CI: set up ccache

[lucascolley]

on top of gh-19

[lucascolley]

@steppi I tried following the config in SciPy, but it looks like ccache isn't having an impact here

[steppi]

@steppi I tried following the config in SciPy, but it looks like ccache isn't having an impact here

Thanks for checking. CI time is still pretty short anyway so let's leave investigating this for the future.

[rgommers]

@lucascolley conda-forge compilers set CFLAGS et al on activation, which overwrites what you put into a GHA yml file. What you seem to be missing here is env = {CC: cache $CC} in pixi.toml. See how gpu.ci.ymland the correspondingpixi.toml` in the main scipy repo are set up.

[lucascolley]

thanks Ralf, that did the trick!

Successful in 1m

1m58s, but we'll take it :) 🚀🚀

The bottleneck now seems to be processing the .gcda files for coverage, not sure if there is a way for cache to help us there.

[steppi]

The failures are because there were some tests for overloads which don't actually exist in xsf and I removed the parquet files for them. The tests themselves are removed in #21. Let's just merge that one first, I'll have the tolerance files ready shortly.

[steppi]

I merged #21. After merging main this should be ready.

[lucascolley]

yep, this is ready. We should see whether -j2 is any better than -j3 at some point, but that can probably wait until prefix-dev/pixi#3538 is fixed, 2 minutes is already pretty speedy

[steppi]

Great! Thanks @lucascolley and thanks @rgommers for explaining why it wasn't working before.