Contouring 1x1 array (issue #8197) #8227
Conversation
lib/matplotlib/tests/test_contour.py
Outdated
| try: | ||
| plt.contour([[0]]) | ||
| except TypeError as exc: | ||
| assert exc.args[0] == 'Input z must be at least a 2x2 array.' |
There was a problem hiding this comment.
we have recently changed over to using py.test. You can see how asserts are tested here: http://doc.pytest.org/en/latest/assert.html#assertions-about-expected-exceptions
There was a problem hiding this comment.
Exception-catching tests changed to pytest, including those elsewhere in test_contour.py. Thanks for the link @WeatherGod.
| @@ -309,3 +309,53 @@ def test_contourf_symmetric_locator(): | |||
| locator = plt.MaxNLocator(nbins=4, symmetric=True) | |||
| cs = plt.contourf(z, locator=locator) | |||
| assert_array_almost_equal(cs.levels, np.linspace(-12, 12, 5)) | |||
|
|
|||
|
|
|||
| def test_contour_1x1_array(): | |||
There was a problem hiding this comment.
This test might need a cleanup decorator, because while the plt.contour() calls all fail, figures and axes are still produced prior to throwing the error.
There was a problem hiding this comment.
master does not need or use the cleanup decorator any more; it's all in the py.test fixture.
pytest.ini
Outdated
| @@ -50,11 +50,13 @@ pep8ignore = | |||
| matplotlib/testing/jpl_units/UnitDbl.py E201 E202 E203 | |||
| matplotlib/testing/jpl_units/UnitDblConverter.py E201 E202 E203 E251 E302 E501 E711 | |||
| matplotlib/testing/jpl_units/UnitDblFormatter.py E201 E202 E251 E302 | |||
| matplotlib/tests/test_contour.py E501 | |||
There was a problem hiding this comment.
The PEP8 exceptions list should only get smaller, not bigger. Probably will be fixed by using the py.test method noted by @WeatherGod.
|
Should I squash these commits into one? |
|
@ianthomas23 if you're comfortable with that workflow, go for it. |
d716c7b to
3e23c4e
Compare
|
Squashed commits. |
|
Is this worth backporting to 2.0.x? This seems like a pretty narrow corner that has been broken for a long time? |
|
I would say "no" to backporting. |
There are two underlying problems here. Firstly, we've never bothered to explicitly check if contour/contourf are called with a 2D array that is smaller than 2x2, but have instead relied upon later code to raise an error. There are different code paths for the legacy and new contouring algorithms, hence two different errors reported by OP. Secondly, in the new contour code (and also tricontour too) I haven't correctly handled all errors raised; I've called PyErr_SetString in each case but not necessarily returned -1 or NULL (depending on context) correctly. This shouldn't really matter as all such C++ code is private and covered by checks in the python wrappers - well it should be be, but evidently isn't in this case.
Hence there are two parts to the fix. Firstly, I've added explicit checks in the python contour code for arrays smaller than 2x2, and this is done twice as there are 2 code paths. I've also put a new test in test_contour.py to test that these errors are correctly raised. Secondly, I've corrected _contour_wrapper.cpp and _tri_wrapper.cpp to correctly report errors, and added a set of tests for each so that each raising of a python error in the C++ code is tested at least once.