Installation fails on M1 mac #114
Comments
|
Seems like Gh actions don't support M1. I use cibuildwheel, and they also confirm that Cirrus CI is the first public CI platform that natively supports macOS Apple Silicon (so GH actions don't have M1 support). You're on your own basically. Try to use cibuildwheel like I do here to build a .whl file on your computer perhaps. I'll leave this issue open. When it's easy to build using Gh actions I will create or merge a PR so we can have M1 support, but I don't have the capacity to investigate other solutions to this problem in detail myself. |
|
I see. Hopefully, M1 support is added to GH actions soon. I was able to install the package locally by just cloning the repository and installing from source. I'm not sure why that worked as opposed to installing from sdist, but hey, it works now. I suppose that since I have cython installed in my environment, this may have re-cythonized the pyx files, but, skimming over the setup code, I can't really see how or why. |
|
@pavlin-policar Could you please share the steps you performed to install the package locally? |
|
I've don't have the bash logs anymore, but it was probably something along the lines of git clone git@github.com:tommyod/KDEpy.git
cd KDEpy
pip install .Hope that helps! |
|
Thanks! I was able to get it installed. |
|
Hi @tommyod , |
|
I don't use tags anymore. If you need a specific verison, get it from pypi instead. I'll consider tagging future releases if I remember it. But the package is so stable by now that new releases are infrequent. |
|
PyPi does not have built wheel file for s390x. So building from source code is tried by pip and that fails because of the bug I mentioned above. So a tag after that fix of removal of cutils.c in PR 89 would help us in using "pip install" on s390x. Thank you |
|
@rohan097 , @tommyod , I am running into the same issue. However, when I follow your steps to install KDEpy, it does install KDEpy successfully, but then when I import the package in python it complains about When I check in Conda it seems that conda does not have a version of cutils that it can install for the M1 x64 architecture: Any thoughts ? Thanks so much! |
|
"cutils" refers to this file: https://github.com/tommyod/KDEpy/blob/master/KDEpy/cutils.pyx Seems like the Cython .pyx code did not build. This issue also reports errors with Mac: #88 I don't have a Mac myself, and I won't prioritize attempting to fix these issues. That being said, I am very open to pull requests fixing Mac-related issues. The fix will have to make sure the wheels are built correctly for Mac.
|
|
I run into a similar issue with version 1.1.0 after running |
|
I had the same problem on OSX-arm64 and the whole problem was building the cutils.pyx, I managed to build it after installing cython, since I'm using conda I installed cython using |
|
Easiest solution so far: conda install -c conda-forge/osx-arm64 cython
pip install git+https://github.com/tommyod/KDEpygit+https://github.com/tommyod/KDEpy can even be added to requirements.txt or environment.yml for pip or conda install respectively. |
|
Tommy, An update on this issue. As I mentioned some months back, I also ran across this. Doing some testing over the weekend, I can no longer replicate with the current KDEpy version. I'm using an M2 Mac (macOS 13.5.1). Testing included the following Python versions:
I tested all 5 environments using a pip install from PyPI and building from the current master branch. All your tests passed in every case. I did notice that all the PyPI installs used the source distribution as there are no arm64 wheels available. Regards, |
Yes please. All help is appreciated. I'll merge PRs :) |
|
Closing this as version 1.1.8 provides wheels for Apple Silicon Macs. We can re-open if anyone continues to have issues. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Hi, I'm having some trouble installing KDEpy on my M1 mac. I've included the install log below. I've seen on pip that you don't provide wheels for M1 ARM processors, but apparently, building from source also fails on this architecture. Am I missing something really obvious?
Log
The text was updated successfully, but these errors were encountered: