Add a 'quit' message that is app-wide, as opposed to mir_event_type_close_surface

[Olivier Tilloy]

Currently in unity8, when requesting to "Quit" a running application from the launcher, what happens is that qtmir goes through all its associated windows/surfaces, and closes them one by one. Supposedly, when the last window is closed, the application terminates itself (but this is not necessarily the case, see https://doc.qt.io/qt-5/qguiapplication.html#quitOnLastWindowClosed-prop).

This prevents application authors from doing proper cleanup before the app is closed. For example in Qt, the QCoreApplication::aboutToQuit() signal is emitted only after all windows have been closed, so the application doesn’t get a chance to e.g. save the list of open windows (real use case for multi-window browser application which saves the current session).

This is similar to unity7’s implementation by the way, but I think it’s wrong. Surely there must be a way (maybe toolkit-specific) to request an app to terminate itself (and fall back to closing all windows if not).

[Olivier Tilloy (osomon)]

Just marked unity7 also affected, as its implementation is similar (it’s Bamf windows instead of Mir surfaces, but essentially it behaves the same).

[Olivier Tilloy (osomon)]

And also marked unity8 affected, because the wording of the action in the context menu is confusing wrt the actual implementation: if we don’t change the actual implementation, we might want to rename "Quit" to "Close all windows".

[Olivier Tilloy (osomon)]

Windows appears to have a WM_QUIT message¹ that’s used to request quitting an application, and Qt handles it correctly in QEventDispatcherWin32::processEvents(). I guess we need something similar for Mir?

¹ https://msdn.microsoft.com/en-us/library/windows/desktop/ms632641(v=vs.85).aspx

[Gerry Boland (gerboland)]

AFAICS Mir has means to ask clients to close windows, but I don't see an API to ask the client itself to quit.

[Alan Griffiths (alan-griffiths)]

AFAIK Mir follows the same model as X11 (which has the WM_DELETE_WINDOW message but no WM_QUIT). I guess you could kill(pid, SIGTERM)?

[Olivier Tilloy (osomon)]

Given that, unlike X11, we have some control over Mir’s design and implementation, maybe this is something that we could consider adding? Or is that out of place?

[Alan Griffiths (alan-griffiths)]

It is not out of place at all - just an enhancement request.

[Launchpad Janitor (janitor)]

Status changed to 'Confirmed' because the bug affects multiple users.

[Gerry Boland (gerboland)]

Does Mir provide enough api to allow application ask user a question before shutdown? In Unity7, typically that is requesting focus and popping up a dialog - should be ok, but good to think about it.

[Daniel van Vugt (vanvugt)]

Clicking on a window's close button sends the app mir_event_type_close_window. That's just an informational message that apps are meant to act on (e.g. pop up a dialog if not terminate themselves).

If you mean logout however, then I guess the right answer is that shells should send mir_event_type_{close_window,quit_app} to all apps on logout, and wait until the apps are no longer running (user responds to dialogs from apps with unsaved work or some timeout expires).

[Daniel d'Andrada (dandrader)]

Maybe a fix for bug 1674749 is what you want.

[Olivier Tilloy (osomon)]

I’m not sure how a fix for bug #1674749 would address the use case in the description: a multi-window application (such as web browser) wants to distinguish between:

 - please close all windows one by one without retaining state for them

 - please quit the app, and save the state for all windows and tabs in them

[Daniel van Vugt (vanvugt)]

Generally speaking fixing bug 1674749 (now bug 1671424) should be enough for most apps in the short term.

I'm reminded of how Microsoft handles this problem...

WM_CLOSE: https://msdn.microsoft.com/en-us/library/windows/desktop/ms632617(v=vs.85).aspx

WM_QUIT: https://msdn.microsoft.com/en-us/library/windows/desktop/ms632641(v=vs.85).aspx
"The WM_QUIT message is not associated with a window and therefore will never be received through a window's window procedure."

WM_POWERBROADCAST: https://msdn.microsoft.com/en-us/library/windows/desktop/aa373247(v=vs.85).aspx

WM_ENDSESSION: https://msdn.microsoft.com/en-us/library/windows/desktop/aa376889(v=vs.85).aspx

Each has a valid use case, and I don't think we can get away from similarly wanting a 'quit' message in Mir eventually.