Can macOS Apple Silicon binary wheels be provided? #1263
Replies: 13 comments 20 replies
-
|
Use only wheels for the installation by running |
Beta Was this translation helpful? Give feedback.
-
|
Good point. I've updated the original text to include this. |
Beta Was this translation helpful? Give feedback.
-
|
We are up to beta 3 now... |
Beta Was this translation helpful? Give feedback.
-
|
@Carreau can you try building this release? |
Beta Was this translation helpful? Give feedback.
-
|
Sure, I'm not a pyodbc user, but I have a M1, happy to run any test you wish. Installing works, importing not, WRT to building wheel, have you tried https://cibuildwheel.readthedocs.io/en/stable/ ? I'm going to try to build the wheel locally, have have to ValueError currently, I'm going to try to narrow down. [Edited] Edit above error I forgot the --pre |
Beta Was this translation helpful? Give feedback.
-
|
Rebuilding works, I needed to install |
Beta Was this translation helpful? Give feedback.
-
|
And feel free to ping me on issues, or patches where you need someone to run some code to try. |
Beta Was this translation helpful? Give feedback.
-
|
Also, I forgot : MacBook Pro 14-inch, 2021, M1, Mac OS Ventura 13.4 (22F66) |
Beta Was this translation helpful? Give feedback.
-
|
I was able to successfully install, but it failed upon import. 16-inch, 2021 MacBook Pro, M1 Pro, Ventura 13.5 |
Beta Was this translation helpful? Give feedback.
-
|
Do you have I believe installing |
Beta Was this translation helpful? Give feedback.
-
|
This error message is a wheel-compatibility issue.
so:
|
Beta Was this translation helpful? Give feedback.
-
|
Thanks for the clarification, the side scroll to see the error got me...you win again lines greater than 100 characters long. |
Beta Was this translation helpful? Give feedback.
-
|
Yes, I do have unixodbc installed via brew. Im a regular pyodbc user, I use it daily. I have not installed unixodbc specifically through my conda/python environment via pip or conda. But I have never had to in the past, the installation through brew has always worked. |
Beta Was this translation helpful? Give feedback.
-
|
Great, @Carreau figured out that you came across a wheel issue. Can you run |
Beta Was this translation helpful? Give feedback.
-
|
On a MacBook Pro with Apple M2 Pro chip, I ran the following commands: mamba create -n pyodbc python=3.11 unixodbc pip
conda activate pyodbc
python -m pip install --only-binary=:all: pyodbc==5.0.0b3
pythonWithin the python REPL, I ran: import pyodbcI got an error. Here's the full REPL log: |
Beta Was this translation helpful? Give feedback.
-
|
Reading through issue #1124 I'm wondering if setting the export LDFLAGS="-L$CONDA_PREFIX/lib"
export CPPFLAGS="-I$CONDA_PREFIX/include"all the steps# Create a conda environment and activate it.
conda create --name pyodbc python=3.11 pip unixodbc
conda activate pyodbc
# Set the flags.
export LDFLAGS="-L$CONDA_PREFIX/lib"
export CPPFLAGS="-I$CONDA_PREFIX/include"
# Install pyodbc.
python -m pip install --only-binary=:all: pyodbc==5.0.0b3
# Try to import pyodbc and print the version.
python -c "import pyodbc; print(pyodbc.version)" |
Beta Was this translation helpful? Give feedback.
-
|
Does not works, and I'm not sure how it could as LDFLAGS and CPPFLAGS will only work if you do |
Beta Was this translation helpful? Give feedback.
-
|
Thx, I'll read through the linked issue more closely. I just scanned it and read that it worked for some people, but didn't read the |
Beta Was this translation helpful? Give feedback.
-
|
No worries, it # in setup.py
pipe = os.popen('odbc_config --cflags --libs 2>/dev/null')
cflags, ldflags = pipe.readlines()
...
settings['extra_compile_args'].extend(shlex.split(cflags))
settings['extra_link_args'].extend(shlex.split(ldflags))Those are in-general flags/options that only are effective at build time, not run time. Unlike In #1124 it is likely that because the arm wheel was not built, pip would download the tgz and build locally by default. But here we know that |
Beta Was this translation helpful? Give feedback.
-
|
|
Beta Was this translation helpful? Give feedback.
-
|
I build the macos arm wheel beta3 for 3.8, 3.9, 3.10, 3.11: If you wish to try the wheel you can find them here: https://github.com/Carreau/test-whl My guess is they will work on conda, but I'm curious if they will work on brew-python. If that works, maybe someone that know more about how to compare .so objects can figure out the problem ? |
Beta Was this translation helpful? Give feedback.
-
|
Run |
Beta Was this translation helpful? Give feedback.
-
|
That makes sense with those errors - somehow the non-working one is not getting linked with unixODBC. |
Beta Was this translation helpful? Give feedback.
-
|
Ha! increasing verbosity We are likely finding the intel brew instead of the aarch64 ! |
Beta Was this translation helpful? Give feedback.
-
|
That is the kind of info we are looking for. Perfect. Perhaps it is this: I'm not sure why it is here. @keitherskine Do you know anything about this? We need to include more links to docs, blogs, etc. for this or I forget ;) |
Beta Was this translation helpful? Give feedback.
-
|
I think I see what is going on. Normally when building the wheel the libraries get bundled, but since we are cross compiling the unixODBC dylibs will be Intel, not ARM. I was just preparing a build with this commented out, but I'm going to revert that and look into how we can get the right unixODBC platform. |
Beta Was this translation helpful? Give feedback.
-
I tried a fork with that removed and it does not help, as far as I understand if you unset this it will try to statically link the libraries. I think this is for packages that need systems dependencies, that can't be expressed in pip. I found the following text here I did not manage to get a correct brew even using CIBW_BEFORE_ALL_MACOS, maybe we need to build unixODBC ? As far as I remember arm64 brew is not on the same path than intel one, I belive so that you can install both, but I don't remember where I read this info. My homebrew in in |
Beta Was this translation helpful? Give feedback.
-
|
Yes, if you remove One potential solution here is to use a macOS runner to build the macOS distributable: |
Beta Was this translation helpful? Give feedback.
-
|
Do you mean self-hosted runners ? Because github-hosted is already what I belive this is using, and it does not list ARM... Nnumpy seem to be using cirrus_ci that supports macos arm64, see their config file |
Beta Was this translation helpful? Give feedback.
-
|
Ah, yes, @Carreau . You're quite right. We would have to use a GitHub Actions self-hosted runner for ARM64, as you say. Also, thank you for the reference to Cirrus CI. Good to know. |
Beta Was this translation helpful? Give feedback.
-
|
It was brought to my attention that github now have M1 runners in beta: https://github.blog/2023-10-02-introducing-the-new-apple-silicon-powered-m1-macos-larger-runner-for-github-actions/ Maybe it can be used to build a working wheel. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks @Carreau. It has now been officially released and @keitherskine has updated the workflow to use the new runner. Version 5.1.0 has just been released with binaries for both Intel & ARM. We'd appreciate feedback from anyone using pyodbc on Macs. |
Beta Was this translation helpful? Give feedback.
-
We're retrying macOS Apple Silicon wheels, for those with M1, M2, etc. chips. If you have one of these, please give this release a try so we know before the final 5.0.0 release in a few days.
This means you would not use Rosetta. Just install normally using
python -m pip install --only-binary=:all: pyodbc==5.0.0b3.For those wondering, the inability to supply these wheels has not been due to any pyodbc code issues. We use Github Actions to build all the different versions and it still doesn't have complete support for macOS Apple Silicon CI runners. At this time, only Intel is supported. However, we can cross-compile and generate ARM wheels on Intel, but we can't test them. That's why we need your help!
Beta Was this translation helpful? Give feedback.
All reactions