Tests with tests ubuntu-latest, python 3.11 and pyside6 always fails

[malik-irain]

This combination always fails in github actions.

[seb5g]

I'm sure you'll find why ;-)

[malik-irain]

It actually doesn't always fail as I couldn't reproduce locally and on some runs in github actions the tests pass. So, it "randomly" (= multithread) segfaults while garbage collecting:

Fatal Python error: Segmentation fault

Thread 0x00007f94d21ff640 (most recent call first):
  File "/opt/hostedtoolcache/Python/3.11.10/x64/lib/python3.11/site-packages/execnet/gateway_base.py", line 534 in read
  File "/opt/hostedtoolcache/Python/3.11.10/x64/lib/python3.11/site-packages/execnet/gateway_base.py", line 567 in from_io
  File "/opt/hostedtoolcache/Python/3.11.10/x64/lib/python3.11/site-packages/execnet/gateway_base.py", line 1160 in _thread_receiver
  File "/opt/hostedtoolcache/Python/3.11.10/x64/lib/python3.11/site-packages/execnet/gateway_base.py", line 341 in run
  File "/opt/hostedtoolcache/Python/3.11.10/x64/lib/python3.11/site-packages/execnet/gateway_base.py", line 411 in _perform_spawn

Current thread 0x00007f94d3892b80 (most recent call first):
  Garbage-collecting
  File "/opt/hostedtoolcache/Python/3.11.10/x64/lib/python3.11/site-packages/pyqtgraph/graphicsItems/GraphicsObject.py", line 20 in itemChange
  File "/opt/hostedtoolcache/Python/3.11.10/x64/lib/python3.11/enum.py", line 1257 in value
  File "/opt/hostedtoolcache/Python/3.11.10/x64/lib/python3.11/enum.py", line 211 in __get__
  File "/opt/hostedtoolcache/Python/3.11.10/x64/lib/python3.11/site-packages/pyqtgraph/graphicsItems/GraphicsObject.py", line 20 in itemChange
  File "/opt/hostedtoolcache/Python/3.11.10/x64/lib/python3.11/site-packages/pyqtgraph/graphicsItems/GraphicsObject.py", line 15 in __init__
  File "/opt/hostedtoolcache/Python/3.11.10/x64/lib/python3.11/site-packages/pyqtgraph/graphicsItems/UIGraphicsItem.py", line 28 in __init__
  File "/opt/hostedtoolcache/Python/3.11.10/x64/lib/python3.11/site-packages/pyqtgraph/graphicsItems/ROI.py", line 1359 in __init__
  File "/opt/hostedtoolcache/Python/3.11.10/x64/lib/python3.11/site-packages/pyqtgraph/graphicsItems/ROI.py", line 618 in addHandle
  File "/opt/hostedtoolcache/Python/3.11.10/x64/lib/python3.11/site-packages/pyqtgraph/graphicsItems/ROI.py", line 559 in addRotateHandle
  File "/home/runner/work/pymodaq_gui/pymodaq_gui/src/pymodaq_gui/managers/roi_manager.py", line 304 in __init__
  File "/home/runner/work/pymodaq_gui/pymodaq_gui/tests/plotting_test/utils/plot_utils_test.py", line 133 in test_get_repr

From what I saw in the code, nothing special is happening at the crash point.

One way to solve it could be to avoid cyclic references, as stated on an old pyqtgraph issue, but more likely there are more general thread-safety problems.

[seb5g]

in the issue you mentioned, it's more likely related to C++ objects being removed while an instance of the python object is still around. Maybe explicitly deleting things like windows and main widgets could help...

[malik-irain]

Yes, but it shouldn't happen, so my guess is that it's more probable that we do something wrong (circular references + multithreading / something else ...) than the libraries do.

I'm not sure about explicit delete, it could mitigate the problem but not really solve it. Also, in practice the del keyword only remove the reference and not the object itself, leaving the object ready to be garbage collected later on when it's reference count reach 0 (which is not an info that we have). Then, to delete the objects we would have to call gc.collect() regularly, which is considered a bad practice.