Performance regression in coursier code paths on Windows #5398
Description
I'm trying to rebootstrap Mill in preparation for releasing 1.0.0, but hit a performance regression in JavaHomeModule#javaHome/defaultResolver().resolve code paths on windows that is making the __.compile on the Mill codebase take a long time. It's not stuck, but the slowdown seems unreasonable:
- Takes 15 minutes on my Windows laptop to compile everything (instead of 1.5 minutes on my Mac laptop with about the same number of cores) and
- Probably timing out in Windows CI (https://github.com/com-lihaoyi/mill/actions/runs/15964889955/job/45031598162?pr=5396) (instead of taking ~7 minutes before https://github.com/com-lihaoyi/mill/actions/runs/15964933177/job/45023491406).
- A second run on my laptop also takes very long, even with the artifacts all cached locally
I can reproduce this slowdown on my Windows-ARM laptop, and it appears in CI on Windows-Intel. Doesn't seem to appear on my macbook pro or on Mac/Linux ARM/Intel CI
The commits since the last bootstrap can be seen in 6bd3a9...43069c. The most likely candidate seems to be #5350, which bumps the coursier version, and may have affected the related codepaths. @alexarchambault can you help investigate this and see if this might be related to the courwsier version bump?