[patch][cla][dijit.Dialog] in 1.6: destroying dialog in `onHide` causes error

[Kenneth G. Franqueiro]

In Dojo 1.6, dijit.Dialog's onHide event is triggered when the hide animation ends. This is generally awesome.

However, the current implementation causes an error in one of the more frequent use-cases I've seen: hooking destruction of the dialog to this event. It throws because the code path that executes onHide expects the _fadeOutDeferred instance member to still be set after onHide completes - however, when the widget is destroyed, the deferred is cancelled, and its canceller function deletes this instance member.

I've attached a patch that simply checks for existence of the member first.

For those who need to work around this in 1.6, it can be done simply by using setTimeout to defer destruction until after the offending code path has completed:

dlg.connect(dlg, 'onHide', function() {
    setTimeout(function() { dlg.destroyRecursive(); }, 0);
});

[Kenneth G. Franqueiro]

Simple isolated test case

[bill]

Sure, I'm fine w/it if you add an automated test (not using robot, just close the dialog using hide() or something).

Note that I have it on my list to get rid of the JS animations in widgets (replacing them with CSS animations), not sure when that will be done (hopefully in 1.7) but that will change the landscape some. Still good to have the test checked in though.

[Kenneth G. Franqueiro]

Revised change, no need for condition; hook on Deferred resolution

[Kenneth G. Franqueiro]

(In [24031]) Dialog: don't fire onHide until after promise is resolved to avoid explosions if Dialog is destroyed within onHide. Added manual test to test_Dialog.html reflecting the issue. Fixes #12436 !strict

[bill]

#17088 is a duplicate of this ticket.