pomazanbohdan/memory-mcp-1file

🛡️ Crash & OOM Fix Release

Critical Fixes

• Grammar load panic → graceful skip if tree-sitter fails (extractor.rs)
• Mutex poison cascade → recover from poisoned mutex instead of crashing (cache.rs)
• 100+ nested runtimes leak → eliminated per-callback runtime creation during model download (service.rs)

Smart Large File Handling (indexer.rs)

Instead of hard-skipping large files, v0.2.7 uses a smart strategy:

1. Skip auto-generated files (.g.dart, .freezed.dart, .min.js, .min.css, lock files) — no useful semantic content
2. Cap chunks per file at 50 for legitimately large source files — processes them but bounds memory
3. Warn on files >1MB in logs for visibility

Indexing Efficiency (scanner.rs)

Removed non-code file extensions from indexing: .json, .yaml, .yml, .toml, .xml, .md — these produced noisy, low-value search results and wasted memory/CPU

Docker Memory Safety (Dockerfile.release)

Added SURREAL_MEMORY_THRESHOLD=256mb to cap SurrealDB cache allocation as defense-in-depth

Files Changed (9 files, +101/-25)

| File | Change |
|------|--------|
| extractor.rs | expect() → is_err() + return None |
| cache.rs | .unwrap() → .unwrap_or_else(\|e\| e.into_inner()) |
| service.rs | Nested rt_inner → try_write() directly |
| indexer.rs | is_generated_file() + MAX_CHUNKS_PER_FILE=50 |
| scanner.rs | Removed 6 non-code extensions |
| Dockerfile.release | ENV SURREAL_MEMORY_THRESHOLD=256mb |
| Cargo.toml | 0.2.6 → 0.2.7 |
| package.json | 0.2.6 → 0.2.7 |

Full Changelog: https://github.com/pomazanbohdan/memory-mcp-1file/compare/v0.2.6...v0.2.7