against canonicalize(tmp.path()), as the sibling test does.
• get_failure_paths: join patches and the file name as separate components.
• user_patches: a slashed() helper normalizes \\ and \ to / before the relative-path contains checks. On Unix it changes nothing.

Proof

Linux, local:

• cargo test -p socket-patch-core --lib vlt_project_root_governs_unlisted_directories: ok
• cargo test -p socket-patch-cli --test get engine_nested_apply_preserves_conflicts_introduced_during_fetch: ok
• cargo test -p socket-patch-cli --test user_patches: 27 passed
• rustfmt --check on the touched files: clean. Clippy reports nothing in the touched files.

Windows, proven: workflow_dispatch run 39919335680 on this head (562fc93) passes both test (windows-latest, 1) and test (windows-latest, 2); both fail on main's nightly. The only red legs in that run are the old-bundler gem legs (#1685), which fail on main in the same way.

Not in this PR

The same nightly also fails e2e-extended gem legs (bundler 1.17.3 and 2.1.4, gem_hosted_unwind_after_unconverged_supersede_undoes_the_new_block). That one is a behaviour regression, not a path-form issue, and is filed as #1685.

No tests were removed or moved.

---

Generated by Claude Code

---

Note

Low Risk
Test-only assertion updates with no runtime or product behavior changes.

Overview
Fixes Windows-only nightly test failures where assertions compared path strings that referred to the same file but used different forms (8.3 short vs canonical paths, mixed \// separators, JSON-escaped backslashes).

governing_root: vlt_project_root_governs_unlisted_directories now asserts against canonicalize(tmp.path()) for stray-lock root and refusal messages, matching sibling tests.

get_failure_paths: Builds the expected user-patch path with separate join("patches") + filename so it matches discovery’s per-component paths on Windows.

user_patches: Adds slashed() to normalize \ and \\ to / before contains checks on skip reasons and --check stdout.

No production code changes; semantics of the tests are unchanged.

^{Reviewed by Cursor Bugbot for commit 562fc93. Configure here.}

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

bugbot run

---

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 562fc93. Configure here.}

[Mikola Lysenko (mikolalysenko)]

Started a workflow_dispatch CI run on this branch (head 562fc93) so the Windows test (windows-latest, *) legs actually run before merge: https://github.com/SocketDev/socket-patch/actions/runs/39919335680. If both Windows test legs pass there, the fix is proven. This PR is the only open fix for the red nightly Windows legs on main.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

The two failing e2e-extended (ubuntu-latest, e2e_redirect_gem_build e2e_vendor_gem_build, 3.1, 1.17.3 | 2.1.4) checks in dispatch run 39919335680 are not caused by this PR. Both legs fail the same way on main in the nightly (39907645034, head aa69125). The failing test is gem_hosted_unwind_after_unconverged_supersede_undoes_the_new_block, which came in with #1664. This PR changes only three test files, none of them gem-related. The failure is tracked in #1685, and no fix exists yet, so nothing is ported here. I did not re-run the checks because the failure reproduces on every run of main. What decides this PR is the two test (windows-latest, *) legs in the same run.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Ready for review (burn-down agent).

• Head: 562fc93
• CI: the PR run on this head is green (ci-ok success). The workflow_dispatch run 39919335680 passed both test (windows-latest, 1|2) legs, which is what this PR fixes. Its only failures are the two e2e-extended (ubuntu-latest, e2e_redirect_gem_build e2e_vendor_gem_build, 3.1, 1.17.3 | 2.1.4) legs and the ci-ok rollup. Those are main-wide: the same legs fail in the main nightly 39907645034 (Nightly: gem unwind-after-unconverged-supersede e2e fails on bundler 1.17.3/2.1.4 #1685, fix in Build: fix nightly gem unwind test on bundler < 2.2 (#1685) #1687).
• Bugbot: reviewed 562fc93, no findings. No open review threads.
• Mergeable, no conflicts; test-only change, CHANGELOG.md untouched.

Slack announcement not sent: this run had no Slack send tool.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Labeled Ready for review (burn-down agent).

• Head: 562fc93
• CI: PR run green (ci-ok success); the workflow_dispatch run that exercised the Windows legs passed them all. Its only red is e2e-extended (ubuntu-latest, e2e_redirect_gem_build e2e_vendor_gem_build, 3.1, 1.17.3), which is main-wide: it also fails on main's nightly 39907645034 (Nightly: gem unwind-after-unconverged-supersede e2e fails on bundler 1.17.3/2.1.4 #1685, fixed by Build: fix nightly gem unwind test on bundler < 2.2 (#1685) #1687)
• Bugbot: reviewed 562fc93, no issues; no open review threads
• Reviewer focus: all three files are test-only (expected paths canonicalized so 8.3 short temp paths compare equal), no production code changes.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

For the reviewer: this PR makes no non-test change. Every hunk in crates/socket-patch-core/src/hosted/governing_root.rs (lines 3289–3323) sits inside its #[cfg(test)] mod tests (from line 1267), in vlt_project_root_governs_unlisted_directories, so library behaviour is unchanged on every platform.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Final review brief

What it does. This is a test-only fix for the three Windows test legs that have been red on every nightly since #1646/#1658. Each assertion had spelled the expected path differently from the path the code builds: an 8.3 short temp dir instead of the canonical root, a mixed \// join, and JSON-escaped \\. The assertions now compare like with like. No product code changes.

Risk: low. Only test files and a #[cfg(test)] module change, and every assertion still checks the same path. The Windows fix is proven by the workflow_dispatch run 39919335680 on this head: both test (windows-latest, 1/2) legs pass.

Look here

• governing_root.rs:3289–3292 and :3323: compare against canonicalize(tmp), as the sibling test already does
• get_failure_paths.rs:1617–1622:``` join patches and the file name as separate components
• user_patches.rs:19–23: the slashed() helper, used at :96 and :232; it changes nothing on Unix

Verified

• Read the full diff. CHANGELOG.md is untouched, no tests were removed, and there's no debug code.
• PR CI on the head is green (ci-ok and clippy success). Bugbot: success.
• The head also carries a ci-ok failure, but it comes from that full-scope dispatch run: its only red legs are the old-bundler gem e2e-extended legs (Nightly: gem unwind-after-unconverged-supersede e2e fails on bundler 1.17.3/2.1.4 #1685, fixed separately in Build: fix nightly gem unwind test on bundler < 2.2 (#1685) #1687/Fix gem supersede on Bundler < 2.2 locks (#1688) #1690), which fail on main in the same way. The PR's own ci-ok is green.

Changes I made: none.

Open questions: none.

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

---

Generated by Claude Code