Investigate getattrlistbulk on macOS / APFS

[Par-python]

Context

bigfiles currently uses the ignore crate's parallel walker, which on macOS ultimately calls readdir per directory. macOS provides a faster bulk API: getattrlistbulk(2), which returns metadata for many directory entries in a single syscall.

For large trees (especially ~/Library, node_modules, Xcode derived data) this can meaningfully reduce syscall overhead, since each readdir + stat pair becomes one bulk call.

Goal

Evaluate whether bigfiles should add a macOS fast path using getattrlistbulk, either as a drop-in replacement for the per-directory step inside the walker or as a --fast mode.

Open questions

• Is there a maintained Rust wrapper? If not, we'd need to bind it via the libc crate.
• How does it interact with the gitignore / --exclude filtering layer? (Likely fine — we still get paths back, just faster.)
• Benchmarks vs ignore walker on representative trees (~, a node_modules, Xcode derived data).
• Symlink and hidden-file behavior parity with the current walker.

References

• Apple docs: https://developer.apple.com/documentation/kernel/1532353-getattrlistbulk
• walkdir issue discussing similar tradeoffs: FilterEntry::filter_entry has a suboptimal signature BurntSushi/walkdir#130

Why this matters

macOS is bigfiles' primary platform today. Any meaningful perf win here directly improves the experience for the largest chunk of current users.

Status

Not started. Tracking for future work; contributions welcome.