How did you track down that comment?

backported to v2.x as 6ab5475

git blame to see which PR that line came from...

Hi,
I went into this issue after installing Qt 5.7 and I was unable to work around it. I cloned the current master branch and installed it, but I still get the error. Being quite basic with git, I'm not able to understand which is the release version this master branch relates to: is it 1.5.2 (for which the issue seems to be unsolved) or 1.5.3 (for which it should be) ?
Thanks for any hints,
Massimo

The 1.5 line is nearing its end. Apart from one last bugfix release, which will be 1.5.3, the line of development goes to 2.x, for which a 2.0 release is anticipated quite soon, and then master, which will be the origin of an eventual 2.1 release--but not soon. The fix in this PR is in master and in 2.x, so if you installed master you should have it. It sounds like it should be backported to 1.5.3. The immediate question, though, is why you are still getting the error if you have installed master. Is it possible that you are seeing a similar error, but not exactly the same one? Or that the python you are using is not the one into which mpl master was installed?

Hi,
I solved it. I messed up the environment so that python3 was still using the pip3 version I previously installed, which is at 1.5.2 version. I uninstalled it so that now the interpreter correctly sees the 2.0-based version ( 2.0.0b3+2000.g22aa800 ) and my loved tool matplotlib is again working with my preferred backend, qt5.
Thank you,
Massimo

Actually, I now think that this issue is annoying enough to warrant a bugfix release asap (say 1.5.2.1 and possibly 2.0b4, respectively based on 1.5.2 and 2.0b3, with only this additional patch in), except if the next releases are planned for very soon.

there is a (sadly) growing list of things to backport to a 1.5.3

I'm seeing this problem with matplotlib 1.5.2, which is the current version that installs with pip install matplotlib for python 2.7.

backported to 1.5.x as 3d2872e

I just synced my fork with master, and now I have this problem as well, after a bunch of gtk warnings. Apparently it should not be happening anymore. The version shows matplotlib==2.0.0b1+1895.g391198e, and the output file.

In the notebook it works fine though.

@rmlopes 391198e is the tip of your PR (which has stalled, sorry about that!), which was forked before this fix was merged.

Thanks, how can I fix it? rebase again?

@rmlopes Yes, rebase + force push should fix this.

Could it be that I'm seeing the same error under Fedora 25?

python-qt5.x86_64                            5.7-4.fc25
python2-matplotlib-qt5.x86_64                1.5.2-0.2.rc2.fc25

Yes, you have a version of Matplotlib that pre-dates this fix (went out the door in 1.5.3).

Ah, so not backported to 1.5.2.x then I suppose?

There is no 1.5.2.x Matplotlib has at most 3 release levels. The official releases of the 1.5.x series were: 1.5.0, 1.5.1, 1.5.2 and 1.5.3. Anything else is done by fedora maintainers. They may chose to backport patches to their packages but that is completely up to them.