test_fontconfig_fonts error on Linux wheel testing

[matthew-brett]

Testing the built wheel, I get the following error (using pytest):

=================================== FAILURES ===================================
__________________________ test_get_fontconfig_fonts ___________________________
[gw0] linux2 -- Python 2.7.6 /venv/bin/python2.7
@pytest.mark.skipif(sys.platform == 'win32', reason='no fontconfig on Windows')
    def test_get_fontconfig_fonts():
>       assert len(get_fontconfig_fonts()) > 1
E       assert 0 > 1
E        +  where 0 = len([])
E        +    where [] = get_fontconfig_fonts()

https://travis-ci.org/MacPython/matplotlib-wheels/jobs/205418240#L623

Should a bare Linux container like the testing container contain any fontconfig fonts? Or is some assumption being made about the configuration of the system?

[tacaswell]

Blah, this test went into to verify that we can parse the output from fontconfig properly.

It is reasonable to expect that a testing system have at least one system-level font installed.

[QuLogic]

Does fc-list --format='%{file}\n' output anything? Does fc-list exist on OSX (or does it need homebrew)?

[QuLogic]

Oops, this is trusty, not OSX; yea I guess maybe a system font should exist there....

[matthew-brett]

fc-list is not installed on the trusty container.

[matthew-brett]

Reasonable to skip this test when fontconfig / fc-list not installed?

[QuLogic]

Interestingly, fontconfig is not listed in the dependencies, optional or not, even though it's used by FontManager regardless of the USE_FONTCONFIG setting. Do the rest of the tests run correctly even without fontconfig?

[matthew-brett]

They appear to : https://travis-ci.org/MacPython/matplotlib-wheels/jobs/205423488

[QuLogic]

Yes, but:

!!!!!!!!!!!! xdist.dsession.Interrupted: stopping after 1 failures !!!!!!!!!!!!!
======== 1 failed, 840 passed, 221 skipped, 8 xfailed in 189.47 seconds ========

That's not even close to all the tests.

[matthew-brett]

Right - but for the 2.0.0 builds, on the same container, the tests did run : https://travis-ci.org/MacPython/matplotlib-wheels/jobs/195613539 . So, at least for 2.0.0, I believe fontconfig was not required.

[tacaswell]

We hit fontconfig (if it is available) and then also search our internal data folders. If the call to fc-list fails, we just treat that as it returning [], thus fc-list is optional, but should be required for testing.

[anntzer]

Closed by #8213, afaict.