MNT: Compat with pytest 5.4.0

[pllim]

Fix #94

Due to pytest-dev/pytest#5975

[bsipocz]

failure is related, I can come back in the evening, but not earlier.

[pllim]

Update: I don't have enough pytest-fu to fix the error. Looking at the code, the thing that seems most intuitive is to have DocTestTextfilePlus inherit from DoctestTextfile but doing that broke other things. I tried several other hacks including copying over a from_parent signature with fspath but that didn't work either. I also tried switching the two parent classes of DocTestTextfilePlus.

https://github.com/pytest-dev/pytest/blob/d7f01a90eba4b3bde06e7827ccaf12252f6116d6/src/_pytest/doctest.py#L364

I wonder if @RonnyPfannschmidt or @nicoddemus has any insights for us.

Relevant CI log: https://travis-ci.org/github/astropy/pytest-doctestplus/jobs/662066066

    x = self._doctest_textfile_item_cls.from_parent(
E   TypeError: from_parent() got an unexpected keyword argument 'fspath'

[nicoddemus]

Hi folks,

I will wait for @RonnyPfannschmidt to chime in as he is more familiar with the from_parent changes. 👍

[RonnyPfannschmidt]

i took a initial look, inheriting from both a item and a collector like module was broken to begin with,
instead of item,

as far as i can tell doctestplus changes a assumption in tying to run a complete file as test,
it should use 2 nodes here, one for collecting the file, one for running the single big test

combining test items and file collectors works purely by accident, the new code added a to the point constructor for the test item which broke the file collector bits that came in as well

[pllim]

Thanks, @RonnyPfannschmidt . But, oh dear, sounds like a complete refactoring is needed to make this compatible and I am not sure if I am up for the task.

[pllim]

This is already fixed in #101 and should be gone after rebase.

[pllim]

This does not work because I think now Python's doctest tries to read in the docstring as it is defined in the test file, not via testdir fixture. I stared at the code here, the code in pytest, and Python doc, but couldn't figure out a path forward.

Simply renaming the old runtest to collect doesn't work either:

https://github.com/astropy/pytest-doctestplus/blob/9975e23d178465c00b75f6b2f178f89e2245b540/pytest_doctestplus/plugin.py#L218

[pllim]

Wait a minute, my local test is totally different from Travis CI results. I cannot even get the tests to collect. My brain hurts.

$ pytest  --doctest-plus
============================= test session starts ==============================
platform linux -- Python 3.7.3, pytest-5.4.1, py-1.8.0, pluggy-0.12.0
Matplotlib: 3.1.1
Freetype: 2.10.0
rootdir: .../astropy/pytest-doctestplus, inifile: setup.cfg, testpaths: tests, pytest_doctestplus
plugins: remotedata-0.3.1, openfiles-0.3.2, arraydiff-0.3, astropy-header-0.1.2, forked-1.1.3, xdist-1.30.0, mpl-0.11, cov-2.8.1, filter-subpackage-0.1.1, hypothesis-5.5.2, ci-watson-0.0.0.dev76+g4b73d3b5, asdf-2.4.0.dev63+gb67cb02, doctestplus-0.6.0.dev0
collected 28 items / 1 error / 27 selected                                     

==================================== ERRORS ====================================
__________________ ERROR collecting tests/test_doctestplus.py __________________
pytest_doctestplus/plugin.py:231: in collect
    test = parser.get_doctest(text, globs, name, filename, 0)
.../doctest.py:668: in get_doctest
    return DocTest(self.get_examples(string, name), globs,
.../doctest.py:682: in get_examples
    return [x for x in self.parse(string, name)
pytest_doctestplus/plugin.py:268: in parse
    result = doctest.DocTestParser.parse(self, s, name=name)
.../doctest.py:644: in parse
    self._parse_example(m, name, lineno)
.../doctest.py:714: in _parse_example
    lineno + len(source_lines))
.../doctest.py:800: in _check_prefix
    (lineno+i+1, name, line))
E   ValueError: line 30 of the docstring for test_doctestplus.py has inconsistent leading whitespace: '    """'
=========================== short test summary info ============================
ERROR tests/test_doctestplus.py - ValueError: line 30 of the docstring for te...
!!!!!!!!!!!!!!!!!!!! Interrupted: 1 error during collection !!!!!!!!!!!!!!!!!!!!
=============================== 1 error in 0.22s ===============================

[pllim]

Closing in favor of #103 . Thanks, @saimn !