Something seems to have changed in Qt 5: you can't get a drop or move event if you don't move at least one pixel from the start point where you were when QDrag::exec() was called. Try putting a breakpoint in the dropEvent of the Draggable Icons Sample, then click a boat and release it without moving the mouse. That generates an "ignore" without any drop signal.
(This is on Kubuntu 13.10 with Qt 5.1.)
When teaching how to start a drag operation, the documentation suggests you might use manhattanDistance() to determine if the mouse has moved enough to really qualify as "the user intending to start a drag". But you don't have to use that; you can start up a QDrag on the click itself.
Anyone know of a workaround to have that same kind of choice on the drop side, or is that choice gone completely? :-/
Why I care: I've long had frustrations trying to get a tight control on mouse behavior in GUI apps—Qt included. There seems to be no trustworthy state transition diagram you can draw of the invariants. It's a house of cards you can disprove very easily with simple tests like:
virtual void enterEvent(QEvent * event) {
Q_ASSERT(!_entered);
_entered = true;
}
virtual void leaveEvent(QEvent * event) {
Q_ASSERT(_entered);
_entered = false;
}
This breaks all kinds of ways, and how it breaks depends on the platform. (For the moment I'll talk about Kubuntu 13.10 with Qt 5.1.) If you press the mouse button and drag out of the widget, you'll receive a leaveEvent when you cross the boundary...and then another leaveEvent when the button is released. If you leave the window and activate another app in a window on screen and then click inside the widget to reactivate the Qt app, you'll get two consecutive enterEvents.
Repeat this pattern for every mouse event, and try and get a solid hold on the invariants...good luck! Nailing these down into a bulletproof app that "knows" it's state and doesn't fall apart (especially in the face of wild clicking and alt-Tabbing) is a bit of a lost cause.
This isn't good if your program does allocations and has heavy processing, and doesn't want to do a lot of sweeping under the rug (e.g. "Oh, I was doing some processing in response to being entered... but I just got entered again without a leave. Hm, I guess that happens! Throw the current calculations away and start again...")
In the past what I've done is to handle all my mouse operations (even simple clicking) with drag & drop. Getting the OS drag & drop facility involved in the operation tended to produce a more robust experience. I can only presume this is because the testers actually had to consider things like task switching with alt-Tab, etc. and not cause multiple drop operations or just forget that an operation had been started.
But the "baked in at a level deeper than the framework" aspect actually makes this one-pixel-move requirement impossible to change. I tried to hack around it by setting a timer event, then faking a QMouseEvent to bump the cursor to a new position once the drag was in effect. However, I surmise that the drag and drop is hooked in at the platform level, and doesn't consult the ordinary Qt event queue: src/plugins/platforms/xcb/qxcbdrag.cpp
