ShadowRoot.getElementById() returns removed node.

• Component: DOM
• Status: RESOLVED FIXED

[Kent Tamura]

tkent doesn't like registering a shadow ID in insertedIntoDocument() and unregistering it in removedFromDocument().

See https://bugs.webkit.org/show_bug.cgi?id=62116#c5

[Hajime Morrita]

I'm sorry for the lack of response on this issue.

Another possible idea is to have insertedIntoScope() in addition or instead of insertedIntoTree().
As we pulled some functionality up from Document to TreeScope,
associated part on Element and Node is also need to be generalized.

[Roland Steiner]

I added a comment at https://bugs.webkit.org/show_bug.cgi?id=62116#c14

[Roland Steiner]

In Summary: 

If an element has an ID it just unconditionally calls treeScope()->registerID() from insertedIntoTree(), and treeScope()->unregisterID() from willRemove().

Nodes that are not in the Document are inserted into a new TreeScope-derived class DocumentAssociated, of which there is exactly 1 instance per Document, owned by the Document.

DocumentAssociated::registerID()/unregisterID() are no-ops.

With this, we can also get rid of all the instances where inDocument() is queried.

[Dimitri Glazkov (Google)]

(In reply to comment #3)
> In Summary: 
> 
> If an element has an ID it just unconditionally calls treeScope()->registerID() from insertedIntoTree(), and treeScope()->unregisterID() from willRemove().
> 
> Nodes that are not in the Document are inserted into a new TreeScope-derived class DocumentAssociated, of which there is exactly 1 instance per Document, owned by the Document.

I remember you talking about this a while back. It seems like a pretty cool idea. What are the compatibility implications? Any observable changes in behavior - especially if we introduce treeScope() accessor on an Element as public API?

> 
> DocumentAssociated::registerID()/unregisterID() are no-ops.
> 
> With this, we can also get rid of all the instances where inDocument() is queried.

[Roland Steiner]

(In reply to comment #4)
> I remember you talking about this a while back. It seems like a pretty cool idea. What are the compatibility implications? Any observable changes in behavior - especially if we introduce treeScope() accessor on an Element as public API?

I guess it'll have the same implications as the shadow root if we allow it to be reached from JS (by accessing it directly via treeScope, or by climbing the parent pointers) - i.e., mainly what is the node type and IDL of this DocumentAssociated node?

[Roland Steiner]

Actually, we might be able to just use a DocumentFragment for this if likewise we extend DocumentFragment to also be a TreeScope.

[Hajime Morrita]

Now TreeScope is landed. We need to test the scope is working as intended.

[Roland Steiner]

Created attachment 128431 [details]
Test

Turns out, removing elements from a shadow DOM doesn't properly update the ID cache. See attached test file (using removeChild) and also the test bug 78473 attachment 127944 [details] (using innerHTML).

[Hajime Morrita]

Created attachment 144505 [details]
Patch

[Hajime Morrita]

Hi Ryosuke, Dimitri, could you take a look?
It's a simple bug fix.

[WebKit Review Bot]

Comment on attachment 144505 [details]
Patch

Clearing flags on attachment: 144505

Committed r118804: <http://trac.webkit.org/changeset/118804>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.