bd sync: 2025-10-26 19:42:46

.beads/bd.jsonl

@@ -66,6 +66,7 @@
 {"id":"bd-158","title":"Implement \"bd daemons list\" subcommand","description":"Create the \"bd daemons list\" command that displays all running daemons in a table with: workspace path, PID, version, socket path, uptime, last activity, exclusive lock status. Include --json flag.","status":"closed","priority":1,"issue_type":"task","created_at":"2025-10-26T17:09:51.140442-07:00","updated_at":"2025-10-26T17:47:47.929666-07:00","closed_at":"2025-10-26T17:47:47.929666-07:00","dependencies":[{"issue_id":"bd-158","depends_on_id":"bd-145","type":"parent-child","created_at":"2025-10-26T17:09:51.150077-07:00","created_by":"daemon"}]}
 {"id":"bd-159","title":"Timestamp-only changes still being exported despite dedup logic","description":"User observed timestamp-only changes in .beads/beads.jsonl causing dirty working tree. Example: bd-128's updated_at changed from 2025-10-25T23:51:09.811006-07:00 to 2025-10-26T14:12:45.207573-07:00 with no other field changes.\n\nThis should have been prevented by the export deduplication logic that's supposed to skip timestamp-only updates.\n\nNeed to investigate why timestamp-only changes are still being exported and fix the dedup logic.","status":"open","priority":1,"issue_type":"bug","created_at":"2025-10-26T17:58:15.41007-07:00","updated_at":"2025-10-26T17:58:15.41007-07:00"}
 {"id":"bd-16","title":"Add lifecycle safety docs and tests for UnderlyingDB() method","description":"The new UnderlyingDB() method exposes the raw *sql.DB connection for extensions like VC to create their own tables. While database/sql is concurrency-safe, there are lifecycle and misuse risks that need documentation and testing.\n\n**What needs to be done:**\n\n1. **Enhanced documentation** - Expand UnderlyingDB() comments to warn:\n   - Callers MUST NOT call Close() on returned DB\n   - Do NOT change pool/driver settings (SetMaxOpenConns, SetConnMaxIdleTime)\n   - Do NOT modify SQLite PRAGMAs (WAL mode, journal, etc.)\n   - Expect errors after Storage.Close() - use contexts\n   - Keep write transactions short to avoid blocking core storage\n\n2. **Add lifecycle tracking** - Implement closed flag:\n   - Add atomic.Bool closed field to SQLiteStorage\n   - Set flag in Close(), clear in New()\n   - Optional: Add IsClosed() bool method\n\n3. **Add safety tests** (run with -race):\n   - TestUnderlyingDB_ConcurrentAccess - N goroutines using UnderlyingDB() during normal storage ops\n   - TestUnderlyingDB_AfterClose - Verify operations fail cleanly after storage closed\n   - TestUnderlyingDB_CreateExtensionTables - Create VC table with FK to issues, verify FK enforcement\n   - TestUnderlyingDB_LongTxDoesNotCorrupt - Ensure long read tx doesn't block writes indefinitely\n\n**Why this matters:**\nVC will use this to create tables in the same database. Need to ensure production-ready safety without over-engineering.\n\n**Estimated effort:** S+S+S = M total (1-3h)","design":"Oracle recommends \"simple path\": enhanced docs + minimal guardrails + focused tests. See oracle output for detailed rationale on concurrency safety, lifecycle risks, and when to consider advanced path (wrapping interface).","status":"closed","priority":1,"issue_type":"task","created_at":"2025-10-22T17:07:56.812983-07:00","updated_at":"2025-10-25T23:15:33.476053-07:00","closed_at":"2025-10-22T20:10:52.636372-07:00"}
+{"id":"bd-160","title":"Critical: Multi-clone sync is fundamentally broken","description":"## Problem\n\nTwo clones of the same beads repo working on non-overlapping issues cannot stay in sync. The JSONL export/import mechanism creates catastrophic divergence instead of keeping databases synchronized.\n\n## What Happened (2025-10-26)\n\nTwo repos working simultaneously:\n- ~/src/beads (bd.db, 159 issues) - worked on bd-153, bd-152, bd-150, closed them\n- ~/src/fred/beads (beads.db, 165 issues) - worked on bd-159, bd-160-164\n\nResult after attempting sync:\n- Databases completely diverged (159 vs 165 issues)\n- JSONL files contain conflicting state\n- Database corruption in fred/beads\n- bd-150/152/153 show as closed in one repo, open in the other\n- No clear recovery path without manual database copying\n- git pull + bd sync does NOT synchronize state\n\n## Root Causes\n\n1. **Database naming inconsistency**: One repo uses bd.db, other uses beads.db - no enforcement\n2. **Daemon state divergence**: Each repo's daemon maintains separate state, never converges\n3. **JSONL import/export race conditions**: Auto-import can overwrite local changes before export\n4. **No conflict resolution**: When databases diverge, there's no merge strategy\n5. **Timestamp-only changes**: bd-159 - exports trigger even with no real changes\n6. **Multiple daemons**: No coordination between daemon instances\n\n## Expected Behavior\n\nTwo developers working in separate clones on non-overlapping issues should:\n1. git pull\n2. bd sync\n3. See each other's changes immediately\n4. Never have database divergence\n\n## Actual Behavior\n\nDatabases diverge catastrophically with no recovery path except:\n- Manual database file copying\n- Destructive force-import\n- Database corruption\n\n## Impact\n\n**Beads is unusable for multi-developer or multi-agent workflows**. The core promise - git-based sync via JSONL - is broken.\n\n## Required Fixes\n\n1. Enforce single canonical database name (beads.db everywhere)\n2. Make bd sync actually synchronize (currently it fails)\n3. Add conflict detection and resolution for diverged databases\n4. Fix daemon state synchronization across clones\n5. Prevent timestamp-only exports (bd-159)\n6. Add database integrity checks\n7. Provide clear recovery procedures for divergence\n8. Test multi-clone workflow in CI\n\n## Severity\n\nP0 - This breaks the fundamental use case of beads. Without reliable sync, the tool is unusable for any multi-agent or team scenario.","status":"open","priority":0,"issue_type":"bug","created_at":"2025-10-26T19:42:43.355244-07:00","updated_at":"2025-10-26T19:42:43.355244-07:00"}
 {"id":"bd-17","title":"Update EXTENDING.md with UnderlyingDB() usage and best practices","description":"EXTENDING.md currently shows how to use direct sql.Open() to access the database, but doesn't mention the new UnderlyingDB() method that's the recommended way for extensions.\n\n**Update needed:**\n1. Add section showing UnderlyingDB() usage:\n   ```go\n   store, err := beads.NewSQLiteStorage(dbPath)\n   db := store.UnderlyingDB()\n   // Create extension tables using db\n   ```\n\n2. Document when to use UnderlyingDB() vs direct sql.Open():\n   - Use UnderlyingDB() when you want to share the storage connection\n   - Use sql.Open() when you need independent connection management\n\n3. Add safety warnings (cross-reference from UnderlyingDB() docs):\n   - Don't close the DB\n   - Don't modify pool settings\n   - Keep transactions short\n\n4. Update the VC example to show UnderlyingDB() pattern\n\n5. Explain beads.Storage.UnderlyingDB() in the API section","status":"closed","priority":1,"issue_type":"task","created_at":"2025-10-22T17:07:56.820056-07:00","updated_at":"2025-10-25T23:15:33.478579-07:00","closed_at":"2025-10-22T19:41:19.895847-07:00","dependencies":[{"issue_id":"bd-17","depends_on_id":"bd-10","type":"discovered-from","created_at":"2025-10-24T13:17:40.32522-07:00","created_by":"renumber"}]}
 {"id":"bd-18","title":"Consider adding UnderlyingConn(ctx) for safer scoped DB access","description":"Currently UnderlyingDB() returns *sql.DB which is correct for most uses, but for extension migrations/DDL, a scoped connection might be safer.\n\n**Proposal:** Add optional UnderlyingConn(ctx) (*sql.Conn, error) method that:\n- Returns a scoped connection via s.db.Conn(ctx)\n- Encourages lifetime-bounded usage\n- Reduces temptation to tune global pool settings\n- Better for one-time DDL operations like CREATE TABLE\n\n**Implementation:**\n```go\n// UnderlyingConn returns a single connection from the pool for scoped use\n// Useful for migrations and DDL. Close the connection when done.\nfunc (s *SQLiteStorage) UnderlyingConn(ctx context.Context) (*sql.Conn, error) {\n    return s.db.Conn(ctx)\n}\n```\n\n**Benefits:**\n- Safer for migrations (explicit scope)\n- Complements UnderlyingDB() for different use cases\n- Low implementation cost\n\n**Trade-off:** Adds another method to maintain, but Oracle considers this balanced compromise between safety and flexibility.\n\n**Decision:** This is optional - evaluate based on VC's actual usage patterns.","status":"closed","priority":3,"issue_type":"feature","created_at":"2025-10-22T17:07:56.832638-07:00","updated_at":"2025-10-25T23:15:33.479496-07:00","closed_at":"2025-10-22T22:02:18.479512-07:00","dependencies":[{"issue_id":"bd-18","depends_on_id":"bd-10","type":"related","created_at":"2025-10-24T13:17:40.325463-07:00","created_by":"renumber"}]}
 {"id":"bd-19","title":"MCP close tool method signature error - takes 1 positional argument but 2 were given","description":"The close approval routing fix in beads-mcp v0.11.0 works correctly and successfully routes update(status=\"closed\") calls to close() tool. However, the close() tool has a Python method signature bug that prevents execution.\n\nImpact: All MCP-based close operations are broken. Workaround: Use bd CLI directly.\n\nError: BdDaemonClient.close() takes 1 positional argument but 2 were given\n\nRoot cause: BdDaemonClient.close() only accepts self, but MCP tool passes issue_id and reason.\n\nAdditional issue: CLI close has FOREIGN KEY constraint error when recording reason parameter.\n\nSee GitHub issue #107 for full details.","status":"closed","priority":0,"issue_type":"bug","created_at":"2025-10-22T17:25:34.67056-07:00","updated_at":"2025-10-25T23:15:33.480292-07:00","closed_at":"2025-10-22T17:36:55.463445-07:00"}