v0.13.8

Performance

Fast startup on large repos (v0.13.8)

• On iree (60k files), the MCP server took 10 seconds to become ready — spent entirely on 8 rglob calls counting files for the staleness check
• Split is_stale() into fast and full modes:
  • Fast (startup + tool calls): samples mtime of 100 cached file paths, no filesystem walk — <0.1s
  • Full (periodic background check every 60s): does the expensive file count to detect new/deleted files
• Startup-to-ready on iree: 10s → <0.1s

Skip O(n²) similarity checks on large repos (v0.13.7)

• find_similar_classes/find_similar_functions were doing pairwise SequenceMatcher comparisons — 87M comparisons on baspacho (13k functions), burning 100% CPU for minutes after indexing
• Skipped when candidate list exceeds 2,000 symbols

Fixes

Zombie server processes (v0.13.5)

• Servers hung indefinitely after sessions ended due to uncancelled background tasks

Infinite reindex loop (v0.13.6)

• Large repos with uncacheable files triggered reindexing every 20 minutes forever (since January on iree)

Unthrottled hook processes

• Removed map.py from project-switch hook (MCP server handles indexing with proper resource limits)
• Added 60s timeout to scan.py in hook

New

Server diagnostics: uv run scripts/servers.py

• Lists all running repo-map servers with PID, uptime, memory, project root, index status, cache info, and recent logs