ENH: PyPy3 testing

[jcpunk]

Is your feature request related to a problem?

With the increasing support for PyPy3 in NumPy, it would be helpful if pandas started testing against PyPy3.

Describe the solution you'd like

Please consider adding pypy3 to the .travis.yml (allowed to fail for now).

API breaking implications

No expected API breaks

Describe alternatives you've considered

This is more of a portability concern than anything else.

Additional context

Something like:

  matrix:
   -  python: pypy3
      env:
        - JOB="pypy3" PATTERN="(not slow and not network and not clipboard)"
  allow_failures:
   -  python: pypy3
      env:
        - JOB="pypy3" PATTERN="(not slow and not network and not clipboard)"

[jreback]

happy to have a PR to do this

just note that allowed_failures are unfortunately not paid as much attention as one would like

[jcpunk]

I don't really understand conda, but took an initial stab in #37665

[lithomas1]

@mattip Do you know when PyPy will make a Python 3.8 alpha release? We are dropping Python 3.7 support in 1.4(the next release which should be released in December).

[mattip]

Thanks for the heads up. Hopefully before December. Are you basing your policies on NEP 29?

[lithomas1]

Yes.

[mattip]

cool

[mattip]

PyPy now has a 3.8 release.

You can get it on github actions by specifying pypy-3.8 in the setup-python action.

You can also get it from pypa/manylinux2014:latest (from 5 days ago): pypy3.8 is in /opt/python/pp38-pypy38_pp73.

On azuure pipelines the images may not include pypy3.8 just yet, see actions/runner-images#4347. Many of the changes have been merged, I don't know how that relates to new image deployment.

[lithomas1]

Thanks @mattip. I'll add the testing in when I have some spare time.