Repository navigation
Lean under churn, measured on an M6 Mac mini with a 3.4M-entry disk against 0.2.1.
- The daemon no longer pegs a core under builds, caches and
node_modules: a cargo build costs it 0.08 s of CPU (was 1.05 s), a 50k-filenode_modulesburst 0.18 s (was 2.20 s), and sustained heavy churn 42-54% of a core (was 75%) - A mixed churn workload takes 32-42% less CPU, with 0 full index rebuilds (was 5) and peak memory of 380 MB (was 847 MB)
- Changes inside cache and build folders (
node_modules,__pycache__, cargotarget/,~/Library/Caches) are applied a few seconds after they settle, and compaction waits for a quiet disk instead of rebuilding the index back to back - Dropped FSEvents back off instead of looping
- Recovery when the FSEvents journal re-bases below the saved event id (changes made while the daemon was off were silently missed)
fsearch versionand--version- Plus several restart and recovery edge cases that could miss changes
- One note: a change inside those cache and build folders shows up about 5 s after it settles; everything else within about half a second
Thanks @sirrayi, @muhammad-luay and @domfahey.