Conversation
|
CI is passing. It took a bit to work around limitations in the gfortran_install and multibuild subrepos, maybe we should switch to the scipy strategy of using the conda-forge |
|
Ah you switched all the builds to
It's not a whole lot better (SDKROOT messing around still required), and actually it looks like for arm64 we're pulling in some other |
|
The more impactful maintenance improvement is probably to switch to |
rgommers
left a comment
There was a problem hiding this comment.
I think we can only really know if this worked when building numpy/scipy wheels against these new versions, but the changes and CI logs LGTM.
|
(I'll leave the merging up to you, since I didn't follow along very closely) |
|
Thanks for the review
Good idea, I will open an issue |
git describe --tags --abbrev=8in OpenBLAS at theOPENBLAS_COMMIT. If I did not updateOPENBLAS_COMMIT, I incremented the wheel build number (i.e. 0.3.29.0.0 to 0.3.29.0.1)This moves the macos builds from a x86_64 host to an arm64 host, hopefully it will Just Work