Can you try running the scripts from a terminal? I experience similar problems and it seems to work better from a terminal. Still a problem of course, but would be interesting to narrow it down. It may still be a Matplotlib issue, but also a Spyder/spyder-kernel/IPython/ipykernel problem...

Hi. thanks for responding so fast.
It works well from a terminal running in the same conda-environment as I use from Spyder.
So it might be related to Spyder/iPython kernel issue...

Here are some related issues:
ipython/ipykernel#775
spyder-ide/spyder#17172

Not explicitly like this, but at least indicating that there are some problems with the related code and that it is not really clear who should change what... But for sure worthwhile digging into!

It may be that this is fixed by #22202

Is there some kind of backtrace, or does it simply die?

It simply just dies, when I try to open the menu.

hey, I thought I'll add my problem here since it seems to be quite related... (and I'm using matplotlib v3.5.2)
On my side, the Qt5Agg gui becomes irresponsive if I click on the button to adjust the axis limits while there is more than 1 axes in the plot... (plt.subplots(2) or ax.twinx() show the same behaviour)

... and it also simply dies (I can't even close or resize the gui any more)

[edit by @tacaswell to put gif in a detalis fold because it was making me slightly motion sick!]

I experience the exact same thing.
The funny thing is that if you manage to hit ok in the pop-up window, before it crashes, it actually stays functional.
So it might be related to the initial window, where you can select which plot you want to configure.

This looks like there is something wrong with the input hook inside of the terminal that is being hosted by spyder. When you see things like "I click on the window and nothing happens" or "I resize and just get blank space padding the graph" that is usually because (for what ever reason) the UI event loop is not running. Hence the windowing system is sending to your application ("user clicked", "user clicked",...) and your application is just being like "yup, good to know, put it on the queue" but not actually doing anything about the input.

If in the terminal you run plt.show(block=True) do thing come back to life / work ?

hey, thanks for the response!

I checked and using plt.show(block=True) or a plain plt.show() does not help.
... also I get the very same behavior in a basic python console... so spyder does not seem to have anything to do with it.

However, I noticed that this is only an issue if interactive mode is on (e.g. plt.ion() is called).
In non-interactive mode, all works as expected.

The behavior of plt.show() depends on the value of plt.isinteractive() (which is stored in an rcparam). For debugging please explicitly set block to True or False just to be sure we are decoupling from any underlying global state.

Can you also test in an IPython shell (as oppossed to plain python) and make sure you are using prompttoolkit (rather then readline) for the prompt?

This is starting to get into the realm of https://matplotlib.org/stable/users/explain/interactive_guide.html and how the event loop is (or is not) managed.

Hey, I can confirm that if I directly use an IPython 7.31.1 shell all works as expected !
(from what I read, prompttoolkit is used from v5.0 on)
The strange thing is that Spyder is using the same ipyhton version but here the error is happening just as with the plain python shell.

Hey, I can confirm that if I directly use an IPython 7.31.1 shell all works as expected !

I think this show that this is fundamentally an issue with spyder, readline, pyqt/pyside and the interaction with the event loop. It is possible that there is something we could do differently on the Matplotlib side to work around this, but I think we would need guidance from down-stream about what those changes will be.

I would point to spyder-ide/spyder#4367 which re-points the problem to the IPython Qt console. I strongly suspect that this is another version of this.

I'm going to close this issue here as

1. the issue seems isolated to usage only with in the qt ipython console
2. the integration of the event loop that hosts the Matplotlib figures (any toolkit and and downstream use) is the primary responsibility of down stream tools. That said we will definitely take any patches that make that integration work better, but failure to work with a given integration is not a bug.

Sorry to play hot-potato 😞 .

attn @cordoba12

hey, ok sounds reasonable...

just as a final comment...
i think there might be 2 different things in question here

1. the fact that spyders ipython console has the issue but a "plain ipython console" does not
2. the fact that the same error occurs in a plain python console

While the first one is certainly on spyder (or somewhere further down the road), the second part remains puzzling ...
(e.g. the issue is not isolated to a qt ipython console)

the fact that the same error occurs in a plain python console

fair, re-opening.

I'll re-close since the original issue was Spyder only. @raphaelquast can you open a new reproducible issue with the problem that you can reproduce without Spyder? Thanks for your understanding.

@jklymak @tacaswell moved to new issue #23103