pomazanbohdan/memory-mcp-1file

🔧 Fix: Incremental Indexing

This release fixes the critical incremental indexing bug that caused full re-indexing on every file change.

Root Cause

incremental_index compared blake3(file_content) vs blake3(chunk₁ ++ chunk₂ ++ ...). Because AST chunking skips small nodes, trims large ones, and reformats content — these hashes never matched. Every file was treated as modified.

What's Fixed

🗄️ File-Level Hash Tracking

• New file_hashes table stores blake3(file_content) per file
• incremental_index now compares actual file content hash → skip if unchanged
• Hashes are stored after full index and updated after incremental re-index

🧹 Symbol Relation Cleanup

• delete_symbols_by_path was using file_path on symbol_relation — an edge table that has no such field
• Fixed to use in/out subquery through code_symbols
• No more orphaned relations after file deletion

🛡️ Prevent Accidental Full Re-index

• index_project tool now returns "already indexed" when project status is Completed or EmbeddingPending
• New force parameter (default: false) to explicitly trigger full re-index
• File changes are handled incrementally by the file watcher

🔄 EmbeddingPending State Handling

• CodebaseManager::start() now treats EmbeddingPending as completed (index data is valid)
• Previously fell through to catch-all branch and did nothing

Files Changed

| File | Change |
|------|--------|
| schema.surql | New file_hashes table |
| traits.rs | 4 new trait methods |
| surrealdb.rs | Implementations + SQL fix |
| indexer.rs | Hash comparison + file hash recording |
| manager.rs | EmbeddingPending handling |
| params.rs | force parameter |
| logic/code.rs | Skip re-index + cleanup |

Tests

• 73 tests passing ✅
• cargo clippy clean ✅