How to track down a bug w/ PyGTK3 which freezes the UI?

[Emmanuel Blot <eblot ml gmail com>]

Hi,

I'm using PyGTK3 [python3.7, Debian 9.7 @ x86_64, Gtk 3.22.11] for an 
embedded UI [X11, Xfce4, kiosk mode].

I noticed that after a while, the UI "freezes": the UI items are not 
repainted anymore, whatever Gtk calls are invoked.

I initially thought the main loop was stuck for some reason, but it is 
definitely not: execution goes on, application threads and UI thread 
keep logging traces. I've also added a timer (GLib.timeout_add) which 
keeps being called at the defined pace.

I double-checked that all calls made from the application threads to Gtk 
loop directly go through GLib.idle_add; all events received from Gtk are 
pushed into an event queue that the application thread manages, to avoid 
any long calls that could be made from the main context.

Any hint on how to debug this kind of issue would be appreciated, as I'm 
out of ideas. GDB is available on the test platform, although I'm not 
sure how to track this issue from there.

Additional details about this problem:

UI items include several text labels, a progress bar, and two basic 
drawing areas, all contained in a page of a notebook. The app is running 
in full screen mode [800x480].

I run the app with some kind of auto-test mode, to run the same sequence 
again and again until it freezes, without any user input - nor any input 
event.

When it freezes, the application does not crash, the CPUs show no 
noticeable load when the UI freezes.

The UI freezes "randomly", from 5 to 100 cycles. Each cycle last ~ 4 
seconds. A cycle makes less than a hundred calls to Gtk API.

The UI provides a modal dialog to quit the app, which can be kicked in 
from the main menu bar. When the UI is frozen, e.g. many calls to change 
the content of the text label, the progress bar fraction value, etc. 
have not been serviced, the dialog nevertheless shows up. Sometimes, its 
contents is painted successfully, sometimes only the dialog frame (inc. 
buttons) is displayed along with its title text, while the dialog 
content area is not (re)painted...

Using the GTK+ inspector, I can see that the text labels, the progress 
bar fraction, etc. are actually updated whereas the UI does not refresh 
their contents. However, the GdkFrameClockIdle counters *all* stop as 
the same time the UI is not updated anymore.

What could cause this kind of trouble, where the UI is not refreshed 
anymore, but main context seems to be up and running?

Thanks,
Manu.