Hi - this should be fixed in https://gitorious.org/mypaint/mypaint/commit/df3afd837ebaa9794f9b0eaaa32bd7b0ae8efcf7 , and I've not been able to reproduce this problem since (I have to use very short fast drags to trigger this race condition). Conversely, sticking debug prints in leave()'s new repeated calls to self._finalize_move_idler() now shows occasional calls there at precisely the point you'd expect an exception with the old code and the right drag-fu. Good enough for me!

Also noticed exceptions from the regular in-move idle callback, so fixed that up too.

Thank you for the excellent bug report. It affects the stable release, unfortunately, so let's leave it open but "Fixed" for now.

When moving layer via shift+mouse button, sometime "AttributeError: 'NoneType' object has no attribute 'model'" error can happen if move didn't finish in time and was scheduled for later.

To reproduce:

Preparation:
1) start mypaint
2) in preferences assign "Move Layer" to <Shift>Button3 (shift-right mouse button)
3) zoom out (zoom ~18% will suffice at my machine)
4) select big brush (e.g deevad/fill radius 6)
5) paint all over the screen. (e.g. paint giant diagonal line from top left to bottom right cornor)

Bug reproduction:
6) Press Shift
7) Click RMB to initiate moving

Next steps must be done quickly
8) move layer. Stop moving.
9) release shift key
10) release RMB

There will be error box (attach#1)

What's going on:
If movement takes too long, mypaint schedules moving to be done later. Sometimes this 'later' comes so much later, that some variables are already cleared and can't be used.

What's happening in the code:

in SpringLoadedModeMixin around gui/canvasevent.py:913
key_release_cb pops mode, around canvaseven:767 pop calls old_mode.leave(), which calls InteractionMode.leave() that clears doc at line canvasevent.py:180 [ok, that's confusing, I'll add stacktrace to the attach #2]

BUT. LayerMoveMode might be not finished with doc yet:
around canvasevent.py:1438 it calls gobject.idle_add(self._finalize_move_idler), so after a while GTK decides to finalize movement.

BUT. At that point LayerMoveMode.doc is already cleared thanks to key_release_cb/pop/leave

So basically
Attach#1: picture
Attach#2: contains two callstacks: first was gained by adding _setattr to LayerMoveMode and printing when doc changes to None

second by adding traceback.print_stack() into _finalize. As you can see _finalize_move_idler got called after

Interesting thing to note:
finalizer calls _drag_cleanup, which calls doc.modes.pop. But key_release_cb also calls doc.modes.pop(well, right now it's not a problem, since _finalizer displays error)

Workaround:
adding ignore_modifier=True to LayerMoveMode.__init__ seems to work
(see https://gitorious.org/~zrzz/mypaint/librarian-mypaint/commit/a59a5b7766c02e834c0af699830d3bdfa3cbfa2a )

but I have no idea what problems it might cause