I'm not sure this is the correct fix either. It is definitely something we should fix though.

Is this a regression from 1.1.1? Are you able to git bisect to identify the commit that caused the problem?

This sort of thing often has interesting interactions with IPython, so we should make sure to test it both there and in the regular Python console.

I think this behavior was always present. It is not a regression. It is just that the loading of the icon from disk in mpl 1.2 takes long enough such that the root window becomes visible. I can reproduce this with mpl 1.1.1 if I add a small delay right after the creation of the Tk window.

It is just that the loading of the icon from disk in mpl 1.2 takes long enough such that the root window becomes visible.

Ah, ok. That could be a "regression" (introduced by me presumably). On this basis, I probably think your proposed change is the right thing to do (see http://stackoverflow.com/questions/1406145/how-do-i-get-rid-of-python-tkinter-root-window).

@fperez it would be great to test this simple fix with IPython, are you able to try it out for us? (the only thing to look out for is that the standard TkAgg behaviour still functions as it always has)

This seems to have no adverse effects on Linux with IPython git master or IPython 0.13. Of course, the real test needs to happen on Windows where the initial bug that this solves happens in the first place.

I tried it on Windows with IPython 0.13 (plain, pylab, and qtconsole) and did not notice anything wrong.

Lets go for it then. @mdboom : I plan on cherry picking this into 1.2 (or @cgohlke can rebase to the 1.2 branch) - do you agree?

Forgive me if I'm wrong, but is the 'root' window here one of the 'extra' windows that was popping up in #1212?

Yes it is, and your "withdraw" call in this PR is the right way to handle it--but this is independent of the fundamental problem with PR #1125 (reverted by PR #1221).

Please cherry pick this into 1.2.

This has been merged and cherry-picked to v1.2.x