pyFDA doesn't shut down properly

[chipmuenk]

When closing the application by closing the window or by clicking the "Quit" button, the python kernel crashes (Windows 7, 64 bit). I don't know whether the same happens under other OS's.

[cfelton]

Similar behavior on OS X and Linux Ubuntu. The python interrupter doesn't crash instantly after a close, it takes awhile (20 seconds). Segmentation fault on Linux and Unexpected failure on OS X.

[chipmuenk]

Under Windows it only takes one or two seconds instead of 20 seconds, but there is a delay. The problem could be due to the hierarchy of widgets:

A simple widget like

%run -m pyfda.input_widgets.input_freq_specs

can be terminated without problem.

A more complex widget like

%run -m pyfda.input_widgets.input_specs

sometimes can be terminated by closing the window (but sometimes the kernel dies), using the "Quit" button never works.

On the other hand, the even more complex widget input_tab_widgets can be closed without problems (so far). Strange.

[chipmuenk]

ahh, that was a bit hasty:

%run -m pyfda.input_widgets.input_tab_widgets

and terminate it using the "Quit" button lets the kernel die as well when attempting to rerun the above line.

[chipmuenk]

Reminder to myself: try
http://enki-editor.org/2014/08/23/Pyqt_mem_mgmt.html
(How to shoot yourself in the foot)
http://stackoverflow.com/questions/14017102/how-to-detect-parent-widget-close-in-pyqt
https://www.commandprompt.com/community/pyqt/x2976.htm
http://stackoverflow.com/questions/3072787/multi-dialog-program-in-pyqt-will-not-close
http://stackoverflow.com/questions/30241684/pyqt4-setparent-vs-deletelater

[chipmuenk]

Problem could be in the _delEntry method of input_freq_specs.py (and other modules).

[chipmuenk]

I had hoped the issue might be related to infinite signal-slot loops - but after cleaning up the signal-slot mess the issue still exists.

Strange:
python -m pyfda.plot_widgets.plot_tab_widgets runs and quits without problem
python -m pyfda.input_widgets.input_filter runs and quits without problem
(same for all other non-hierarchical subwidgets)
python -m pyfda.input_widgets.input_specs runs but crashes when trying to quit it

Ideas, anyone?

[vuki]

I googled for this issue. It seems to be a general problem with the way PyQt is written. You're not doing anything wrong, it's just that the order of calling the destructors is undefined.

It seems that the only workaround is sip.setdestroyonexit(False). I tried this in your code and it works.

See here for info:

http://pyqt.sourceforge.net/Docs/PyQt5/pyqt4_differences.html#object-destruction-on-exit

http://stackoverflow.com/questions/24522327/consequences-of-avoiding-pyqt-program-crash-on-exit-by-using-sip-setdestroyonexi

[chipmuenk]

That's a good point, I've read about the sip module and sip.setdestroyonexit() a few times as well during the last week, but I had and have done some really nasty things in the software like deleting widgets that don't exist any longer, creating instances without relationship to the main window and so on. As a consequence, some other bugs (issue #6 and #7) disappeared with PR #23 as well.

So I guess I'll use sip.setdestroyonexit() as an intermediate fix but try to clean up the plotting routines inheritance mess anyway.

[chipmuenk]

@vuki: Could you tell me at which level you set sip.setdestroyonexit(False) ? I tried it in main() and in the constructor of the pyfdax class, but it had no effect.

[vuki]

Works for me. File pyfdax.py, top of the file:

from sip import setdestroyonexit

Then:

if __name__ == '__main__':
    setdestroyonexit(False)
    main()

Any place before calling app_exec_() should work. I use python 3.5 if that matters.

BTW, when i press Quit button in the left pane, it does not ask me for confirmation.

[chipmuenk]

Confirmed and fixed, unfortunately it only works for Python 3.x.
Under Python 2.x the problem can be reduced to adding plot_tab_widgets with just one (any) plot_widget. When no plot widget is instantiated, the plot_tab_widgets are blank (naturally) but the application shuts down cleanly.

[chipmuenk]

After fiddling around with the dynamic creation and deletion of subwidgets, this seems to be one of the problems: Adding in input_filter.py

337: fb.fil_inst.destruct_UI() # disconnect signals from dyn. widget

an attempt to disconnect signals before deleting corresponding subwidgets crashes the subwidget upon exit (started with python -m pyfda.input_widgets.input_filter

[chipmuenk]

It seems, the problem can be boiled down to Qt objects without parents, creating havoc when the app is closed. Currently, the app can be started and closed without problems when no plot widget is instantiated.

The following parts of the app are / have been especially suspect:

• Dynamic creation and deletion of the amplitude, weight and frequency spec fields depending on the filter type -> now subwidgets are created as needed, superfluous ones are hidden instead of deleted
• Instantiation of all filter widgets at startup to build the filter tree. This instantiated filter widgets at the same time, leaving them dangling in the air. -> Construction of filter widgets is now outside the init method
• Dynamic creation and destruction of filter widgets: Now, the top level widget is referenced and deleted in a proper fashion, signals are connected and disconnected
• The handling of the common plot_utilities and the Matplotlib menu bar in the plot routines is very fishy -> major rework needed ...

[chipmuenk]

Fixed in v0.1rc1! Unexpectedly, the last root cause was a double instantiation of a QSpacerItem under the same name "spacer" ...