Comment 124 for bug 1446865

So, looking into the Qt sources a bit more it seems that

a) Session management is FUBAR in Qt5 and actually was likewise FU in Qt4 - adding the proposed flag will only restore the completely FU condition of Qt4, ie. allow clients (KDE) to workaround the brokeness ...
b) this actually *only* seems to affect the xcb platform ("GuiApplicationPrivate::commitData()" is only called by "QPlatformSessionManager::appCommitData()" which is only called by "sm_performSaveYourself(QXcbSessionManager *sm)" in QXcbSessionManager)

Restoring the Qt4 state however won't cut it: KApplication is deprecated and there's no Guarantee for a KMainWindow around. Also every "plain" Qt5 client that can be found on KDE users desktops (qupzilla, trojita, ... whatever) won't be covered.

If you "seem to be the expert on session management" we need to define a roadmap out of this mess, or it's never gonna happen.

What do you think about this process:

1. emit commitDataRequest()
2. at this point either the logout should be canceled (requires API addition on QSessionManager to wire up sm_cancel; internal requirement only though. Maybe kept in d_ptr contexts) or there thould be no windows with QWidget::isWindowModified() or modal transients around since that indicates some user interaction should have taken place but didn't
3. If we detect missing interaction, spam a QMessageBox and inform/ask the user:
   "It looks like the application should have intercepted the logout request, but didn't. Do you wish to cancel the logout and explicitly save data? Since this is an application bug and this failsafe check will be removed with Qt6, please file a bug against the application."
4. If this still didn't cancel the logout, start closing windows depending on an application attribute (see below) and a trumping environment variable QT_CLOSE_WINDOWS_ON_LOGOUT_REQUEST (to pass the user ultimate control over the behavior)
5. Remove attribute, failsafe check and (actually even documented, we can NO WAY just remove this in Qt5) window closing in Qt6

I'd go for the application attribute since indeed there's little point in adding functions to implement a short-term workaround.
Actually, I might even be in favor of a dynamic property (QObject::setProperty()) since it will spare the clients compile time checks - we an add this to KFooBar now, even though Qt 5.6 (or whatever) is not on the horizon.