SCOPE DESTRUCTION: Leak when asynchronously continuing the linking of directives on scopes that are already destroyed

[lgalfaso]

When a directive uses a templateUrl, this directive is inside an ng-switch or an ng-if and the following events happen in this specific order

• The html of the ng-switch starts it's compilation
• The directive inside the ng-switch starts it's compilation, the template is requested from the server
• The synchronous part of the compilation ends
• The condition on the ng-switch changes so the directive should be displayed
• ng-switch transcludes the content of the ng-switch-when (only what it can as the template did not finish loading)
• The expression at ng-switch changes so the directive should not be displayed
• ng-switch destroys the scope
• The template finishes loading, the scope (even when it is already destroyed) is added to the element.data (here is where we have the leak)

Here is a plunker that shows this
http://plnkr.co/edit/tNMYYHgoebX1Vx0hWZAb?p=preview

Note: This issue is based on a modification of #9197

[caitp]

did we really need to create a new issue for this? it's basically the same thing as the one you were commenting on is it not?

[lgalfaso]

@caitp I created the new issue to be sure it is tracked as a leak and not as something that can cause no harm. That said, I am ok with just moving everything to the original issue

[lgalfaso]

Posted an update on #9079 that solves this issue

[petebacondarwin]

I commented on the two commits in #9079. I believe the commit that prevents double destroy events could cause memory leaks and proposed a different solution of actually recursing down the scope hierarchy and calling $destroy on all descendants, based on the fact that #9281 would make the scope hierarchy more conducive to this - currently this would fail to destroy transcluded scopes.
I think the second commit is the right thing to do assuming that we really have destroyed the descendant scopes.