Suppress/fix some test warnings. #11705
Conversation
36d6dde to
886be24
Compare
|
This is failing on windows which does not really have a float128 type (iirc). |
886be24 to
f936ce8
Compare
|
should be fixed |
f936ce8 to
da6fd6a
Compare
| @@ -1112,7 +1112,8 @@ def _transform(self, a): | |||
| """ | |||
| Inplace transformation. | |||
| """ | |||
| masked = np.abs(a) > self.linthresh | |||
| with np.errstate(invalid="ignore"): | |||
There was a problem hiding this comment.
Why does this throw an invalid warning in the first place? It's not obvious to me that _transform should be getting invalid values.
There was a problem hiding this comment.
You can definitely pass nans in (they're ignored at a later stage, by the low-level path handling machinery).
| assert res == is_opentype_cff_font(f) | ||
| for f in fontManager.ttflist: | ||
| if 'otf' in f.name: | ||
| with open(f.name, 'rb') as fd: |
There was a problem hiding this comment.
This is not hit when the tests are run. Maybe we have to otf fonts on travis?
There was a problem hiding this comment.
Dunno, and I can't say I really care about this test given that it's a literaly copy-paste of the function's definition (https://github.com/matplotlib/matplotlib/blob/master/lib/matplotlib/font_manager.py#L1267).
There was a problem hiding this comment.
As discussed on this week's call, this is actually a real drop in coverage due to a typo (f.name vs. f.fname)
da6fd6a to
f1aed96
Compare
|
fixed |
The only remaining test that throws a warning is
test_image.py::test_full_invalid ("converting a masked element to nan"),
which I didn't suppress in the test because I think that should indeed
not throw (we don't throw warnings when passing partially-nan data
either).
f1aed96 to
c9a8edb
Compare
|
Anyone know why the csd test would be failing? |
|
Probably something with numpy 1.15 being just released? |
The only remaining test that throws a warning is
test_image.py::test_full_invalid ("converting a masked element to nan"),
which I didn't suppress in the test because I think that should indeed
not throw (we don't throw warnings when passing partially-nan data
either).
(I think ultimately we should make CI fail in the presence of warnings;
looks like this can be done by adding the suitable warnings filter to
pytest.ini https://docs.pytest.org/en/latest/warnings.html.)
PR Summary
PR Checklist