chore(windows): build and pass the test suite natively on Windows (#1501) #1503

Closed

claude wants to merge 1 commit from claude/1501-windows-native-tests into main

pull from: claude/1501-windows-native-tests

merge into: frankenbit:main

frankenbit:main

frankenbit:i/336-sha-pin-propagation

frankenbit:sentry/1342-approval-whitelist-presence

Conversation 1 Commits 1 Files changed 14 +174 -25

claude commented 2026-10-02 11:35:09 +02:00

Member

• Copy link
• Copy Markdown

Refs #1501

Head 7163298 on base 55a12b9 (live main at push time).

Why

#1485 gave Windows contributors an LF checkout, but a native go test ./... still fails there. cmd/rt and internal/release do not compile (syscall.Flock). rt register-check never matches an allow-list directory against a Windows path. Several tests assume POSIX permissions, an executable bit, glibc regex, or Unix path forms. Release binaries stay linux/amd64, and Linux behaviour does not change.

What's in the PR

Cut lock (internal/release): real fix

• lock.go keeps the shared lifecycle (acquireLock / release). The OS primitive moves to lock_flock.go (//go:build !windows, flock as before) and lock_windows.go (LockFileEx, LOCKFILE_EXCLUSIVE_LOCK|LOCKFILE_FAIL_IMMEDIATELY, whole-file range). ERROR_LOCK_VIOLATION maps to ErrConcurrentCut.
• The tag is !windows rather than unix, so the flock file builds for exactly the targets the untagged lock.go built for.
• LockFileEx is resolved from kernel32 through stdlib syscall, without adding golang.org/x/sys. That is right while Windows is contributor-only. x/sys/windows becomes right if Windows release runners land (Phase 8) or more Win32 calls accumulate.

register.ScanTree: real fix

• scanOneFile matches allow-list and self-exclusion entries, and reports FileHit.File, in filepath.ToSlash form; it still reads through the native path. ToSlash is the identity on Linux, so register-check output there is byte-identical.

Windows skips, in the existing t.Skip reason style (the tests still run on Linux)

• TestDiscoverUnreadableIsUngradeable, and TestWritePostCutManifestCustomTagFormat/config_is_UNREADABLE_by_mode_is_could_not_grade. On Windows the latter was green for the wrong reason (see #1501).
• TestRunPostBumpHooks_theHookDoesNotReceiveTheGitCredential: Windows has no executable bit and cannot exec #!/bin/sh.
• TestRepinFile_PreservesModeAndSkipsNoMatch: only the mode half is ungraded on Windows (t.Log); the unmatched-file half still runs.
• TestHeaderEmDashLocaleAndByteClassMutation: the live-pattern subtests run on Windows; only the mutant half skips, because MSYS regex rejects [-—] under C.utf8 too.

cmd/rt test-side fixes, not skips

• TestChangelogBodyCheckBadThreshold builds rt.exe on Windows.
• TestFragmentCheckSummaryWordWarnExecutes and TestFragmentCheckLengthWarnCountsRunes now hand the gate a relative dir via a new relFragDir, because a C:\ path correctly takes the safe-char stderr fallback. The expected path uses /, as the gate builds it.
• TestPrepStagesTheRegeneratedTwinIntoTheIndex uses path.Join for git index keys.

TestPushManifestAuthedRejectsSecondPush: the deadline goes from 2s to 30s. The retry has no backoff, and the "xx" attempt counter stays as the discriminating assertion. Under load, the 2s deadline cancelled attempt 2 before its hook ran, and the error text was the same.

Fragment: changelog.d/1501-windows-native-tests.internal.md (patch lane; nothing a consumer can observe).

Acceptance criteria (#1501)

• internal/release and cmd/rt build on Windows; the Windows backend returns ErrConcurrentCut in-process and cross-process, and frees the lock when the holder is killed. TestNewCutter_concurrentCut is green on Windows and goes red with tryLockFile neutered. A throwaway cross-process probe (not committed) showed: the second process gets ErrConcurrentCut, the lock is free immediately after the holder is killed, and a double release is a no-op.
• On Windows, ScanTree matches and reports in / form, and Linux output is unchanged. TestScanTree is red without ToSlash and green with it.
• Tests needing POSIX mode bits, an executable bit or glibc regex skip on Windows with a reason, and still run on Linux. With -v: SKIP on Windows, PASS on Linux (non-root) for each.
• The four cmd/rt path and .exe assumptions are fixed in the tests, not skipped.
• Native Windows go test ./... passes on an LF checkout, and golang:1.25 stays green as non-root (see Test plan).

Test plan

• Native Windows (Go 1.27.1), fresh core.autocrlf=true clone of 55a12b9 checked out at 7163298, go test -count=1 ./...: 29/29 ok.
• golang:1.25, same commit, as a non-root user so the existing root skips do not fire: 29/29 ok.
• go vet ./... for windows, linux and darwin; gofmt -l (Go 1.25) clean; FRAGMENT_CHECK_COVERAGE=content rt fragment-check changelog.d PASS.
• Reviewer: CI covers the Linux half. The Windows half needs a Windows checkout: go test ./....

Review coordination

• Author: announce HOLD in the PR thread before requesting or starting an
  approval read if more commits or a PR-body correction is still expected.
• Reviewer: wait for that explicit HOLD before approving. A
  REQUEST_CHANGES asks the author to fix the PR; it does not freeze the
  author from pushing.
• Reviewer: immediately before submitting a review, run
  rt review-submit-check --owner OWNER --repo REPO --pr N --reviewed-head SHA
  with the full 40-character server head that was read. rc=0 means only that
  the head still matches; rc=1 requires a fresh read, and rc=2 means the
  check could not grade. This does not check approval eligibility or review
  rows, and it does not submit the review. Omit commit_id when submitting,
  name the exact reviewed SHA in the review body, and read the returned row
  back.
• Author: after a body-only correction, announce it to the reviewer and obtain
  a fresh read; the head and a human rejection do not move or clear by
  themselves.

What this PR does NOT do

• Windows release binaries or release runners.
• Strengthen the post-cut 0o000 arm to assert the could-not-read message (out of scope in #1501; not measured on Linux).
• A Linux-discriminating test for the ScanTree fix. On Linux a backslash is a filename byte and ToSlash is the identity, so only a Windows run can catch a regression there.
• Reformat cmd/rt/mirror_drift_check_test.go. Go 1.27's gofmt rewrites its indentation and Go 1.25's does not; this PR does not touch the file.

🤖 Generated with Claude Code

Refs #1501 Head `7163298` on base `55a12b9` (live `main` at push time). ## Why #1485 gave Windows contributors an LF checkout, but a native `go test ./...` still fails there. `cmd/rt` and `internal/release` do not compile (`syscall.Flock`). `rt register-check` never matches an allow-list directory against a Windows path. Several tests assume POSIX permissions, an executable bit, glibc regex, or Unix path forms. Release binaries stay linux/amd64, and Linux behaviour does not change. ## What's in the PR **Cut lock (`internal/release`): real fix** - `lock.go` keeps the shared lifecycle (`acquireLock` / `release`). The OS primitive moves to `lock_flock.go` (`//go:build !windows`, flock as before) and `lock_windows.go` (`LockFileEx`, `LOCKFILE_EXCLUSIVE_LOCK|LOCKFILE_FAIL_IMMEDIATELY`, whole-file range). `ERROR_LOCK_VIOLATION` maps to `ErrConcurrentCut`. - The tag is `!windows` rather than `unix`, so the flock file builds for exactly the targets the untagged `lock.go` built for. - `LockFileEx` is resolved from kernel32 through stdlib `syscall`, without adding `golang.org/x/sys`. That is right while Windows is contributor-only. `x/sys/windows` becomes right if Windows release runners land (Phase 8) or more Win32 calls accumulate. **`register.ScanTree`: real fix** - `scanOneFile` matches allow-list and self-exclusion entries, and reports `FileHit.File`, in `filepath.ToSlash` form; it still reads through the native path. `ToSlash` is the identity on Linux, so register-check output there is byte-identical. **Windows skips, in the existing `t.Skip` reason style (the tests still run on Linux)** - `TestDiscoverUnreadableIsUngradeable`, and `TestWritePostCutManifestCustomTagFormat/config_is_UNREADABLE_by_mode_is_could_not_grade`. On Windows the latter was green for the wrong reason (see #1501). - `TestRunPostBumpHooks_theHookDoesNotReceiveTheGitCredential`: Windows has no executable bit and cannot exec `#!/bin/sh`. - `TestRepinFile_PreservesModeAndSkipsNoMatch`: only the mode half is ungraded on Windows (`t.Log`); the unmatched-file half still runs. - `TestHeaderEmDashLocaleAndByteClassMutation`: the live-pattern subtests run on Windows; only the mutant half skips, because MSYS regex rejects `[-—]` under `C.utf8` too. **`cmd/rt` test-side fixes, not skips** - `TestChangelogBodyCheckBadThreshold` builds `rt.exe` on Windows. - `TestFragmentCheckSummaryWordWarnExecutes` and `TestFragmentCheckLengthWarnCountsRunes` now hand the gate a relative dir via a new `relFragDir`, because a `C:\` path correctly takes the safe-char stderr fallback. The expected path uses `/`, as the gate builds it. - `TestPrepStagesTheRegeneratedTwinIntoTheIndex` uses `path.Join` for git index keys. **`TestPushManifestAuthedRejectsSecondPush`**: the deadline goes from 2s to 30s. The retry has no backoff, and the `"xx"` attempt counter stays as the discriminating assertion. Under load, the 2s deadline cancelled attempt 2 before its hook ran, and the error text was the same. **Fragment**: `changelog.d/1501-windows-native-tests.internal.md` (patch lane; nothing a consumer can observe). ## Acceptance criteria (#1501) - [x] `internal/release` and `cmd/rt` build on Windows; the Windows backend returns `ErrConcurrentCut` in-process and cross-process, and frees the lock when the holder is killed. `TestNewCutter_concurrentCut` is green on Windows and goes red with `tryLockFile` neutered. A throwaway cross-process probe (not committed) showed: the second process gets `ErrConcurrentCut`, the lock is free immediately after the holder is killed, and a double `release` is a no-op. - [x] On Windows, `ScanTree` matches and reports in `/` form, and Linux output is unchanged. `TestScanTree` is red without `ToSlash` and green with it. - [x] Tests needing POSIX mode bits, an executable bit or glibc regex skip on Windows with a reason, and still run on Linux. With `-v`: SKIP on Windows, PASS on Linux (non-root) for each. - [x] The four `cmd/rt` path and `.exe` assumptions are fixed in the tests, not skipped. - [x] Native Windows `go test ./...` passes on an LF checkout, and `golang:1.25` stays green as non-root (see Test plan). ## Test plan - [x] Native Windows (Go 1.27.1), fresh `core.autocrlf=true` clone of `55a12b9` checked out at `7163298`, `go test -count=1 ./...`: 29/29 ok. - [x] `golang:1.25`, same commit, as a non-root user so the existing root skips do not fire: 29/29 ok. - [x] `go vet ./...` for windows, linux and darwin; `gofmt -l` (Go 1.25) clean; `FRAGMENT_CHECK_COVERAGE=content rt fragment-check changelog.d` PASS. - [ ] Reviewer: CI covers the Linux half. The Windows half needs a Windows checkout: `go test ./...`. ## Review coordination - Author: announce `HOLD` in the PR thread before requesting or starting an approval read if more commits or a PR-body correction is still expected. - Reviewer: wait for that explicit `HOLD` before approving. A `REQUEST_CHANGES` asks the author to fix the PR; it does not freeze the author from pushing. - Reviewer: immediately before submitting a review, run `rt review-submit-check --owner OWNER --repo REPO --pr N --reviewed-head SHA` with the full 40-character server head that was read. `rc=0` means only that the head still matches; `rc=1` requires a fresh read, and `rc=2` means the check could not grade. This does not check approval eligibility or review rows, and it does not submit the review. Omit `commit_id` when submitting, name the exact reviewed SHA in the review body, and read the returned row back. - Author: after a body-only correction, announce it to the reviewer and obtain a fresh read; the head and a human rejection do not move or clear by themselves. ## What this PR does NOT do - Windows release binaries or release runners. - Strengthen the post-cut `0o000` arm to assert the could-not-read message (out of scope in #1501; not measured on Linux). - A Linux-discriminating test for the `ScanTree` fix. On Linux a backslash is a filename byte and `ToSlash` is the identity, so only a Windows run can catch a regression there. - Reformat `cmd/rt/mirror_drift_check_test.go`. Go 1.27's gofmt rewrites its indentation and Go 1.25's does not; this PR does not touch the file. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

claude added the

kind

bug

label 2026-10-02 11:35:09 +02:00

claude added 1 commit 2026-10-02 11:35:09 +02:00

chore(windows): build and pass the test suite natively on Windows (#1501) ... 7163298262

The cut lock moves its OS primitive behind a build tag: lock_flock.go
(!windows, unchanged flock) and lock_windows.go (LockFileEx with
FAIL_IMMEDIATELY). Contention still surfaces as ErrConcurrentCut,
in-process and cross-process, and the OS frees the lock when the holder
dies.

register.ScanTree matches allow-list and self-exclusion entries, and
reports hits, in slash form. On Windows the allow-list directory never
matched a WalkDir path; on Linux ToSlash is the identity.

Tests that need POSIX mode bits, an executable bit, or glibc regex skip
on Windows with a reason, and still run on Linux. The four cmd/rt tests
that assumed a bare "rt" binary, an absolute path inside the annotation
safe-char set, or backslash git index keys are fixed instead. The
second-push test's 2s deadline becomes a 30s hang guard: under load it
cancelled attempt 2 before the hook ran.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

claude referenced this pull request 2026-10-02 22:25:09 +02:00

bug(windows): native go test ./... still fails on an LF checkout — cmd/rt does not build, and POSIX-only tests run unguarded #1501

claude commented 2026-10-02 22:31:40 +02:00

Author

Member

• Copy link
• Copy Markdown

Superseded by #1505, which landed as 96d2646 and covers this PR's scope and more (committed cross-process and killed-holder lock tests, graded 0o000 arms on Windows, a by-name post_bump_hooks refusal). #1501 is ticked against that landing. Shutting this unmerged and deleting claude/1501-windows-native-tests.

Superseded by #1505, which landed as `96d2646` and covers this PR's scope and more (committed cross-process and killed-holder lock tests, graded `0o000` arms on Windows, a by-name `post_bump_hooks` refusal). #1501 is ticked against that landing. Shutting this unmerged and deleting `claude/1501-windows-native-tests`.

claude closed this pull request 2026-10-02 22:31:40 +02:00

claude referenced this pull request 2026-10-02 22:34:52 +02:00

bug(release): rt does not build on Windows — the cut lock calls syscall.Flock, and register-check ignores the allow-list there #1502

Some checks are pending

Hide checks

fork-pr-approval-notice / explain fork workflow approval (pull_request_target) Successful in 5s

Details

ac-closure-check / ac-closure check (pull_request) Successful in 7s

Details

ac-closure-check / check (pull_request) Successful in 0s

Details

base-divergence-check / check (pull_request) Successful in 6s

Details

changelog-body-check / changelog body Cold-Read linter (pull_request) Successful in 7s

Details

changelog-body-check / check (pull_request) Successful in 0s

Details

check-self-bootstrap / check (pull_request) Successful in 5s

Details

fragment-check / coverage (pull_request) Successful in 7s

Details

fragment-check / density (pull_request) Successful in 7s

Details

fragment-check / check (pull_request) Successful in 0s

Details

gitea-twin-check / check (pull_request) Successful in 6s

Details

go-ci / lint + build + test (pull_request) Successful in 31s

Required

Details

go-ci / record reviewed vs landed commit (pull_request) Has been skipped

Details

manifest-check / manifest-vs-tag consistency (pull_request) Successful in 7s

Details

manifest-check / check (pull_request) Successful in 0s

Details

prep-order-check / check (pull_request) Successful in 7s

Details

readme-pin-check / mirror (pull_request) Successful in 8s

Details

readme-pin-check / digest (pull_request) Successful in 6s

Details

register-check / register-drift check (pull_request) Successful in 7s

Details

register-check / check (pull_request) Successful in 0s

Details

tests / workflow-schema (pull_request) Successful in 4s

Details

tests / bats (pull_request) Successful in 32s

Required

Details

tests / dated-examples (pull_request) Successful in 5s

Details

tests / contract-paths (pull_request) Successful in 4s

Details

tests / shellcheck (pull_request) Successful in 4s

Details

toolkit-self-gates / toolkit-self gates (PR's own rt) (pull_request) Successful in 7s

Required

Details

workflow-parse-check / toolkit-self parse guard and controls (pull_request) Successful in 4s

Details

workflow-parse-check / workflow parse and schema (pull_request) Successful in 4s

Details

workflow-parse-check / check (pull_request) Successful in 0s

Details

go-ci / page landing-tree failure (pull_request) Has been skipped

Details

pr / gates (pull_request)

Required

tests / checks (pull_request)

Required

Pull request closed

This pull request cannot be reopened because the branch was deleted.

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels   

bump

major

Override the cut bump to MAJOR (release-toolkit#107)

  

bump

minor

Override the cut bump to MINOR (release-toolkit#107)

  

bump

patch

Override the cut bump to PATCH (release-toolkit#107)

  

kind

bug

Something is broken

  

kind

chore

Maintenance, tooling, refactor

  

kind

docs

Documentation work

  

kind

epic

Tracker: its acceptance criteria are sub-issues; closed by hand when all are ticked

  

kind

feature

New feature or capability

  

loop/paused

rt-bugfix-loop kill switch: while this is on the loop log issue, no round claims or merges

  

mirror-drift

Automated: a release here has no asset-bearing counterpart on the mirror. One tracker per drift episode; the check never reopens a closed one (#1586)

  

priority

critical

Critical: must be addressed immediately

  

priority

high

Important, should be addressed soon

  

priority

low

Low priority, address when convenient

  

priority

medium

Normal priority

  

size

L

Large: significant feature, ~6-16h

  

size

M

Medium: moderate feature, ~2-6h

  

size

S

Small: simple change, ~1-2h

  

size

XL

Extra large: major feature or refactor, 16h+

  

status

blocked

Cannot start until a named dependency resolves

No labels

bump

major

bump

minor

bump

patch

kind

bug

kind

chore

kind

docs

kind

epic

kind

feature

loop/paused

mirror-drift

priority

critical

priority

high

priority

low

priority

medium

size

L

size

M

size

S

size

XL

status

blocked

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

frankenbit/release-toolkit!1503

No description provided.