Unable to change filename when saving from figure window

[adamjstewart]

Bug report

Bug summary

When I try to save a figure by clicking the save button in the figure window, it doesn't allow me to change the filename. Even if I explicitly click the Finder-like window that pops up, all text is sent to the Terminal.

Code for reproduction

import matplotlib.pyplot as plt

plt.plot()
plt.show()

Actual outcome

I am unable to change the filename, and all text is sent to the Terminal. This may be related to the fact that I'm seeing the following error message in the Terminal:

objc[36742]: Class FIFinderSyncExtensionHost is implemented in both /System/Library/PrivateFrameworks/FinderKit.framework/Versions/A/FinderKit (0x7fff8965cc90) and /System/Library/PrivateFrameworks/FileProvider.framework/OverrideBundles/FinderSyncCollaborationFileProviderOverride.bundle/Contents/MacOS/FinderSyncCollaborationFileProviderOverride (0x112aa1cd8). One of the two will be used. Which one is undefined.

Perhaps the error message is forcing the focus to be on the Terminal? It would be interesting to see if anyone can reproduce this without any error message.

Expected outcome

I should be able to change the filename.

Matplotlib version

• Operating system: macOS 10.13.6 (High Sierra)
• Matplotlib version: 2.2.3 (conda)
• Matplotlib backend: MacOSX
• Python version: 3.6.5 (conda)

[adamjstewart]

Update: I switched to the Qt5Agg backend and this bug no longer occurs. So this is definitely a bug in the MacOSX backend.

Side note: I noticed that the Qt5Agg backend is significantly more powerful than the MacOSX backend. The MacOSX backend figure window does not have the "Edit axis, curve and image parameters" button that the Qt5Agg backend has. Is there some fundamental limitation of the MacOSX backend that prevents this, or does it just need a developer to work on? Also, where can I find a side-by-side comparison of each backend in terms of functionality and speed?

[WeatherGod]

I swear, I remember this bug from a long time ago. Could this be a framework/non-framework build issue again?

…

On Mon, Sep 3, 2018 at 5:58 PM Adam J. Stewart ***@***.***> wrote: Update: I switched to the Qt5Agg backend and this bug no longer occurs. So this is definitely a bug in the MacOSX backend. Side note: I noticed that the Qt5Agg backend is significantly more powerful than the MacOSX backend. The MacOSX backend figure window does not have the "Edit axis, curve and image parameters" button that the Qt5Agg backend has. Is there some fundamental limitation of the MacOSX backend that prevents this, or does it just need a developer to work on? Also, where can I find a side-by-side comparison of each backend in terms of functionality and speed? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#11879 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AARy-EA_2MTJSqlZzlqrkCqYLxj5ysmVks5uXaYCgaJpZM4WClpX> .

[jklymak]

Yes that is exactly what this is. @adamjstewart if you do pip install python.app and then run pythonw instead of python this should work. OTOH I thought the macOS backend was not supposed to work at all if you didn’t do this.

[adamjstewart]

pythonw did the trick! The default backend for conda-installed matplotlib is MacOSX. I wonder if this could be changed.

[anntzer]

This should (essentially) be fixed in 3.0 with the backend switching code as long as 1) conda doesn't force the default backend to OSX anymore and/or 2) conda's python (not pythonw) doesn't pretend it is a framework build (from the PoV of our test). @jklymak given that you know this part of the code well now :-) can you check that?

[tacaswell]

I am not sure we can fix this if it is related to if the GUI event loop is spinning correctly or not which depends on the details of the python build...

[jklymak]

@anntzer - sorry to be dense - what do you want tested?

[anntzer]

Looks like the original bug arises from running the OSX backend on a non-framework build(?), which should not happen anymore after the backend-switching+framework-detection work(?)
Can you confirm my interpretation?

[dstansby]

I can't get the OSX backend to get past the expected error message now; closing this, but anyone feel free to re-open if I've missed something.