Travis fails for 2.7 with EXTRA_ and PRE_ wheels

[yarikoptic]

That wheel it tries to install doesn't work for me either:

$> venvs-pandas-whl/bin/pip install https://7933911d6844c6c53a7d-47bd50c35cd79bd838daf386af554a83.ssl.cf2.rackcdn.com/pandas-0.24.0.dev0+20181030041926_865.ga2e599499-cp27-cp27mu-manylinux1_x86_64.whl
Collecting pandas==0.24.0.dev0+20181030041926-865.ga2e599499 from https://7933911d6844c6c53a7d-47bd50c35cd79bd838daf386af554a83.ssl.cf2.rackcdn.com/pandas-0.24.0.dev0+20181030041926_865.ga2e599499-cp27-cp27mu-manylinux1_x86_64.whl
  Downloading https://7933911d6844c6c53a7d-47bd50c35cd79bd838daf386af554a83.ssl.cf2.rackcdn.com/pandas-0.24.0.dev0+20181030041926_865.ga2e599499-cp27-cp27mu-manylinux1_x86_64.whl (10.0MB)
    100% |████████████████████████████████| 10.0MB 2.9MB/s 
Collecting python-dateutil>=2.5.0 (from pandas==0.24.0.dev0+20181030041926-865.ga2e599499)
  Downloading https://files.pythonhosted.org/packages/74/68/d87d9b36af36f44254a8d512cbfc48369103a3b9e474be9bdfe536abfc45/python_dateutil-2.7.5-py2.py3-none-any.whl (225kB)
    100% |████████████████████████████████| 235kB 3.0MB/s 
Collecting numpy>=1.12.0 (from pandas==0.24.0.dev0+20181030041926-865.ga2e599499)
  Downloading https://files.pythonhosted.org/packages/9e/eb/c9eda9f4865d669e0bb37ce5c780e86c63daa54ca827b95a171429012d08/numpy-1.15.3-cp27-cp27mu-manylinux1_x86_64.whl (13.8MB)
    100% |████████████████████████████████| 13.8MB 2.0MB/s 
Collecting pytz>=2011k (from pandas==0.24.0.dev0+20181030041926-865.ga2e599499)
  Downloading https://files.pythonhosted.org/packages/f8/0e/2365ddc010afb3d79147f1dd544e5ee24bf4ece58ab99b16fbb465ce6dc0/pytz-2018.7-py2.py3-none-any.whl (506kB)
    100% |████████████████████████████████| 512kB 1.6MB/s 
Collecting six>=1.5 (from python-dateutil>=2.5.0->pandas==0.24.0.dev0+20181030041926-865.ga2e599499)
  Using cached https://files.pythonhosted.org/packages/67/4b/141a581104b1f6397bfa78ac9d43d8ad29a7ca43ea90a2d863fe3056e86a/six-1.11.0-py2.py3-none-any.whl
Installing collected packages: six, python-dateutil, numpy, pytz, pandas
Successfully installed numpy-1.15.3 pandas-0.24.0.dev0+865.ga2e599499 python-dateutil-2.7.5 pytz-2018.7 six-1.11.0
venvs-pandas-whl/bin/pip install   4.75s user 0.71s system 54% cpu 9.987 total
1 15135.....................................:Tue 30 Oct 2018 06:56:43 PM EDT:.
hopa:/tmp
$> venvs-pandas-whl/bin/python -c 'import pandas'
Traceback (most recent call last):               
  File "<string>", line 1, in <module>
  File "/tmp/venvs-pandas-whl/local/lib/python2.7/site-packages/pandas/__init__.py", line 35, in <module>
    "the C extensions first.".format(module))
ImportError: C extension: No module named enum not built. If you want to import pandas from the source directory, you may need to run 'python setup.py build_ext --inplace --force' to build the C extensions first.

and I also wondered what are those extra wheels collections added 96f1dac in

  global:
    - EXTRA_WHEELS="https://5cf40426d9f06eb7461d-6fe47d9331aba7cd62fc36c7196769e4.ssl.cf2.rackcdn.com"
    - PRE_WHEELS="https://7933911d6844c6c53a7d-47bd50c35cd79bd838daf386af554a83.ssl.cf2.rackcdn.com"

really are (adding a comment why each one is needed would be nice).

May be those runs should get added to allow_failures: section of .travis.yml so they do not "fail" the PRs (it seems everyone now would just ignore them)?

[effigies]

Those wheels repositories are stolen from nibabel's. I believe they have nightly builds of a lot of scientific packages, but I haven't dug too deeply. @matthew-brett might know off the top of his head.

Agree we should add comments.

As to allowed failures, I'm okay with that in principle, though I worry that people won't even see that they're failing if the check on GitHub is green.

[yarikoptic]

And now they see red all the time and can't do anything about it, can't merge in Mobile view... Not sure what is better ;-)

[matthew-brett]

EXTRA_WHEELS are for wheels of stable packages that the main packages don't provide, for some reason. Often it's for older versions of packages, or no-longer-supported platforms - like 32-bit Matplotlib.

PRE_WHEELS is for nightly and release-candidate wheels. It's particularly useful for Numpy, Scipy and Pandas builds, so you can see if you're going to break with the upcoming releases.

[effigies]

Thanks, Matthew. @yarikoptic We can try to put the pre-release builds in allow_failures and see if it bites us.

[tyarkoni]

As I've said before, I'd really like to stop supporting Python 2.7... I believe the informal agreement from #57 and #86 was that the 2.7 lovers among us (cough, @mih) agreed to take on the responsibility of fixing 2.7-related bugs as they arose... if nobody's going to open a PR for this, can I just drop 2.7 from the matrix and solve the problem that way?

[effigies]

I've opened an issue in MacPython/pandas-wheels#36. This seems to be a Pandas build error, not anything actually 2.7 related.

[yarikoptic]

+1 for not dropping 2.7 just yet, but soon! indeed while issues aren't per se 2.7 related IMHO would be nice to keep compatibility going.
shame on us with @mih - pymvpa is not yet python3 proper yet :-( although we do not use pybids directly there yet, some time we will. not sure if it would be python3 support or pybids use which would come first ;)

[tyarkoni]

Any chance a fix is imminent, @yarikoptic, @mih? It would be nice to have working tests again... how about we just drop the pre-wheels line but keep the other 2.7 options? (This will presumably fail once pre becomes current, though.)

[effigies]

I don't think this is a @yarikoptic, @mih issue. I think @TomAugspurger is the maintainer of the pandas nightly builds.

[tyarkoni]

Oh, sorry, missed your earlier comment that it's a Pandas build error. Well, in that case, I'd still suggest commenting out the PRE line until it's fixed. Does that sound reasonable?

[effigies]

I would rather move it to allow_failures:, as @yarikoptic suggested, as that will at least let us see when it's fixed.