Bug 423131

SUMMARY

Powerdevil delays registering an Inhibit request from dbus for 5 seconds (commit eca79138c). If the requesting application/service dies during those 5 seconds, the Inhibit is still registered.


STEPS TO REPRODUCE
1. Launch an application that sends a request to Inhibit via org.freedesktop.PowerManagement.Inhibit.Inhibit or directly to Powerdevil via its dbus interface. I used the elisa music player.
2. Have said application send an Inhibit. With elisa, just play some music.
3. Exit the application before 5 seconds after sending an Inhibit. With elisa, that's just exiting the application quickly after pressing play.

OBSERVED RESULT
Using:
watch -n1 dbus-send --session --dest=org.kde.Solid.PowerManagement.PolicyAgent --type=method_call --print-reply /org/kde/Solid/PowerManagement/PolicyAgent org.kde.Solid.PowerManagement.PolicyAgent.ListInhibitions

A new inhibition is listed after the application has exited, 5 seconds after the inhibit was sent.

EXPECTED RESULT
No new inhibition is listed by the application that just exited.


SOFTWARE/OS VERSIONS
Linux/KDE Plasma: Gentoo Linux (unstable)
(available in About System)
KDE Plasma Version: 5.19.1
KDE Frameworks Version: 5.71.0
Qt Version: 5.15.0

ADDITIONAL INFORMATION
I found this while debugging bug #421662 . While arguably elisa should be sending an inhibit on a controlled exit (it can't reasonably do this if it gets a SIGTERM or SIGKILL, or triggers a SIGSEGV, though), there is code in Powerdevil to handle cases where the external service exits without getting rid of the inhibitor. In fact, if I exit elisa after the Inhibit has posted successfully, I do see the Inhibit being removed by Powerdevil.

The root of the problem is that currently, due to commit eca79138c, Powerdevil quasi-registers the pending request, but it doesn't actually fully register it so that the service is being listened to in case it dies. Specifically, in daemon/powerdevilpolicyagent.cpp:

    m_busWatcher.data()->addWatchedService(service);

isn't called until after the 5 second timer goes off.

From what I can tell, this file hasn't changed since 5.19.1 and the current master commit b009a7d856220b9ca64e34cc887069798c801fe9.

I'm experimenting to see what could be a possible solution, but I'm not familiar with the codebase. In addition, due to contract clauses with my employer, I can't provide patches without their permission, at least until I am no longer employed by them. I may be able to provide ideas, though, once I come up with something that works.