-
-
Notifications
You must be signed in to change notification settings - Fork 2.2k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Version 0.18.2 -- aarch64 wheels #5299
Comments
I'm trying to build 32 bit wheels with verbose testing in: |
It looks like the failure occurs within a call to
<\details> |
Earlier in the log, pip does not find a NumPy wheel, so it gets built from source instead. It looks like there is an aarch64 wheel for Python 3.9 on PyPI, but it was built with manylinux2014: https://pypi.org/project/numpy/#files |
@grlee77 i guess not only do i have to bump the normal python 3.9 requirements for x86, but i have to bump manylinux even more for aarch64. |
It seems that the very first release of numpy for aarch was using manylinux2014 https://pypi.org/project/numpy/1.20.0/#files I can probably pin to that. |
It seems that 2014 is the current default for aarch64 |
@jni not sure what is left to release 0.18.2. I think it would be beneficial to do it sooner rather than later since people have been asking for an aarch build. Thoughts? |
@hmaarrfk
Needed So after a bit of digging and manually installing the dependencies of
Ok, lemme clone My first run of
But when running
Full log of the install step: Help? |
I guess the requirements.txt are not fully in sync with the pyproject.toml. can you try removing the pyproject.toml file? It lists the requirement of the lowest version of bumpy that exists on pypi as a wheel for aarch. It seems that you have built it yourself though with a lower version. Is that correct? |
@hmaarrfk I have not yet successfully installed scikit-image:
When I removed the
|
@theschles i'm really sorry to ask, but could you please open a new issue. The goal of this issue is to track progress on the 0.18.2 release which we plan to get out soon (in order to have aarch wheels). In the issue, can you please provide the output of |
@hmaarrfk I think we can go ahead and release assuming all is green on the 0.18.x branch. Separately, @grlee77 has started working with @leej3 and @viniciusdc, also at Quansight, to overhaul our release process, so maybe this is a good chance to get them to witness the mess 😂. Perhaps we can schedule a day next week to have a big release party? ;) |
Sign me up! |
@janaknat @theschles I've pushed 3.7 and 3.9 aarch64 wheels for 0.18.2rc1. You can try installing them with At any rate, the wheels the action did upload are built on manylinux2014 for arch. I can't see which part of the config sets that — it seems to be implicitly the only architecture it can use, because the env is set to manylinux2010. Anyway, let us know whether these wheels work for you, if so, I'll tag 0.18.2 and push those. The 0.18.2 release notes can be found here. |
That rank filter failure may be resolved by backporting #5175 |
@jni The workflow that failed was due to i686 having an error. Hopefully, with the PR suggested by @grlee77 it gets resolved. With respect to manylinux, aarch64 is only supported starting with manylinux2014. cibuildwheel automatically does the right thing for aarch64. I can test the wheels. Any particular test you'd like to see me run to check for correctness? |
I personally think we can skip hardware testing. We already release aarch stuff on conda-forge and it seems to be fine in the wild. |
Whatever it is you're using skimage for! 😂 ie if your feedback is that the wheels are fit for purpose we'll push the release — now with the latest test-skipping patch! 🎉 |
@jni, @janaknat , I gave scikit-image 0.18.2rc1 wheels a test on Arm64 with Python3.7 and Python3.9. The binary portions load fine and seem to be working. I think they are good to go with a release. Its a shame the 3.8 wheels are not there. The i686 build may be handling the pickled arrays differently and that is why its failing with a bad magic. I found this reference: https://stackoverflow.com/questions/40054872/numpy-load-wrong-magic-string-error Maybe explicitly telling the |
@geoffreyblake thank you for the feedback! The missing wheel was in fact fixed by #5383. I just pushed a new tag, hopefully will push a new release candidate in a few hours when the builds finish. 🤞 |
Ok skimage 0.18.2rc2 is released, with aarch64 wheels for Python 3.{7,8,9}! @geoffreyblake @janaknat if you give the go-ahead with these I will release 0.18.2 proper! Apologies for the long cycle, we are getting there! 🙏 |
@jni sciki-image 0.18.2rc2 appears to be working fine on aarch64 for python 3.{7,8.9}. I believe it is ready for release. |
@jni Any ideas on when the aarch64 wheels get released? |
Ooops! I tagged the release but never uploaded the wheels and source to PyPI! This is why the work of @leej3 and @viniciusdc in #5397 is so important! 😅 Release happening right now! 🚨🚨🚨 |
Release made! So happy that I can finally close this! Thank you everyone for all your contributions and for your patience! 🙏 |
Thank you @jni ! |
On the road to building aarch64 wheels, we are trying to release 0.18.2
Presently, from what I know, we need to:
cc: @janaknat
The text was updated successfully, but these errors were encountered: