OpenAI Contributes Git Improvements for Large Repository Performance and Stability

Addressing Static Analysis Timeouts

One patch resolved an issue where a reused loop index caused Coccinelle's static analysis to exceed a six-hour CI limit. By declaring separate loop indexes, static analysis can now complete normally, ensuring continuous integration processes are not stalled by this specific problem.

Optimizing Large Repository Performance

Another significant contribution improved bitmap selection for large repositories. Previously, bitmap selection undervalued merge commits, leading to inefficient processing. The updated approach chooses bitmap commits based on the history they make reachable, rather than path length alone. This change reduced the measured build time for MIDX bitmap writes from 984.73 seconds to 200.06 seconds while selecting the same number of bitmaps.

Enhancing Parallel Checkout Diagnostics

To better diagnose performance issues during parallel checkouts, new Trace2 counters were added. These counters record each worker's item count, bytes processed, failures, elapsed time, and slowest item. This data helps determine if long checkout times are due to uneven work distribution or a single slow file.

Improving Filesystem Monitoring and Status Caching

Further patches addressed potential inconsistencies in filesystem monitoring and introduced status caching. One change ensures that scanned file descriptors and metadata remain available until the final filesystem-monitor check, preventing issues from concurrent worktree changes. Additionally, an optional .csts file was documented, allowing for the caching of `git status --porcelain=v2` results on APFS worktrees, with fallbacks to ordinary status if the cache is missing, stale, or malformed.