TST: Test no-file for source

[larsoner]

Not sure this is the best solution but it at least allows things in MNE-Python to pass. Turns out the real issue was us subclassing list and that having to associated file. But I added another test that failed a different way along the way as well for where we create functions dynamically.

cc @stefmolin @stefanv since you've probably thought about this more than I have! And @jarrodmillman since we should probably get some form of this in for 1.6.

BTW the time to test docstrings in MNE-python went from this before #476:

5.15s call     mne/tests/test_docstring_parameters.py::test_docstring_parameters

to:

49.58s call     mne/tests/test_docstring_parameters.py::test_docstring_parameters

So we should indeed prioritize time/efficiency at some point soon :)

[stefmolin]

I would tweak this slightly.

numpydoc_ignore_comments = {}

try:
    with open(filepath) as file:
        last_declaration = 1
        declarations = ["def", "class"]
        for token in tokenize.generate_tokens(file.readline):
            if token.type == tokenize.NAME and token.string in declarations:
                last_declaration = token.start[0]
            if token.type == tokenize.COMMENT:
                match = re.match(IGNORE_COMMENT_PATTERN, token.string)
                if match:
                    rules = match.group(1).split(",")
                    numpydoc_ignore_comments[last_declaration] = rules
except (OSError, TypeError):  # can be None, nonexistent, or unreadable
    pass

return numpydoc_ignore_comments

[larsoner]

I actually prefer the current approach as I always try to limit the scope of try/except as much as possible. With your code there is a single return path, which is nice. However the scoping for the try/except is broader, which means you risk accidentally catching a TypeError somewhere in the remaining 10 lines that do more than just "try to open the file". And I don't think the interpretation of the current code is too bad: "try to open the file and if you can't, return an empty dict. if you can, proceed to parse it"

[larsoner]

I also realized that extract_ignore_validation_comments was being called for every :py:obj: to be validated, and that this is probably a big overhead when you only need to do it once per source file (as each of those could have tens/hundreds/thousands(?) of docstrings in them). Based on this:

$ find mne-python/mne -name "*.py" | wc -l
609
$ find matplotlib/lib/ -name "*.py" | wc -l
290
$ find pandas/pandas/ -name "*.py" | wc -l
1376
$ find scipy/scipy/ -name "*.py" | wc -l
924
$ find numpy/numpy/ -name "*.py" | wc -l
495

I added a @functools.lru_cache(maxsize=2000) to extract_ignore_validation_comments. This dropped the time from ~50s to ~10s for MNE-Python, which seems well worth the slight memory usage increase required to store the output dicts.

[jarrodmillman]

Thanks!