add arm64 macos build

[ianhi]

hopefully fixes: https://github.com/AllenCellModeling/aicspylibczi/issues/101

[ianhi]

oops i may have put in the wrong place

[ianhi]

ah ya beat me to it

[psobolewskiPhD]

Y'all lemme know (how) and I can test anything M1 related!

[ianhi]

I don't think this should affect any of the tests here so the only way to check is to run the deply script and try to use the generated wheel

[evamaxfield]

I don't think this should affect any of the tests here so the only way to check is to run the deply script and try to use the generated wheel

Agree. I leave this PR approve and deploy to @toloudis

[codecov-commenter]

Codecov Report

Merging #102 (c963e50) into main (0c9a19d) will not change coverage. The diff coverage is n/a.

@@           Coverage Diff           @@
##             main     #102   +/-   ##
=======================================
  Coverage   98.64%   98.64%           
=======================================
  Files           3        3           
  Lines         148      148           
=======================================
  Hits          146      146           
  Misses          2        2           

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 0c9a19d...c963e50. Read the comment docs.

[toloudis]

I think we can test prior to merging by changing the conditions that the action runs under (and remove the publish step), then watch actions and see if anything fails. if the build succeeds, then we can reset the conditions and merge and move forward. Otherwise I can just merge and go for it blind instead of half blind :)

[toloudis]

i'll just do it, let's build it and see.

[psobolewskiPhD]

Should just show up here right? https://pypi.org/project/aicspylibczi/#files Edit: getting sleepy!! Edit2: Ill check tomorrow!

[ianhi]

Should just show up here right?

@psobolewskiPHD I think you can only get it from: https://github.com/AllenCellModeling/aicspylibczi/suites/4994059858/artifacts/146261793

[evamaxfield]

Whenever we push a new version, likely just a patch release, it should also be pushed to pypi

[toloudis]

yes - it won't go to pypi until we do a version bump and trigger the release publish process

[psobolewskiPhD]

Bad news, unless I screwed something up:

pip install /Users/piotrsobolewski/Downloads/artifact/aicspylibczi-3.0.4-cp3-cp39-macosx_11_0_arm64.whl
Processing /Users/piotrsobolewski/Downloads/artifact/aicspylibczi-3.0.4-cp39-cp39-macosx_11_0_arm64.whl
Requirement already satisfied: numpy>=1.14.1 in /Users/piotrsobolewski/Dev/miniforge3/envs/napari-env/lib/python3.9/site-packages (from aicspylibczi==3.0.4) (1.22.0)
Installing collected packages: aicspylibczi
Successfully installed aicspylibczi-3.0.4

But:

> lipo -info _aicspylibczi.cpython-39-darwin.so
Non-fat file: _aicspylibczi.cpython-39-darwin.so is architecture: x86_64

Edit: so it looks like wheels are packaged, but the wrong compiled? library is included.

[evamaxfield]

Sorry, why is this bad news? It looks like that install worked!

[psobolewskiPhD]

The install worked, but the library in the wheel is x86-64: Non-fat file: _aicspylibczi.cpython-39-darwin.so is architecture: x86_64 It should be arm64: artifact/aicspylibczi-3.0.4-cp3-cp39-macosx_11_0_arm64.whl So the wheel/bundling is making a "thing" for arm64, but it's not actually compiling an arm64 version of the library and placing it inside the bundle. Edit: this means that in a native arm64 environment this wheel will not work and will throw an architecture error.

[evamaxfield]

Here are the docs from cibuildwheel themselves if you find any insights in them... i can't see what went wrong: https://cibuildwheel.readthedocs.io/en/stable/faq/#apple-silicon

[evamaxfield]

And, looking at their direct example. It's the code we had before this PR :confused:

EDIT: correction we had auto universal2 instead of x86_64 universal2 I guess thats another configuration we could try????

[psobolewskiPhD]

I can't say I understand how any of this works. I don't get what cibuildwheel does, relative to the setup.py/cfg files. I assume the issue is that when the _aics bit is compiled it's not cross compiled, but just x86. I've fixed other things to compile by set the cmake flag CMAKE_OSX_ARCHITECTURES to arm64—but this library compiles just fine if i clone the repo. I think it must be something related to cmake because: https://github.com/AllenCellModeling/aicspylibczi/issues/101#issuecomment-1017252716 Trying to compile via pip also fails.