When deleting the panel the notifications applet is in, the containment is removed but the corona is still there and rightfully emits availableScreenRectChanged.
This will cause us to crash when we try to access the no-longer existing containment.
BUG: 378508
FIXED-IN: 5.8.7
Even though we no longer emit the signal when tearing down, we still do (correctly) when manually removing the panel. The containment goes away nonetheless.
No longer crashes when I remove the panel. Resizing the panel still repositions notification popups.
| Automatic diff as part of commit; lint not applicable. |
| Automatic diff as part of commit; unit tests not applicable. |
Your reasoning behind the bug is missing some steps.
QObject destruction deletes children *then* emits QObject::destroyed
Applets are children of the containment. (see Containment::addApplet())
So if the notification applet is in the panel, it's fine (and won't crash)
What you actually have is: panel containment -> system tray -> system tray containment -> notification applet
System tray must be not setting applets as children of the system tray containment? Or not deleting it's containment when it deletes the applet?
i added the applet to a panel directly and it would crash without this patch, with is fine. Frankly i didn't test the in tray case.
Heh, finally found the bug.
I was a bit wrong.
QObject::destroyed() is emitted before we delete the children.
But that still doesn't explain everything, the containment won't get deleted till after the child applet is deleted; so why would containment() be null?
We know it's not deleted yet, so it's not really "gone" like your comment says,
The issue is that containment() is a qobject_cast recursive search of parents. (for some reason)
qobject_cast calls metaobject() which is virtual. Calling a virtual method during an objects destructor can result in unexpected behaviour.
Our parent is still valid, but we've called Containment::~Containment, Applet::~Applet and we're currently in QObject::~QObject, so when we call a method on a base class it doesn't get dynamically dispatched as it should.
So when we call parent()->metaObject() we now get the metaobject for QObject not the metaobject of Containment.
This means the qobject_cast fails, and that's why Applet::containment() is null even though the containment is still there.
| applets/notifications/lib/notificationsapplet.cpp | ||
|---|---|---|
| 54 | I think, corona (witch is parent of containment) is still alive and that's why slot is invoked. | |
| applets/notifications/lib/notificationsapplet.cpp | ||
|---|---|---|
| 54 | Obviously, the corona manages all the containments, we only delete a single containment, though. | |
RE: your message on IRC this morning.
Yes, you should submit this.
I explained in my comment why it's needed. It's not quite for the reason you initially said, but the solution is still valid.
Ideally fix your comment about the containment being gone.