search

xgcm / xcape

Fast convective parameters for numpy, dask, and xarray
Other
30 stars 11 forks source link

Use meson to build the wheel #55

Open ocefpaf opened 3 months ago

ocefpaf commented 3 months ago

This modernize the packaging and wheel building with:

It is building on all platforms but the tests are passing only on Linux.

Ping @weiji14 and @scottyhq.

ocefpaf commented 3 months ago

Linux: 34 passed, 0 failed Windows: 12 failed, 22 passed macOS: 8 failed, 26 passed

I don't think this package was tested on Windows or macOS before, so maybe the failures that are showing up were already there. MacOS ones seems to be related to numeric precision.

Edit: I won't fix the failing test on Windows and macOS b/c I don't have experience with CAPE to do this. My PR here goes as far as fixing the CIs and build it for Linux.

If you don't want a failing CI I can remove Windows and macOS tests to keep in line with what existed before in the Travis-CI config.

weiji14 commented 3 months ago

Edit: I won't fix the failing test on Windows and macOS b/c I don't have experience with CAPE to do this. My PR here goes as far as fixing the CIs and build it for Linux.

If you don't want a failing CI I can remove Windows and macOS tests to keep in line with what existed before in the Travis-CI config.

No experience with CAPE either, but from a conda-forge packaging perspective, I'd reckon it's ok to drop the Windows CI since it's not being built. As for macOS, you could maybe xfail those tests for now, and someone could look at whether the numerical precision errors are ok later?

ocefpaf commented 3 months ago

I'd reckon it's ok to drop the Windows CI since it's not being built. As for macOS, you could maybe xfail those tests for now, and someone could look at whether the numerical precision errors are ok later?

I'm not a dev it maintainer here so, while I agree with you, it is not up to me to decide. I would merge the PR as is though, to remind to come back to the failing tests ASAP.