Qt slow in Spyder when using ipykernel 5.3.3 (5.3.2 OK)

[8bit-Dude]

I am using Spyder 4.1.4 on Miniconda3-py37-4.8.3.
I noticed that matplotlib was unsually slow in IPython when using the Qt backend.
Downgrading ipykernel with conda to version 5.3.0 can however fix the issue.

[blink1073]

@ccordoba12 might this be related to #525?

[DiegoF90]

I also have the same issue, matplotlib seems to be very slow with the new Spyder 4.1.4.
I am not sure if it is related to #525, but I had already done the change suggested by @ccordoba12 here and faced this issue later.

[DiegoF90]

I do not know if this is related or f any use but, I was trying to use Tkinter as the graphics backend to see if the speed improved, but after restarting Spyder to apply this changes I got:

ERROR:tornado.application:Exception in callback functools.partial(<function Kernel.enter_eventloop.<locals>.advance_eventloop at 0x000001D8E9EA4AF0>)
Traceback (most recent call last):
  File "C:\Users\1520797\AppData\Local\conda\conda\envs\gmv\lib\site-packages\tornado\ioloop.py", line 743, in _run_callback
    ret = callback()
  File "C:\Users\1520797\AppData\Local\conda\conda\envs\gmv\lib\site-packages\ipykernel\kernelbase.py", line 314, in advance_eventloop
    eventloop(self)
  File "C:\Users\1520797\AppData\Local\conda\conda\envs\gmv\lib\site-packages\ipykernel\eventloops.py", line 238, in loop_tk
    app.tk.createfilehandler(stream.getsockopt(zmq.FD), READABLE, notifier)
AttributeError: '_tkinter.tkapp' object has no attribute 'createfilehandler'

Which is the file referenced from which the problem seems to come from in #525 ?

[DiegoF90]

I realized that I was already using 5.3.0 but with the changes suggested in here (which I am guessing were included in 5.3.3) and matplotlib was still very slow. I then rolled back the changes and matplotlib started working as normal again. Maybe the changes have to be further reviewed so that they fix this but do not introduce this slow behaviour for matplotlib?

[ccordoba12]

@DiegoF90, thanks for confirmation. Could you let us know in what way Matplotlib is slow? I did some simple testing for PR #525 and things were working fine for me.

[Nowa-Ammerlaan]

I have the same problem with 5.3.3, plot windows are slow, the save button is completely unresponsive. There are no issues when running the same script through any terminal directly with python. (See also the issue that @ccordoba12 mentioned above)

Downgrading to version 5.3.2 fixed these problems

[ccordoba12]

@blink1073, one solution to this would be to exclude the control channel completely from the Qt event loop for now. That will solve the issue fixed in #525 and also this one.

However, I don't know what implications that could have for the JupyterLab debugger. Pinging @SylvainCorlay about it.

[DiegoF90]

Hi @ccordoba12 thanks for looking into this. As @AndrewAmmerlaan mentions, the Figure window after plotting even simple graphs becomes very unresponsive. It takes longer than usual just to switch to this window. Other processes that are usually done in almost real time take second: e.g. if you activate the grid ax.grid(), the Figure window shows at the bottom left the x and y coordinates at the same time as you move around the mouse. Showing the coordinates after 5.3.3. takes around to second to update for each position.

[8bit-Dude]

I also confirmed that the issue does not occur in version 5.3.2.

[raphaelquast]

... just to add myself to the list here...
i can also confirm that matplotlib is very very slow after updating from 5.3.2 to 5.3.3

[ccordoba12]

Hey everyone, could you help us to test pull request #531 to see if it fixes this problem and also the warnings problem that pull request #525 tried to fix?

We'd really appreciate your help in order to not release another faulty version. Thanks!