Re: Building skimage wheels for OS X using Travis CI
On 07/01/2014 10:35 AM, Matthew Brett wrote:
Hi,
On Tue, Jul 1, 2014 at 4:29 PM, Jonathan Helmus <jjhelmus@gmail.com> wrote:
On 07/01/2014 09:42 AM, Matthew Brett wrote:
Hi,
On Tue, Jul 1, 2014 at 1:08 AM, Stéfan van der Walt <stefan@sun.ac.za> wrote:
On Tue, Jul 1, 2014 at 2:01 AM, Matthew Brett <matthew.brett@gmail.com> wrote:
I've forked the repo to the scikit-image organization and added you to the packaging team with the necessary administrative rights. Oops - sorry - I transferred the repo to scikit-image, but now I don't have admin rights. I think you should have those back now.
I put in a PR to give the right credentials for travis and rackspace - I hope. Merged.
Could you enable travis testing on that repo also? It looks like it's activated from my side. Yup - all working : https://travis-ci.org/scikit-image/scikit-image-wheels/builds/28845582
By the way - would you consider uploading the latest built wheels to pypi?
http://a365fff413fe338398b6-1c8a9b3114517dc5fe17b7c3f8c63a43.r19.cf2.rackcdn... Sure, is this the latest wheel 0.10.1 that I pushed out earlier this evening? I'm off to bed for now, but I've also given you maintainer permission on PyPi so that you don't have to wait on me. Thanks - I uploaded the automated builds to pypi (using twine). Any problems - please do let me know.
Cheers,
Matthew
Great work Matthew! Thanks for getting the idea integrated scikit-image's framework, I was busy this weekend finishing a poster for SciPy. One though I had was to use Travis CI's PyPI deploy [1] to upload the wheels directly to pypi. I tested this out a bit last night and the PyPI deploy seems to expect a certain layout of the package, Python install, etc which just leads to trouble. Having the files on Rackspace and having to manually push them to pypi for each release seems reasonable.
Next up, is to use AppVeyor [2] to build wheel files for Windows. I has able to get a wheel for Python 3.3 built [3] but Python 2.7 looked more difficult. Wow - nice discovery, that looks as though it will be very useful.
Will you run into the same problem with conda, that the wheels aren't compatible with Python.org python?
Cheers,
Matthew
I'm not sure about the conda issue. NumPy doesn't provide wheels for windows (yet) as they are still trying to figure out what BLAS library to include. If they did, it would be possible to use the pre-installed Python on the system which appears to be the official Python.org interpreter. The bigger issue is that only Visual C++ 10.0 is installed (VS 2010) which can only build binaries for Python 3.3 and 3.4. VC++ 9.0 is needed for Python 2.6 and 2.7 but AppVeyor doesn't provide it and installing the SDK seems a bit extreme. I'm hoping to chat with folks at SciPy and see if anyone has any good suggestions for Windows based build VMs. - Jonathan Helmus
Hi, On Tue, Jul 1, 2014 at 4:47 PM, Jonathan Helmus <jjhelmus@gmail.com> wrote:
On 07/01/2014 10:35 AM, Matthew Brett wrote:
Hi,
On Tue, Jul 1, 2014 at 4:29 PM, Jonathan Helmus <jjhelmus@gmail.com> wrote:
On 07/01/2014 09:42 AM, Matthew Brett wrote:
Hi,
On Tue, Jul 1, 2014 at 1:08 AM, Stéfan van der Walt <stefan@sun.ac.za> wrote:
On Tue, Jul 1, 2014 at 2:01 AM, Matthew Brett <matthew.brett@gmail.com> wrote:
> > I've forked the repo to the scikit-image organization and added you > to > the packaging team with the necessary administrative rights.
Oops - sorry - I transferred the repo to scikit-image, but now I don't have admin rights.
I think you should have those back now.
I put in a PR to give the right credentials for travis and rackspace - I hope.
Merged.
Could you enable travis testing on that repo also?
It looks like it's activated from my side.
Yup - all working : https://travis-ci.org/scikit-image/scikit-image-wheels/builds/28845582
By the way - would you consider uploading the latest built wheels to pypi?
http://a365fff413fe338398b6-1c8a9b3114517dc5fe17b7c3f8c63a43.r19.cf2.rackcdn...
Sure, is this the latest wheel 0.10.1 that I pushed out earlier this evening? I'm off to bed for now, but I've also given you maintainer permission on PyPi so that you don't have to wait on me.
Thanks - I uploaded the automated builds to pypi (using twine). Any problems - please do let me know.
Cheers,
Matthew
Great work Matthew! Thanks for getting the idea integrated scikit-image's framework, I was busy this weekend finishing a poster for SciPy. One though I had was to use Travis CI's PyPI deploy [1] to upload the wheels directly to pypi. I tested this out a bit last night and the PyPI deploy seems to expect a certain layout of the package, Python install, etc which just leads to trouble. Having the files on Rackspace and having to manually push them to pypi for each release seems reasonable.
Next up, is to use AppVeyor [2] to build wheel files for Windows. I has able to get a wheel for Python 3.3 built [3] but Python 2.7 looked more difficult.
Wow - nice discovery, that looks as though it will be very useful.
Will you run into the same problem with conda, that the wheels aren't compatible with Python.org python?
Cheers,
Matthew
I'm not sure about the conda issue. NumPy doesn't provide wheels for windows (yet) as they are still trying to figure out what BLAS library to include. If they did, it would be possible to use the pre-installed Python on the system which appears to be the official Python.org interpreter. The bigger issue is that only Visual C++ 10.0 is installed (VS 2010) which can only build binaries for Python 3.3 and 3.4. VC++ 9.0 is needed for Python 2.6 and 2.7 but AppVeyor doesn't provide it and installing the SDK seems a bit extreme. I'm hoping to chat with folks at SciPy and see if anyone has any good suggestions for Windows based build VMs.
Yeah - the windows binary problems are really awful: https://github.com/numpy/numpy/wiki/windows-dll-notes https://github.com/numpy/numpy/wiki/Numerical-software-on-Windows Carl Kleffner is currently working on a mingw-w64 toolchain that can compile numpy and scipy with static linking of the c++ mingw-w64 runtime: https://github.com/numpy/numpy/wiki/Mingw-static-toolchain I think that would be the way to go for general wheel building. I've Cc'ed Carl in case he has any thoughts on that. Cheers, Matthew
participants (2)
-
Jonathan Helmus -
Matthew Brett