Description
CI for more "exotic" platforms like ARM is on our roadmap. Since there's a free ARMv8-a machine pool available from shippable I went ahead and drafted a feature branch in my NumPy fork that contains a shippable.yml with a few basic commands to build & test NumPy.
The console results are available here, and hopefully you can see those. In short, python runtests.py seems to build without problems and produce this test output:
1 failed, 4627 passed, 406 skipped, 87 deselected, 10 xfailed in 226.59 seconds
There are probably various discussion points, including whether we really want to do this, but a few initial observations:
- the above is with Python 2.7 because there seem to be some challenges getting pre-installed 3.6 to work by default with their ARM shared pool image
- seems like a lot of skipped tests above & we'd have to make sure that the default virtual env isolation is appropriate and there's no other NumPy system installation causing issues, etc. (I think we are very cautious about this stuff on Travis)
- cython is (so far) the only dependency that I pip installed so would indeed have to check the environment stuff quite carefully to see what is preloaded or prevent interaction with it
- I did have to email them to get permission to access the free pool, but they were very fast / responsive (minutes usually so far)
Other Note: I didn't initiate this as a PR directly to NumPy so that I could avoid trying to register the actual NumPy organization with the service until i.e. core devs / steering committee have weighed in, etc.