tk window immediately resizes down to zero-height upon showing up.

[anntzer]

Bug report

Bug summary

tk window (whether tkagg or mplcairo.tk) immediately resizes down to zero-height upon showing up.

Code for reproduction

$ MPLBACKEND=tkagg python -c "from pylab import *; plot(); show()"

Actual outcome

As described above.

Expected outcome

Window stays at original size.

Matplotlib version

• Operating system: arch linux
• Matplotlib version: master
• Matplotlib backend (print(matplotlib.get_backend())): tkany
• Python version: 38
• Jupyter version (if applicable):
• Other libraries:

This bisects to #16828 (attn @QuLogic), even though it's not clear how that's related... perhaps the dynamic changes to figure dpi confuse tk, but it's not clear why this is different from before?

[QuLogic]

I'm guessing this is related to the flaky test on AppVeyor. The only thing that's hit by that test is the fix to Text.get_window_extent: https://github.com/matplotlib/matplotlib/pull/16828/files#diff-15e8fbe2a4d40deb738e8627eba02318L894

My best guess is that figure.dpi has never been correctly set, and it's been relying on this accidental change to do so.

[QuLogic]

So it seems that's not it, as I can never trigger a case where dpi is not None, but self.figure.dpi is None (or something other than dpi). The strange thing is that if I print out all the properties that would be changed in Figure._set_dpi, nothing changes. So neither the forward call to Figure.set_size_inches (which might hit the canvas/toolbar size), nor the callbacks for the dpi_changed event do this. Yet somehow on a later call, the size is smaller...

There's definitely something else weird going on; if you add an extra dpi changed on the old code:

MPLBACKEND=tkagg python -c "from pylab import *; plot(); gcf().dpi=100; show()"

then there will be one call to Figure.set_size_inches with [6.4 4.46], but it magically fixes itself. I'm guessing what happens is that since the dpi change in the Text.get_window_extent is part of the draw/stale mechanism, it is enough to propagate that height change.

[FBIShowYourHand]

a question about how to use Chinese,when I want to show Chinese,it always shows me errors,thanks

[tacaswell]

was #16828 backported to the v3.2.x branch?

[tacaswell]

Ah, I understand now, not a regression on 3.2.x, but was backported there because it fixed some other bugs.

[QuLogic]

You're right though; this bug was never in 3.2, so this issue shouldn't be tagged on it.