* fix(sync): use sync-branch worktree for --full --no-pull (#1173)
Bug 1: PullFromSyncBranch was copying uncommitted worktree changes to
main repo when remoteAhead==0. This corrupted the 3-way merge because
local changes appeared as remote changes. Fixed by copying only the
committed state from HEAD instead of the working directory.
Bug 2: doExportOnlySync was checking main repo for changes via
gitHasBeadsChanges, but when sync-branch is configured, changes go to
the worktree, not main. Fixed by detecting sync-branch config and using
CommitToSyncBranch which operates on the worktree.
Fixes#1173
* refactor(sync): consolidate sync-branch detection and commit/push logic
Extract repeated patterns into reusable helpers:
- SyncBranchContext struct: holds branch name and repo root
- getSyncBranchContext(): detects sync-branch config from store
- commitAndPushBeads(): handles both sync-branch and regular git workflows
This eliminates duplicated sync-branch detection code (was in 3 places)
and the duplicated commit/push conditional logic (was in 2 places).
Net reduction of ~20 lines while improving maintainability.
* fix: remove unused bool return from commitAndPushBeads
Fixes#1076.
When git worktrees have sparse-checkout enabled, git add operations can
fail with 'outside of sparse-checkout cone' errors. Adding the --sparse
flag allows git to stage files outside the sparse-checkout cone.
Changes:
- Add --sparse flag to git add in daemon_sync_branch.go (line 165)
- Add --sparse flag to git add in migrate_sync.go (line 343)
- Add --sparse flag to git add in sync_branch.go (line 308)
- Add --sparse flag to git add in worktree.go (line 732)
All changes include comments referencing #1076 for traceability.
Centralizes repository context resolution via RepoContext API, fixing bugs where git commands run in the wrong repo when BEADS_DIR points elsewhere or in worktree scenarios.
Tests were failing because beads.FindDatabasePath() follows the
project's .beads/redirect file, causing tests to find unexpected
databases. Fixed by:
- Setting BEADS_DIR in tests that need isolation from git repo detection
- Clearing BEADS_DIR in TestMain to prevent global contamination
- Updating migration test schema to include owner column
This ensures tests work correctly in crew directories that have
redirect files pointing to shared .beads directories.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Executed-By: beads/crew/dave
Rig: beads
Role: crew
Two issues caused `bd migrate sync` to fail when run from a git worktree:
1. Used GetGitDir() instead of GetGitCommonDir() for worktree path
- GetGitDir() returns the worktree-specific path (.bare/worktrees/main)
- GetGitCommonDir() returns the shared git dir (.bare) where new
worktrees can actually be created
2. Used strings.Index instead of LastIndex in GetRepoRoot()
- When user paths contain "worktrees" (e.g., ~/Development/worktrees/),
Index finds the first occurrence and incorrectly strips the path
- LastIndex finds git's internal /worktrees/ directory
Added GetGitCommonDir() to internal/git/gitdir.go for reuse.
Fixes GH#639 (remaining unfixed callsite in migrate_sync.go)
In contributor mode (bd init --contributor), .beads/ is excluded in
.git/info/exclude to prevent committing upstream issue databases.
However, this exclusion applies to ALL worktrees, including the sync
worktree where .beads/ must be committed.
The fix adds -f flag to git add in commitInWorktree() to force-add
files even when gitignored. This follows the existing pattern in
beads where -f is used for worktree operations (git worktree add -f).
Fixes: bd sync failing with "git add failed in worktree: exit status 1"
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
Problem:
- Sync branch configuration was not consistently saved across storage types
- State divergence between database and config.yaml files
Solution:
- Update Set to write configuration to both yaml and database backends
- Add regression test for yaml configuration persistence
Impact:
- Guarantees configuration consistency across system reboots
- Prevents sync settings from being lost when reading from yaml
bd sync fails with exit status 128 when the daemon is started from a
terminal with different path casing than what git has stored. This
happens on macOS case-insensitive filesystem when directory names
are renamed (e.g., MyProject to myproject) but terminal sessions
retain the old casing.
The fix uses realpath(1) on macOS to get the true filesystem case
when canonicalizing paths:
- CanonicalizePath() now calls realpath on macOS
- git.GetRepoRoot() canonicalizes repoRoot via canonicalizeCase()
- syncbranch.GetRepoRoot() uses utils.CanonicalizePath()
This ensures git worktree paths match exactly, preventing the
exit status 128 errors from git operations.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Executed-By: beads/crew/dave
Rig: beads
Role: crew
- Add TestSet cases for main/master rejection in syncbranch_test.go
- Add TestInitWithSyncBranch to verify --branch flag works
- Add TestInitWithoutBranchFlag to verify no auto-detection (root cause)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This implements sync branch integrity guards that detect when the remote
sync branch has been force-pushed since the last sync, preventing silent
data corruption.
Changes:
- Add internal/syncbranch/integrity.go with:
- CheckForcePush() - detects force-push via stored remote SHA comparison
- UpdateStoredRemoteSHA() - stores current remote SHA after successful sync
- ClearStoredRemoteSHA() - clears stored SHA when resetting
- GetStoredRemoteSHA() - retrieves stored SHA for inspection
- Update cmd/bd/sync.go to:
- Add --accept-rebase flag for non-interactive reset to remote
- Add force-push detection before sync branch pull operations
- Prompt user for confirmation when force-push detected
- Update stored remote SHA after successful sync
The implementation:
1. Tracks the remote sync branch commit SHA in config after each sync
2. On subsequent syncs, checks if stored SHA is ancestor of current remote
3. If not (force-push detected), warns user with details and prompts
4. User can accept reset or abort to investigate manually
5. --accept-rebase flag allows scripted/non-interactive recovery
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Two issues fixed:
1. `bd init` was auto-detecting current branch (e.g., main) as sync.branch
when no --branch flag was specified. This caused worktree conflicts.
2. Added validation to reject main/master as sync.branch values.
When sync.branch is set to main, the worktree mechanism creates a checkout
of main at .git/beads-worktrees/main/, which prevents git checkout main
from working in the user's working directory.
The sync branch feature should use a dedicated branch like 'beads-sync'.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Executed-By: beads/crew/dave
Rig: beads
Role: crew
Code review caught that ".beads" would incorrectly match prefixes like
".beads-backup". Changed to match ".beads/" (with trailing slash) to
ensure we only match the actual .beads directory.
Added test cases for this edge case.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The bug: In a bare repo + worktrees setup, jsonlRelPath was calculated
relative to the project root (which contains all worktrees), resulting in
paths like "main/.beads/issues.jsonl". But the sync branch worktree uses
sparse checkout for .beads/*, so files are at ".beads/issues.jsonl".
This caused copyJSONLToMainRepo to look in the wrong location, silently
returning when the file was not found.
Fix: Add normalizeBeadsRelPath() to strip leading path components before
".beads", ensuring correct path resolution in both directions:
- copyJSONLToMainRepo (worktree -> local)
- SyncJSONLToWorktreeWithOptions (local -> worktree)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Strip (bd-xxx), (gt-xxx) suffixes from code comments and changelog
entries. The descriptions remain meaningful without the ephemeral
issue IDs.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Analysis found these commands are dead code:
- gt never calls `bd pin` - uses `bd update --status=pinned` instead
- Beads.Pin() wrapper exists but is never called
- bd hook functionality duplicated by gt mol status
- Code comment says "pinned field is cosmetic for bd hook visibility"
Removed:
- cmd/bd/pin.go
- cmd/bd/unpin.go
- cmd/bd/hook.go
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Modern git (2.28+) uses 'main' as default branch, not 'master'.
Tests were failing because they assumed 'master' branch exists.
Changes:
- Use 'git init --initial-branch=main' instead of bare 'git init'
- Change 'git checkout master' to 'git checkout main'
- Add git.ResetCaches() after os.Chdir() to clear cached git state
- Ensures test isolation when changing directories
Replace fetchAndRebaseInWorktree with contentMergeRecovery in pushFromWorktree.
The problem: When push fails due to non-fast-forward, the old code used git
rebase to recover. But git rebase is text-level and does not invoke the JSONL
merge driver. This could resurrect tombstones - if remote had a tombstone and
local had closed, the rebase would overwrite the tombstone.
The fix: Use the same content-level merge algorithm that PullFromSyncBranch
uses. This respects tombstone semantics - recent tombstones always win over
live issues.
Changes:
- Add contentMergeRecovery() that does content-level merge instead of rebase
- Update pushFromWorktree to call contentMergeRecovery
- Mark fetchAndRebaseInWorktree as deprecated (kept for reference)
- Add tests for tombstone preservation during merge recovery
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The error message 'path exists but is not a valid git worktree' was appearing
in daemon.log when the daemon attempted to use an existing worktree that was
in the git worktree list but had other issues (broken sparse checkout, etc.).
Root cause:
- CreateBeadsWorktree only checked isValidWorktree (is it in git worktree list)
- CheckWorktreeHealth was called separately and checked additional things
- If the worktree passed isValidWorktree but failed health check, an error
was logged and repair was attempted
Fix:
- CreateBeadsWorktree now performs a full health check when it finds an
existing worktree that's in the git worktree list
- If the health check fails, it automatically removes and recreates the
worktree
- Removed redundant CheckWorktreeHealth calls in daemon_sync_branch.go and
syncbranch/worktree.go since CreateBeadsWorktree now handles this internally
This eliminates the confusing error message and ensures worktrees are always
in a healthy state after CreateBeadsWorktree returns successfully.
Add done channel to runGitCmdWithTimeoutMsg and runCmdWithTimeoutMessage
so goroutines exit immediately when command completes, rather than
waiting for the full timeout duration.
Follow-up to PR #678.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The timerChan and timeoutChan variables were created but never read from,
causing lint errors. Removed them since we don't need channel synchronization -
we only want the timeout to trigger a message, not block anything.
When git push or other git operations hang waiting for credential/browser
auth, show a helpful message to the user after 5 seconds of inactivity
instead of appearing frozen.
Added:
- runCmdWithTimeoutMessage() in internal/syncbranch/worktree.go
- runGitCmdWithTimeoutMsg() in cmd/bd/sync.go
- Both functions print a message after timeout delay with advice
to check for browser auth prompts
Fixes issue #647 (bd sync frozen waiting for browser auth)
* test(doctor): add comprehensive tests for fix and check functions
Add edge case tests, e2e tests, and improve test coverage for:
- database_test.go: database integrity and sync checks
- git_test.go: git hooks, merge driver, sync branch tests
- gitignore_test.go: gitignore validation
- prefix_test.go: ID prefix handling
- fix/fix_test.go: fix operations
- fix/e2e_test.go: end-to-end fix scenarios
- fix/fix_edge_cases_test.go: edge case handling
* docs: add testing philosophy and anti-patterns guide
- Create TESTING_PHILOSOPHY.md covering test pyramid, priority matrix,
what NOT to test, and 5 anti-patterns with code examples
- Add cross-reference from README_TESTING.md
- Document beads-specific guidance (well-covered areas vs gaps)
- Include target metrics (test-to-code ratio, execution time targets)
* chore: revert .beads/ to upstream/main state
* refactor(doctor): add category grouping and Ayu theme colors
- Add Category field to DoctorCheck for organizing checks by type
- Define category constants: Core, Git, Runtime, Data, Integration, Metadata
- Update thanks command to use shared Ayu color palette from internal/ui
- Simplify test fixtures by removing redundant test cases
* fix(doctor): prevent test fork bomb and fix test failures
- Add ErrTestBinary guard in getBdBinary() to prevent tests from
recursively executing the test binary when calling bd subcommands
- Update claude_test.go to use new check names (CLI Availability,
Prime Documentation)
- Fix syncbranch test path comparison by resolving symlinks
(/var vs /private/var on macOS)
- Fix permissions check to use exact comparison instead of bitmask
- Fix UntrackedJSONL to use git commit --only to preserve staged changes
- Fix MergeDriver edge case test by making both .git dir and config
read-only
- Add skipIfTestBinary helper for E2E tests that need real bd binary
* test(doctor): skip read-only config test in CI environments
GitHub Actions containers may have CAP_DAC_OVERRIDE or similar
capabilities that allow writing to read-only files, causing
the test to fail. Skip the test when CI=true or GITHUB_ACTIONS=true.
* fix(syncbranch): support bare repos and worktrees with git-common-dir
Replace hardcoded .git/beads-worktrees/ path with dynamic detection using
git rev-parse --git-common-dir. This correctly handles:
- Regular repositories (.git is a directory)
- Git worktrees (.git is a file pointing elsewhere)
- Bare repositories (no .git directory, repo IS the git dir)
- Worktrees of bare repositories
The new getBeadsWorktreePath() helper uses git's native API to find the
shared git directory, ensuring beads worktrees are created in the correct
location regardless of repository structure.
Updated functions:
- CommitToSyncBranch
- PullFromSyncBranch
- CheckDivergence
- ResetToRemote
Fixes#639
* test(syncbranch): add regression tests for getBeadsWorktreePath
Add comprehensive tests for the worktree path calculation to ensure proper
handling of various repository structures:
- Regular repos: uses .git/beads-worktrees path
- Bare repos: uses <bare-repo>/beads-worktrees (no .git subdirectory)
- Worktrees: uses main repo's .git/beads-worktrees (git-common-dir)
- Fallback: legacy behavior when git command fails
- Relative paths: ensures absolute path conversion
These tests ensure the fix for GH#639 doesn't regress.
---------
Co-authored-by: Charles P. Cross <cpdata@users.noreply.github.com>
When bd sync pushes from the sync-branch worktree, the pre-push hook
would detect uncommitted JSONL changes and suggest running bd sync -
which is circular since that's what the user is already doing.
Fix: Set BD_SYNC_IN_PROGRESS=1 environment variable when pushing from
worktree, and update pre-push hook to skip checks when this var is set.
Updated files:
- internal/syncbranch/worktree.go: Set env var on push command
- cmd/bd/templates/hooks/pre-push: Check for env var and exit early
- examples/git-hooks/pre-push: Same check
- .beads-hooks/pre-push: Same check
When sync branch diverges significantly from remote, provide clear
recovery options instead of a confusing rebase conflict error.
- Add CheckDivergence() to detect and report sync branch divergence
- Add ResetToRemote() to reset local sync branch to remote state
- Add --reset-remote and --force-push flags for recovery
- Improve error message when rebase fails to include recovery steps
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* feat: auto-disable daemon in git worktrees for safety
Implement worktree daemon compatibility as proposed in the analysis.
The daemon is now automatically disabled when running in a git worktree
unless sync-branch is configured.
Git worktrees share the same .beads directory, and the daemon commits
to whatever branch its working directory has checked out. This causes
commits to go to the wrong branch when using daemon in worktrees.
- Add shouldDisableDaemonForWorktree() helper that checks:
1. If current directory is a git worktree (via git rev-parse)
2. If sync-branch is configured (env var or config.yaml)
- Modify shouldAutoStartDaemon() to call the helper
- Modify daemon connection logic in main.go to skip connection
- Add FallbackWorktreeSafety constant for daemon status reporting
- Update warnWorktreeDaemon() to skip warning when sync-branch configured
- In worktree WITHOUT sync-branch: daemon auto-disabled, direct mode used
- In worktree WITH sync-branch: daemon enabled (commits go to dedicated branch)
- In regular repo: no change (daemon works as before)
- Added comprehensive unit tests for shouldDisableDaemonForWorktree()
- Added integration tests for shouldAutoStartDaemon() in worktree contexts
- Manual E2E testing verified correct behavior
- Updated WORKTREES.md with new automatic safety behavior
- Updated DAEMON.md with Git Worktrees section
* feat: check database config for sync-branch in worktree safety logic
Previously, the worktree daemon safety check only looked at:
- BEADS_SYNC_BRANCH environment variable
- sync-branch in config.yaml
This meant users who configured sync-branch via `bd config set sync-branch`
(which stores in the database) would still have daemon disabled in worktrees.
Now the check also reads sync.branch from the database config table,
making daemon work in worktrees when sync-branch is configured via any method.
Changes:
- Add IsConfiguredWithDB() function that checks env, config.yaml, AND database
- Add findBeadsDB() to locate database (worktree-aware via git-common-dir)
- Add getMainRepoRoot() helper using git rev-parse
- Add getConfigFromDB() for lightweight database reads
- Update shouldDisableDaemonForWorktree() to use IsConfiguredWithDB()
- Update warnWorktreeDaemon() to use IsConfiguredWithDB()
- Add test case for database config path
* refactor: use existing beads.FindDatabasePath() instead of duplicating code
Remove duplicate getMainRepoRoot() and findBeadsDB() functions from
syncbranch.go and use the existing beads.FindDatabasePath() which is
already worktree-aware.
Changes:
- Replace custom findBeadsDB() with beads.FindDatabasePath()
- Remove duplicate getMainRepoRoot() (git.GetMainRepoRoot() exists)
- Remove unused imports (exec, strings, filepath)
- Clean up debug logging in tests
---------
Co-authored-by: Charles P. Cross <cpdata@users.noreply.github.com>
When sync.branch is configured to the same branch as the current branch,
git worktree creation fails because the same branch cannot be checked out
in multiple locations.
This fix detects when sync.branch equals the current branch and falls back
to direct commits on the current branch instead of using the worktree-based
approach.
Changes:
- Add IsSyncBranchSameAsCurrent() helper in syncbranch package
- Add GetCurrentBranch() helper function
- Update sync.go to detect this case and skip worktree operations
- Add unit tests for the new functionality
Closes#519🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When multiple polecats run bd sync simultaneously, they race to push to
the shared beads-sync branch. Previously this caused non-fast-forward
errors that blocked all polecats.
Now pushFromWorktree:
- Detects non-fast-forward errors from git push output
- On conflict: fetches remote, rebases local commits on top, retries
- Uses exponential backoff for transient failures (up to 5 retries)
- Aborts rebase cleanly if it fails to leave worktree in good state
Fixes gt-zqor.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Adds comprehensive Git worktree support for beads issue tracking:
Core changes:
- New internal/git/gitdir.go package for worktree detection
- GetGitDir() returns proper .git location (main repo, not worktree)
- Updated all hooks to use git.GetGitDir() instead of local helper
- BeadsDir() now prioritizes main repository's .beads directory
Features:
- Hooks auto-install in main repo when run from worktree
- Shared .beads directory across all worktrees
- Config option no-install-hooks to disable auto-install
- New bd worktree subcommand for diagnostics
Documentation:
- New docs/WORKTREES.md with setup instructions
- Updated CHANGELOG.md and AGENT_INSTRUCTIONS.md
Testing:
- Updated tests to use exported git.GetGitDir()
- Added worktree detection tests
Co-authored-by: Claude <noreply@anthropic.com>
Closes: #478
- Fix gosec G204/G304 warnings by adding exclusions for safe subprocess
launches and file reads in doctor.go, jira.go, migrate_sync.go, and
syncbranch/worktree.go
- Fix misspell: "cancelled" -> "canceled" in sync.go
- Fix unparam: mark unused ctx params in jira.go placeholder functions
- Fix errcheck: explicitly ignore fmt.Sscanf return in doctor.go and
use closure pattern for deferred os.RemoveAll in worktree.go
- Increase Windows test timeout from 10m to 20m to prevent CI timeouts
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- bd-dtm: Changed stderr printing to use SafetyWarnings in worktree.go
- bd-ciu: Fixed non-deterministic output order in formatVanishedIssues
- bd-dmd: Removed duplicate safety check message in sync.go
- bd-k2n: PushSyncBranch now recreates worktree if cleaned up
- bd-c5m: Fixed string(rune()) in tests to use strconv.Itoa
- bd-8uk: Added test for SafetyWarnings population
- bd-1kf: Fixed mergePriority to handle negative priorities
- bd-xo9: Documented sync.require_confirmation_on_mass_delete config
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Add auto-push functionality to PullFromSyncBranch for true one-command sync:
- After successful content merge, auto-push to remote by default
- Safety check: warn (but dont block) if >50% issues vanished AND >5 existed
- Vanished = removed from JSONL entirely, NOT status=closed
Changes:
- Add push parameter to PullFromSyncBranch function
- Add Pushed field to PullResult struct
- Add countIssuesInContent helper for safety check
- Add test for countIssuesInContent function
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
When multiple clones commit to beads-sync branch and histories diverge,
git merge would fail. This replaces git's commit-level merge with a
content-based merge that extracts JSONL from base/local/remote and
merges at the semantic level.
Key changes:
- Add divergence detection using git rev-list --left-right
- Extract JSONL content from specific commits for 3-way merge
- Reset to remote's history then commit merged content on top
- Pre-emptive fetch before commit to reduce divergence likelihood
- Deletions.jsonl merged by union (keeps all deletions)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
