I also asked this here: https://vercel.community/t/performance-of-hashing-on-different-file-systems/6101

Summary

I'm been noticing some very shocking things with APFS, that are honestly quite concerning.

Here is the test I setup: https://github.com/NullVoxPopuli/disk-perf-git-and-pnpm/blob/main/README.md
along with a number of datapoints from various machines and operating systems.
Interactive table here

This doesn't test turbo specifically, but more file creation and deletion.

turborepo has to read all files in a package (governed by inputs), to use as part of what creates the hash for a task.

Now, I know read is different from create and delete, but still... I wonder.

Is there a test we can do with turbo in a sufficiently sized repo on both linux/ext4 and macOS/APFS to see how long it takes just to compute hashes for tasks?

As an aside, if anyone is curious about disk perf on APFS, I could really use your help if you have an M2, M3, or M4,

Instructions here: disk-perf-git-and-pnpm/README.md at main · NullVoxPopuli/disk-perf-git-and-pnpm · GitHub

tl;dr: time ( a git clean ) and time ( dep install )

thank you!!