Reuse InsetLocator to make twinned axes follow their parents. #18456
Conversation
|
I think this idea is great, but can we rename |
|
I renamed the thing to TransformedBoundsLocator, given that it locates an axes using tranformed bbox bounds. |
|
The tests help codify expected behaviour. If someone rewrites all of this from scratch it would be nice if this behaviour would be explicitly tested. |
|
My point is that the behavior is already codified in all the tests for twinned axes, and that it would really require conscious malfeasance :-) to go back to the old behavior (you need to put back the manual position tracking). |
|
? you have a case that was broken for a previous implementation. This implementation presumably fixes that case, so having a test would be preferable. |
|
It was broken because of some rather roundabout way of trying to keep track of twinning (by equating twinning with sharing), which I am arguing is very unlikely to be reintroduced. |
|
@anntzer You've been able to change library behavior without a test failing. That's bad. Therefore, we should have a test. Given that the test should be all of 6 lines, that doesn't seem like a huge maintenance burden for the project to assume. |
|
sure... |
| @@ -6513,3 +6513,12 @@ def test_multiplot_autoscale(): | |||
| ax2.axhspan(-5, 5) | |||
| xlim = ax1.get_xlim() | |||
| assert np.allclose(xlim, [0.5, 4.5]) | |||
|
|
|||
|
|
|||
| def test_sharing_does_not_link_positions(): | |||
There was a problem hiding this comment.
does not or does? I'm confused, as the final assert checks for equality.
There was a problem hiding this comment.
it's checking that the position of the shared axes did not change (i.e. that it did not get linked to the position of the sharee).
From a positioning PoV, a twinned axes is just like an inset axes whose position exactly matches the parent's position. Doing so removes the need for the heuristic in `Figure.subplots_adjust`/`GridSpec.update` where a non-gridspec-managed Axes would track the position of any Axes with which it shared either xaxis or yaxis, which was a proxy for twinning (per 721b949). This would cause incorrect behavior in rare cases such as ```python from pylab import * ax = subplot(221) axes([.6, .6, .3, .3], sharex=ax) subplots_adjust(left=0) ``` where the `subplots_adjust` call would make the second axes go on top of the first.
From a positioning PoV, a twinned axes is just like an inset axes whose
position exactly matches the parent's position. Doing so removes the
need for the heuristic in
Figure.subplots_adjust/GridSpec.updatewhere a non-gridspec-managed Axes would track the position of any Axes
with which it shared either xaxis or yaxis, which was a proxy for
twinning (per 721b949). This would cause incorrect behavior in rare
cases such as
where the
subplots_adjustcall would make the second axes go on top ofthe first.
PR Summary
PR Checklist
pytestpasses).flake8on changed files to check).flake8-docstringsandpydocstyle<4and runflake8 --docstring-convention=all).doc/users/next_whats_new/(follow instructions in README.rst there).doc/api/next_api_changes/(follow instructions in README.rst there).