ci: windows arm64 tests/wheels

[lgray]

for completeness.

py3.8 not available on windows arm64 images, not tested there.

[lgray]

Well that's an interesting outcome at the least:

C:\a\correctionlib\correctionlib\pcg-cpp\include\pcg_uint128.hpp(76,1): error C1189: #error:  Unable to determine target endianness [C:\Users\runneradmin\AppData\Local\Temp\tmpprtg5pe4\build\correctionlib.vcxproj]

Along with some strange test errors.

[nsmith-]

Probably due to missing support in pcg-cpp imneme/pcg-cpp#99

[lgray]

Oh, cool hopefully it gets merged soon.

[ikrommyd]

@lgray master branch has been migrated to the maintained fork o pcg-cpp. Do you wanna rebase and continue?

[lgray]

ah - cool I'll poke it today :)

[lgray]

awkward-cpp and cramjam need windows-arm64 wheels but other than that we're good to go here! Looks like we get arm64 wheels for awkward-cpp with the next release, cool.

@nsmith- please review and merge at your leisure.

[ikrommyd]

scikit-hep/awkward#3855 awkward-cpp now does. They will show up in the next release. We can't control cramjam though but we can ask?

[lgray]

concerning awkward-cpp - yeah I noticed and updated my comment ^_^

I guess it all works so I'll make a PR over there.

[ikrommyd]

Why do we get cramjam here? We just get it through uproot? Uproot is considering switching to numcodecs which also doesn't have windows arm wheels so may need to ask them instead 🤣

[lgray]

Yes, for uproot. numcodecs also appears to be a bit of a ghost town. look at their PRs.

[ikrommyd]

I know they maintain it as much as needed. It's a core dependency of zarr so it will definitely live.

[lgray]

As long as zarr lives, you mean :-)

[nsmith-]

uproot isn't a hard dependency here, so this must just be for testing. In that case we can go ahead

[ikrommyd]

Why did we skip 3.9 and 3.10 windows arm64 wheels here? cibuildhweel says it can do them here https://cibuildwheel.pypa.io/en/stable/

[lgray]

There are no bumpy wheels for 3.9 and 3.10 in win-arm64. It's largely pointless to host ours in that case.

[ikrommyd]

Oh I see, I naively thought one couldn't build correctionlib for whatever reason. Of course yeah. We will not build numpy.