GTK: Fancy html5 draggy stuff.

GTK: Fancy html5 draggy stuff.

Also, don't crash in release mode when the tab contents is deleted during a drag.

1) Source side: when the tab contents is going down, abort the current drag (if any). Initiate drags on a GtkInvisible (matching firefox, among other programs I assume). This lets us give up ownership of the drag widget and let the drag unwind.
2) Destination side: make a seemingly harmless GTK assert point to a crbug bug. This warrants further investigation.

BUG=16249
TEST=drag around ntp thumbnails

BUG=18557
In release mode, you should be able to do all the following without crashing or getting any asserts besides the two mentioned in 18557
TEST=drag something from gedit over a tab that closes
TEST=drag something from the render view over a tab that closes
TEST=drag something from the render view off of the browser from within a tab that closes
TEST=repeat all the above with a tab that is being swapped out according to the repro steps in bug 16073


Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=22542

[Evan Martin, 2009-08-05 00:22:10]

http://codereview.chromium.org/159889/diff/1/3
File chrome/browser/tab_contents/tab_contents_view_gtk.cc (right):

http://codereview.chromium.org/159889/diff/1/3#newcode879
Line 879: MessageLoopForUI::current()->RemoveObserver(this);
Are you sure this runs in all cases?  I wonder if there's a ScopedObserver or
something, in case this doesn't get called if the tab closes itself while we're
mid-drag.

[tony, 2009-08-05 00:24:49]

http://codereview.chromium.org/159889/diff/1/3
File chrome/browser/tab_contents/tab_contents_view_gtk.cc (right):

http://codereview.chromium.org/159889/diff/1/3#newcode741
Line 741: tab_contents()->render_view_host()->DragSourceMovedTo(
This is probably where the d&d bugs I fixed on Windows matter.  You need to make
sure that tab_contents() is valid if (1) the drag source tab is closed or (2)
the drag source tab is swapped out.

http://codereview.chromium.org/159889/diff/1/3#newcode870
Line 870: tab_contents()->render_view_host()->DragSourceCancelledAt(
I bet tab_contents() can be dead by here as well.

http://codereview.chromium.org/159889/diff/1/3#newcode886
Line 886: tab_contents()->render_view_host()->DragSourceEndedAt(
And here.

[Evan Stade, 2009-08-05 00:30:08]

yea, you guys are right, it does crash if the tab is closed during the drag. But
it also does that without this patch. Will look into fixing that problem now as
well.

Tony, what happens when the tab is swapped out, and how can I test it?

[tony, 2009-08-05 00:41:23]

On 2009/08/05 00:30:08, Evan Stade wrote:

> Tony, what happens when the tab is swapped out, and how can I test it?

The steps in comment #3 should cause a tab swap:
http://code.google.com/p/chromium/issues/detail?id=16073

[Evan Stade, 2009-08-05 20:55:30]

updated

[tony, 2009-08-05 21:32:43]

LGTM