The first commit fixes the nonsensical transform mentioned in#14899, while changing the test introduced in#14727 to match the behavior of autoscale which we decided in#14899 to keep at least for now (as refactoring it is going to require more work).

The second commit fixes the fact that I missed, in#14727, that I could just writetrf.transform(array) rather thanmap(trf.transform, array), i.e.Transform.transform works on arrays (and is documented to do so).

attn@efiring

PR Checklist

Has Pytest style unit tests
Code isFlake 8 compliant
New features are documented, with examples if plot related
Documentation is sphinx and numpydoc compliant
Added an entry to doc/users/next_whats_new/ if major new feature (follow instructions in README.rst there)
Documented in doc/api/api_changes.rst if API changed in a backward-incompatible way

Fix nonsensical transform in mixed-mode axes aspect computation.

418a1ad

anntzer added the Release criticalFor bugs that make the library unusable (segfaults, incorrect plots, etc) and major regressions. label

Aug 6, 2019

anntzer added this to thev3.2.0 milestone

Aug 6, 2019

efiring approved these changes

Aug 6, 2019

View reviewed changes

Copy link

Member

efiring left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others.Learn more.

You might want to either remove the comment in the test, or modify the test so that it tests the zoomed-in case where the requested view limits are within larger data limits. This need not block the PR, though.

lib/matplotlib/tests/test_axes.py Outdated

		aspect=1, adjustable="datalim")
		ax.margins(0)
		ax.apply_aspect()
		assert ax.get_xlim() == pytest.approx(np.array([1/10, 10]) * np.sqrt(10))
		# Currently the autoscaler chooses to reduce the x-limits by half a decade
		# on each end, but this may change later.

Copy link

Member

efiringAug 6, 2019

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others.Learn more.

I don't know why it would, unless the test itself is changed. I think the point of confusion is that in this test you are plotting only one point, so apply_aspect is understandably and correctly shrinking an axis instead of expanding one. xlim and ylim pertain to view limits, not data limits. It looks like I made a big mistake, long ago, in calling the adjustable option "datalim", because it is not actually adjusting data limits--they remain determined by the actual plot calls and Artist additions--but rather, view limits. Maybe we should change the option name (with deprecation, in a separate PR) to "viewlim" or "viewlimits"?