On a quick scheme, it looks like all of the changes are simple re-naming of functions. The logic to pick which one should be centralized in one place, probably at the top of backend_wx.py.

 if is_phoenix:
    EventLoop = wx.GUIEventLoop
else:
    EventLoop = wx.EventLoop

And then

self._event_loop = EventLoop()

We should also do enough compatibility munging that the user-code does not care which version of wx is being used.

Sorry, don't understand the last comment.
I don't find this line "self._event_loop = EventLoop()" in the code and don't see what would be gained in moving GUIEventLoop/EventLoop definition to a centralized place they are only used in the start_event_loop method.

That was just the first example I hit when scrolling down the code.

The point is to centralize all of the version logic in one place, instead of scattered through out the code where it is used (see what we do with qt_compat which smooths over the differences between pyqt4, pysides and pyqt5). It will make the code more readable and easier to fully deprecate the old api in the future.

I had a quick look at qt_compat and I am not sure that I am the right person to make such a change.

Why? You have already done the hard part. I am just asking that centralize it to improve readability and maintainability.

O.K., I actually started on it and I am making progress.

On the backend_wx and backend_wxagg files I get the following warning when I commit, but not on e.g. the sample files, is this a problem on my side or?:
warning: LF will be replaced by CRLF in lib/matplotlib/backends/backend_wx.py.
The file will have its original line endings in your working directory.

Did the wx_compat.py changes in a new branch locally. What do you prefer that I update this PR or close it and create a new one?

I wouldn't worry too much about line endings at this point. It is probably a mixture of windows vs. unix line endings in the codebase. We can always feed the files through a converter when we are done.

As for the wx_compat.py changes, just keep making them in this same branch and push them up to your repo. GitHub will track the branch and update this PR automatically. "PRs are the beginning of the conversation!"

I don't understand the failure report on Travis. The failure seems to be in the test and not the backend_wxagg.py file.

You need to remove the file from the black list. It is complaining a file that should fail is passing.

I only use the wxagg backend, so just noticed while testing that the wx backend is broken when I use wxPython 2.9+, it seems to work with 2.8.12.
I don't see any errors, I just see an empty canvas.
Will keep searching, but if anyone has an idea what might cause this, please let me know.

The wx backend works for me with wxPython 3.0 and (wxWidgets 3.01) i.e. doing:

%matplotlib wx 
import matplotlib.pyplot as plt
plt.plot(range(100))

In IPython works as expected.

I use embedding_in_wx2 for testing and activate these lines (and commend the WXAgg lines):

matplotlib.use('WX')
from matplotlib.backends.backend_wx import FigureCanvasWx as FigureCanvas

If this works for you too then I must have messed something up.

embedding_in_wx2 works for me with both the WX and WXAgg backend but the toolbar is missing unless I remove the OSX workaround. (Im on a mac with OSX 10.9), Perhaps that is no longer relevant. It mentions OS X 10.3

I am on Windows (8.1), so it is a specific to that platform. BTW, just tested with MPL 1.3.1 and I see the same issues. Will concentrate on more wxAgg testing and later come back to this issue.

I suspect some issues with sizer handling as that is something which has changed in 2.9+.

I can still not find a solution for the wxPython 2.9+ issue (no graph shown) when using the 'WX' backend. This only seems to happen on Windows as Jens doesn't see it on OSX.

In 2.9.0 the DC code was reorganized in wxWidgets - http://wxpython.org/Phoenix/docs/html/DC.html, but searching the mail lists and googling didn't help me finding a solution.

Hopefully someone with better wxPython know how will find a solution.

I found the reason for the 'wx' backend not showing the graph on windows. The MemoryDC is held due to the _cache in GraphicsContextWx a hack is to do this in gui_repaint:

dc, gc = self.renderer.gc._cache.pop(self.bitmap)
dc.SelectObject(wx.NullBitmap)
del dc

before calling DrawBitmap.

Why are we caching the GC, in my tests there is always only one Bitmap and therefore only one entry in the cache.

With these latest changes the 'WX' and 'WXAgg' backends work for me on Windows with wxPython 2.8.12.1, 2.9, 3.0.1.1 and 3.0.2 Phoenix with Python 2.7.8
To get the 'WX' one to work on 2.9+ I had to use a workaround in FigureCanvasWX.gui_repaint as 'self.bitmap' was held by 'GraphicsContextWx._cache and therefore the DrawBitmap call failed silently - I have a question open on wxPython if this can be fixed in a future release of 3.0.x.

Would be great if this could be tested on Mac and Linux. Hope to test it soon on Windows with Py3.4

Great to hear!

Did an initial test with embedding_in_wx2 using Python 3.4 with Phoenix build 3.0.2.dev77889 both on Windows and Linux Mint 17 and they worked for both 'WX' and 'WXAgg' backends.
Only hick I am seeing is with the toolbar button 'subplots' after one used that the application is not closing correctly, i.e. there is a crash without exception but it must be after wx.App.OnExit is run. So, looks like some Phoenix issue not certainly related to the mpl use of it.

Do I need to do anything more on this?

Any change of getting it into 1.4.3?

At a mimimum his needs a re-base

The failure in 3.4 is:
matplotlib.tests.test_mathtext.test_mathfont_stix_07.test ... FAIL

Can someone help me figure this one out?

@wernerfb That's most likely a random test fail. I have restarted the travis instance.

You can ammend a commit message this way, if you wish:
http://stackoverflow.com/questions/179123/edit-an-incorrect-commit-message-in-git

On Thu, Apr 23, 2015 at 4:29 AM, Jens Hedegaard Nielsen <
[email protected]> wrote:

@wernerfb https://github.com/wernerfb That's most likely a random test
fail. I have restarted the travis instance.

—
Reply to this email directly or view it on GitHub
#3421 (comment)
.

great job squashing! One thing I just noticed. Could you please put back the import six lines, and add import six to wx_compat.py? We have a standing rule that all module files in matplotlib will have that line and the from __future__ line, regardless if it is actually used or not in the file. This makes it easier on ourselves to be able to assume that the appropriate tools are available everywhere.

Great! @tacaswell this probably needs a whats_new entry in the documentation, right? After that, I think we are all good.

Yes. I think you can look at the entry for qt5 from the 1.4 release cycle for a template.

@wernerfb Thank you for bearing with us through what has ended up being a very long process!

@tacaswell no problem. Re entry for what's new, you want me to do that? If yes, is the correct file 'whats_new.rst'?

Please create a new file in
https://github.com/matplotlib/matplotlib/tree/master/doc/users/whats_new,
the readme has directions.

On Fri, Apr 24, 2015 at 12:41 PM Werner F Bruhin [email protected]
wrote:

@tacaswell https://github.com/tacaswell no problem. Re entry for what's
new, you want me to do that? If yes, is the correct file 'whats_new.rst'?

—
Reply to this email directly or view it on GitHub
#3421 (comment)
.

ping @wernerfb ? Looks like we only have one little typo, "to set" --> "for setting", and a little bit more clarification in the whats_new doc. If you have run out of time for this, let me know and I will see what I can do to wrap this up.

@WeatherGod will get to it in the coming week, was away on vacation.

@wernerfb Great job! Thank you for your patience. I think the extra attention to detail here will really pay off! Cheers!

@tacaswell could you please maybe have a look on my issue too? this seems to be more than my knowledge (as beginner)
:CellProfiler/CellProfiler#3636
and: https://forum.image.sc/t/installation-of-cellprofiler-problems-with-ilastik-and-wx-modules/19534
thanks a lot