ModifyAxes.inset_axes so it can accept thepolar,projection andaxes_class keywords, and therefore return subclasses ofAxes. My target use-case is with Cartopy, so we can now adaptthis gallery example to do something like

importcartopy.crsasccrsimportmatplotlib.pyplotaspltimportnumpyasnpdefsample_data(shape=(73,145)):"""Return ``lons``, ``lats`` and ``data`` of some fake data."""nlats,nlons=shapelats=np.linspace(-np.pi/2,np.pi/2,nlats)lons=np.linspace(0,2*np.pi,nlons)lons,lats=np.meshgrid(lons,lats)wave=0.75* (np.sin(2*lats)**8)*np.cos(4*lons)mean=0.5*np.cos(2*lats)* ((np.sin(2*lats))**2+2)lats=np.rad2deg(lats)lons=np.rad2deg(lons)data=wave+meanreturnlons,lats,datalons,lats,data=sample_data()ax=plt.subplot(111,projection=ccrs.PlateCarree())ax.contourf(lons,lats,data)ax.coastlines()axins=ax.inset_axes([0.5,0.03,0.47,0.47],projection=ccrs.PlateCarree())axins.contourf(lons,lats,data)axins.set_xlim(-10,5)axins.set_ylim(50,60)axins.coastlines()box,lines=ax.indicate_inset_zoom(axins,edgecolor="black")plt.show()

PR Checklist

Tests and Styling

Has pytest style unit tests (andpytest passes).I'm unsure what extra tests should be added for this. Advice welcome!
IsFlake 8 compliant (installflake8-docstrings and runflake8 --docstring-convention=all).

Documentation

New features are documented, with examples if plot related.
New features have an entry indoc/users/next_whats_new/ (follow instructions in README.rst there).
[N/A] API changes documented indoc/api/next_api_changes/ (follow instructions in README.rst there).
Documentation is sphinx and numpydoc compliant (the docs shouldbuild without error).

rcomer changed the title~~Use fig.add_axes within Axes.inset_axes~~Use Figure.add_axes within Axes.inset_axes

Mar 5, 2022

Copy link

MemberAuthor

rcomer commentedMar 5, 2022

Looks like I have brokenone test consistently across platforms. I will have to come back to this another day.

Copy link

Member

jklymak commentedMar 5, 2022

Inset axes are explicitly children of their parent axes so please don't also add the figure as a parent. I'm not quite sure what the best way to get a projection into an inset_axes is...

Copy link

Contributor

greglucas commentedMar 6, 2022

This seems like a useful thing to have available in some form. Might be worth having a look at#22060 and whetherhttps://github.com/matplotlib/matplotview would work with GeoAxes too...

Ahh, the axes being a child of the figure or not would actually explain why mpl_toolkit inset_axes warn withtight_layout() (because they are added!), but standard inset_axes don't. (xref:#22611)

matplotlib/lib/mpl_toolkits/axes_grid1/inset_locator.py

Line 294 ine75b529

parent_axes.figure.add_axes(inset_axes)

The easiest way might be to allowaxes_class to be passed in to theinset_axes() method, similar to mpl_toolkits and figure's add_axes methods.

matplotlib/lib/mpl_toolkits/axes_grid1/inset_locator.py

Lines 380 to 385 ine75b529

	axes_class : `matplotlib.axes.Axes` type, default: `.HostAxes`
	The type of the newly created inset axes.

	axes_kwargs : dict, optional
	Keyword arguments to pass to the constructor of the inset axes.
	Valid arguments include:

However, there might be some more surgery needed outside of AxesGrid to support the projection or other keyword args.

matplotlib/lib/matplotlib/figure.py

Lines 535 to 538 in09b17b1

	axes_class : subclass type of `~.axes.Axes`, optional
	The `.axes.Axes` subclass that is instantiated. This parameter
	is incompatible with projection and polar. See
	:ref:`axisartist_users-guide-index` for examples.

rcomer mentioned this pull request

Mar 6, 2022

[Bug]: inset_axes position same as parent axes position. Breaks indicate_axes_zoom#22606

Closed

Copy link

MemberAuthor

rcomer commentedMar 6, 2022

Thanks for the feedback. I did have a hunch that it couldn't really be this simple!

Inset axes are explicitly children of their parent axes so please don't also add the figure as a parent.

I'm afraid I don't have the background knowledge to understand this. Is there somewhere I can read up on the matplotlib parent/child relationships and why they matter? I guess the comment abouttight_layout gives a clue.

matplotlibview

Looks useful - thanks! I will have a play with that at some point.

The easiest way might be to allow axes_class to be passed in to the inset_axes() method, similar to mpl_toolkits and figure's add_axes methods.

I'm not so keen on that API as it seems inconsistent with how the user is usually encouraged to create their Axes instances. But I guess it was done in those other functions for a reason.

Anyway it seems that whatever the answer is, it's not this. So for now I will just close this.

rcomer closed this

Mar 6, 2022

Copy link

Member

jklymak commentedMar 6, 2022

@rcomer, maybe convert this to an issue? I think it is reasonable to ask that inset_axes accepts projections, I'm just not sure off the top of my head how to do it (I have limited Matplotlib (and internet) bandwidth at the moment.

rcomer mentioned this pull request

Mar 9, 2022

[ENH]: enable passing of projection keyword to Axes.inset_axes#22630

Closed

Copy link

MemberAuthor

rcomer commentedMar 9, 2022

maybe convert this to an issue?

Issue now at#22630. Thanks!

rcomer reopened this

Mar 18, 2022

rcomer marked this pull request as draft

March 18, 2022 16:12

Copy link

MemberAuthor

rcomer commentedMar 18, 2022

The important functionality seems to be contained withinFigure._process_projection_requirements, so I tried using that directly. The Cartopy example still works, so I just wanted to see how it would fly with the CI.

rcomer closed this

Mar 25, 2022

Copy link

Member

jklymak commentedMar 25, 2022

@rcomer, did this not work? It looks like something similarshould work, and it is certainly reasonable functionality...

Copy link

MemberAuthor

rcomer commentedMar 25, 2022

@jklymak yes the CI all seemed fine and it works for my Cartopy example. However at#22630 @mylasem said they were working on it, so I thought I would leave it to them for now. I'd be happy to pick this up again at some point if@mylasem's version doesn't land for whatever reason.

Copy link

Member

jklymak commentedMar 25, 2022

In general, Matplotlib does not have a "reserve an issue" policy, and this PR predates@mylasem 's comment. I think they should work within the context of this PR if at all possible.

rcomer reopened this

Mar 31, 2022

rcomer changed the title~~Use Figure.add_axes within Axes.inset_axes~~Axes.inset_axes: enable Axes subclass creation

Mar 31, 2022

Copy link

MemberAuthor

rcomer commentedMar 31, 2022

OK, since@mylasem has not commented further and I had some time, I have:

Added some basic tests to verify that the expected type is returned for our three methods of specifying it.
Updated theAxes.inset_axes docstring with the keyword descriptions - this was a straight copy/paste from theFigure.add_axes docstring, and then added the word "inset".
Updated the what's new entry (and the PR title and OP) to reflect that we're not going viaFigure.add_axes.

rcomer commented

Mar 31, 2022

View reviewed changes

lib/matplotlib/tests/test_axes.py Outdated

		@@ -6614,6 +6614,30 @@ def test_zoom_inset():
		axin1.get_position().get_points(), xx, rtol=1e-4)


		def test_inset_polar():
		from matplotlib.projections.polar import PolarAxes

Copy link

MemberAuthor

rcomerMar 31, 2022•
edited
Loading

Choose a reason for hiding this comment

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

I imported the class within the test as I noticed another test in this module does that. If that's not necessary, then I think we could parametrize these three tests into one.

Copy link

Member

tacaswellApr 4, 2022

Choose a reason for hiding this comment

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

I do not see any reason to do this, I think pulling all of the imports to the top is OK.

Copy link

MemberAuthor

rcomerApr 5, 2022

Choose a reason for hiding this comment

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

I moved the imports to the top, but in the end couldn't convince myself it was worth parametrizing the two small tests that are still the same: since they're so small there isn'tmuch repetition going on and I suspect readability is better as they are than with a parametrized version.