Your Mac is slowing you down (a rant about file systems)
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
A detailed benchmark-driven rant compares file system performance for developers doing heavy agentic dev work with parallel work trees and node module installs. APFS on Mac performs dramatically worse than Linux file systems like ext4 and XFS for deletions, installs, and work tree creation - sometimes 5-10x slower. XFS with LZ4/VDO compression emerges as the best option, cutting storage use by up to 3x while maintaining speed comparable to ext4. ZFS and Btrfs are ruled out as poor fits despite their other strengths. The creator also covers Rift by Anomaly, a copy-on-write work tree tool for Mac OS that appears to be an abandoned project.
Questions this post answers
Why is APFS so slow for pnpm installs and git clean compared to Linux file systems?
APFS lacks efficient copy-on-write and linking primitives that Linux file systems like XFS and ext4 handle well, causing massive overhead when creating hard links or copying many small files. Benchmarks showed an M1 Max took 31 seconds to clean and 44 seconds to install node modules, versus 6 and 4.5 seconds on an AMD Ryzen machine running ext4, a 5-10x difference even though the Mac's SSD is faster. Developers weighing a move off Mac for agentic coding can compare real benchmarks like these on daily.dev.
What file system setup is best for running multiple AI agent work trees in parallel?
XFS combined with LZ4 compression via VDO offers the best balance of speed and space savings for parallel work tree creation and node module installs. Testing with 125 work trees showed stock ext4 used 75GB while XFS with VDO used only 24GB, and switching to hard links for pnpm cut work tree creation time from 7 seconds to 2.85 seconds. Anyone tuning their dev box for parallel agent workflows can track file system tradeoffs like this on daily.dev.
Is Rift by Anomaly a good alternative to git worktrees on Mac OS?
Rift virtualizes files using copy-on-write so agents can work instantaneously without persisting new files until edited, making it fast even on APFS, and it works with bun or node directly. However its last commit was two months prior with no further activity, suggesting the project may be abandoned despite its promising approach. Developers evaluating copy-on-write tooling for git worktrees can follow projects like this on daily.dev.