I have to install my wxpython correctly before I can continue the review. Soon...

@fariza : That should not be too hard. If you're on Linux, then you should look here, as otherwise PIP will build from sources:https://wxpython.org/pages/downloads/

@anntzer can you help reviewing this?

Would be much happier if (e.g.)https://github.com/matplotlib/matplotlib/blob/master/lib/matplotlib/tests/test_backends_interactive.py was modified to add testing for the tool manager (right now there's no CI at all on any of this). Note that this could be done independently of (and before) this PR.

For me, that's currently out of scope. I'm not familar enough with this yet.

(I'm fine if others want to review this and get it in; I would just be more comfortable if this was tested by CI. Otherwise, I would want to find time to test all the features myself, which I don't have the bandwidth for right now.)

Needs a rebase.

Let's just try to keep things moving and push back the discussion to when backend_tools becomes the default, if that ever happens...

There is a problem. I think it is in the doc building steps

autodoc: failed to import module 'matplotlib.backends.backend_wxagg'; the following exception was raised:Traceback (most recent call last):  File "/home/circleci/.local/lib/python3.6/site-packages/sphinx/ext/autodoc/importer.py", line 140, in import_module    __import__(modname)  File "/home/circleci/project/lib/matplotlib/backends/backend_wxagg.py", line 7, in <module>    from .backend_wx import (  File "/home/circleci/project/lib/matplotlib/backends/backend_wx.py", line 1795, in <module>    class StatusbarWx(StatusbarBase, wx.StatusBar):TypeError: metaclass conflict: the metaclass of a derived class must be a (non-strict) subclass of the metaclasses of all its bases

Unfortunately, I don't know how to interpret this message or how to fix it.
There are no metaclasses involved.
The inheritance structure forStatusbarWx is the same as e.g. forToolbarWx which does not trigger the message:

class ToolbarWx(ToolContainerBase, wx.ToolBar):    def __init__(self, toolmanager, parent, style=wx.TB_HORIZONTAL):        ToolContainerBase.__init__(self, toolmanager)        wx.ToolBar.__init__(self, parent, -1, style=style)   ...class StatusbarWx(StatusbarBase, wx.StatusBar):    def __init__(self, parent, *args, **kwargs):        StatusbarBase.__init__(self, *args, **kwargs)        wx.StatusBar.__init__(self, parent, -1)

That's due to the slightly atrocious mocking we use to make docs build even though wx is not installed (athttps://github.com/matplotlib/matplotlib/blob/master/doc/sphinxext/mock_gui_toolkits.py#L111). The solution is to remove the API docs for wx (similarly tohttps://github.com/matplotlib/matplotlib/pull/9993/files#diff-dd295241396a7490c85d37c53ae65dc2) and (not structly necessary, but may as well) remove the mocking mentioned above.
As argued elsewhere, the docs for the GUI integrations should be replaced by a single hand-written docs documenting the common API (and critical differences, when applicable) of all toolkits anyways...

This change makes the test pass, but then the documentation would be gone, right?
Do you think that adding aStatusBar toclass MyWX(MagicMock) would work?
That seems to be the difference betweenToolBar andStatusBar.

OK, I've added theStatusBar. The two ci/circleci still fail, but with something where I would guess it's not related to the PR.

The doc build failure is#11107, please fetch, rebase, and push.
By the way, feel free to remove the "needs rebase" label after a rebase, makes it easier to keep track of PR status.

I'm still not familiar with git and still looking for a better GUI tool. I probably should have used "fetch" instead of "pull". The commit history looks like a mess, but the code diff is correct.

This did re-trigger the ci/circleci, but they still fail....

You have extra commits because you did some weirdness when rebasing. There are actuallytwo copies of your commits now, one ending ate68cfac that's the original, and one ending ata601003 that was rebased yesterday.

When you rebase and try to push, git tells you to fetch & merge (or pull) first. While this is the safe option, it's also wrong in this case because it pulls the old copy of your commits. Youwant to replace the old commits with the rebased ones, so you need to dogit push --force origin wxtoolmanager.

This is not too difficult to fix (these instructions assume you have no other commits or changes):

git checkout wxmanagergit reset --hard a60100352f12916541a4a7521d7a27a1ecf50696git push --force origin wxmanager

However, you have a pair of commits that do something and then immediately revert it; I would also take this opportunity to remove those. After the reset, dogit rebase --interactive upstream/master, and your editor will popup with a list of commits. Remove theavoid 'TypeError: metaclass ... line and its revert, then exit the editor. Then git will replay your commits and leave out those two. You can follow up with the force-push after that.

@QuLogic thanks a lot. That looks much better now. I think the "--force" was the problem. I could not imagine that such a standard task would require such a flag. A nice GUI with preview etc. would be great...

Concerning--force: Well, you rewrote the history. This is potentially harmful if multiple people work on that branch. You should only ever rewrite the history if you're the only using the branch.--force is your explicitly consent that you know what you are doing.

@QuLogic : as the changes are done, would you mind updating your review result? Or maybe even merge the PR? Once this and PR#11081 are merged, I would like to start work on Wayland support to fix issue#10417

Thanks@DietmarSchwertberger , sorry these sat so long.

@tacaswell : no problem. Thanks for merging. Maybe the guidelines should be updated to encourage self merging when two reviewers have approved.
Anyway, I've been quite bit busy with other projects recently. After 3.0 I will create a PR for issue#2716 and then a PR for an improved wx backend including Windows Metafile support. This should be the basis for the discussion whether to keep or finally remove the wx backend.