fix: recreate a prunable worktree instead of no-op'ing on create#107
Merged
Conversation
When a worktree dir was deleted out-of-band (a partial teardown, a manual `rm -rf`, an interrupted run) while its git admin record survived, `git worktree list` still reports the path — as "prunable". AddOrAttach's idempotency check matched that path and returned `(false, base, nil)`, so `bough create` reported the repo "already registered" and never recreated its dir. A `claude --worktree <name>` over such a worktree came up half-materialised: only the sub-repos whose dirs happened to survive were present, the rest silently missing. Confirm the dir exists on disk (os.Stat) before treating a path match as a materialised no-op. A registered-but-missing entry now falls through to the existing prune + add recovery and is re-materialised on its existing branch (attach path). Adds TestRunner_AddOrAttach_RecreatesPrunableWorktree (real git: create → rm -rf the dir → re-add must recreate on the same branch). Verified end-to-end against a real `bough create`: the shipped v0.13.0 binary leaves the deleted repo missing; the fixed binary re-attaches it.
…linked temp /review findings on the prunable-worktree fix: - gitwt.go: os.Stat gated recovery on statErr==nil, so a non-ENOENT stat failure (EACCES on a parent, a stale NFS mount) on a *present* worktree misclassified it as prunable and turned a should-be resume no-op into a hard error. Gate recovery on errors.Is(statErr, os.ErrNotExist) only. - gitwt_test.go: TestRunner_AddOrAttach_RecreatesPrunableWorktree did not guard the regression on macOS: t.TempDir() under /var -> /private/var meant the post-rm literal dst never matched git's resolved worktree-list path, so the loop took the no-match fall-through and never exercised the os.Stat guard (the guard could be reverted and the test still passed). Resolve the temp root with EvalSymlinks so the match -> stat-fails -> break path runs on every OS. Proven: reverting the guard now fails the test.
This was referenced Jul 16, 2026
ikeikeikeike
commented
Jul 16, 2026
| // momentarily-stale NFS mount) must keep the old no-op rather | ||
| // than tear down a worktree that is actually present — the | ||
| // intent is "the dir was deleted", not "the dir is unreadable". | ||
| if _, statErr := os.Stat(dst); !errors.Is(statErr, os.ErrNotExist) { |
Member
Author
There was a problem hiding this comment.
Retrospective review: this ENOENT-only guard only catches the "whole directory deleted" shape of prunability. git worktree list --porcelain also flags an entry prunable when the dir survives on disk but its own .git file (or git's admin-side gitdir record) was removed independently — verified against real git 2.53.0, which emits prunable gitdir file points to non-existent location for exactly that shape while os.Stat(dst) reports no error at all. So this guard silently no-ops on that case, reintroducing a variant of the half-materialised-worktree bug this PR fixed.
Filed as PR #119: adds a Prunable field to Worktree (parsed from the porcelain prunable line, which List() was discarding), and gates recovery on wt.Prunable || ENOENT instead of ENOENT alone — trusting git's own verdict while keeping the EACCES/NFS protection this commit added for the non-prunable case.
ikeikeikeike
added a commit
that referenced
this pull request
Jul 16, 2026
fix: retrospective review fixes for merged PR #107
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
When a worktree dir is deleted out-of-band (a partial teardown, a manual
rm -rf, an interrupted run) while its git admin record survives,git worktree liststill reports the path — as "prunable".AddOrAttach's idempotency check matched that path and returned(false, base, nil), sobough createreported the repo "already registered" and never recreated its dir. Aclaude --worktree <name>over such a worktree came up half-materialised — only the sub-repos whose dirs happened to survive were present.Confirm the dir exists on disk (
os.Stat) before treating a path match as a no-op. A registered-but-missing entry now falls through to the existing prune + add recovery and is re-materialised on its existing branch (attach path).go build/golangci-lint ./internal/gitwt/internal/ cmd/ pkg/)bough createbefore/afterNew test:
TestRunner_AddOrAttach_RecreatesPrunableWorktree(real git: create →rm -rfthe dir → re-add must recreate on the same branch).