933125 - make scoped style sheets work in anonymous content

Status: RESOLVED WONTFIX

(Core :: CSS Parsing and Computation, defect, P4)

Description

[Cameron McCormack (:heycam)]

Scoped style sheets should work in non-native anonymous content, both <style scoped> elements from the outside applying to elements in the shadow tree, and <style scoped> elements within the shadow tree applying to its siblings and any descendent real and non-native anonymous content.

Comment 1

[Cameron McCormack (:heycam)]

Followup from bug 926717.

Comment 2

[Cameron McCormack (:heycam)]

Attachment: WIP (v0.1)

Mostly works, but the layout/reftests/scoped-style/scoped-style-xbl-005.xhtml test doesn't.  Looks like Element::BindToTree isn't called for the <span> element that is in the deeper shadow tree, so my SetIsElementInStyleScope call in there isn't getting hit.

Comment 3

[Cameron McCormack (:heycam)]

Attachment: patch

Comment 4

[Cameron McCormack (:heycam)]

(In reply to Cameron McCormack (:heycam) from comment #2)
> Looks like Element::BindToTree isn't called for the <span> element that is in the
> deeper shadow tree

I was wrong about that bit.

Comment 5

[Cameron McCormack (:heycam)]

Hmm, got a couple of oranges for one of the added reftests, but not on all platforms:

https://tbpl.mozilla.org/?tree=Try&rev=074eea814efc

Comment 6

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 825624 [details] [diff] [review]
patch

>+  FlattenedChildIterator it(aElement);

Hmm.  So in cases when the <style> is a kid of an anonymous element, this will iterate too many things.  But I'm not sure we have a sane way to iterate the set of things we _really_ want here.  Blake, do we?

Also, we should raise a web components spec issue about how <style> in anonymous content interacts with document.styleSheets (ideally it doesn't show up there, I'd think; add tests for this?).

> HasScopedStyleSheetChild(nsIContent* aContent)

Again, using FlattenedChildIterator is a bit weird here.  If this incorrectly reports "true" is that ok?

> UpdateIsElementInStyleScopeFlagOnSubtree

And the same worries about FlattenedChildIterator here.

At the very least all of this stuff needs extensive comments explaining why the non-intuitive FlattenedChildIterator uses make sense and what assumptions make them OK, and perhaps some asserts of these assumptions if that's possible.

Comment 7

[Cameron McCormack (:heycam)]

Hmm, when you say it will iterate too many things, which additional things is it going to be iterating that we don't want?

Comment 8

[Boris Zbarsky [:bzbarsky]]

Consider this DOM:

  <div><span></span></div>

Say the <div> has a binding attached that has:

  <content>
    <p>
      <style scoped></style>
      <children></children>
    </p>
  </content>

The FlattenedChildIterator on <p> will see the <span> as a child of <p>.  But it seems pretty wrong to me for the <style scoped> there to apply to that <span>....

And from the other direction:

  <div><style scoped></style></div>

and the binding:

  <content>
    <p>
      <span>Some stuff</span>
      <children></children>
    </p>
  </content>

I don't think we want the <p> to be treated as a style scope for the <style> in this case.

Comment 9

[Boris Zbarsky [:bzbarsky]]

Or I guess to actually answer comment 7, it'll iterate DOM kids of the bound element whose insertion point is under the element whose flattened children we're iterating.

Comment 10

[Blake Kaplan (:mrbkap) (inactive)]

Comment on attachment 825624 [details] [diff] [review]
patch

After talking to dglazkov on IRC, it seems like nobody has really thought about what this means, so we probably should hold off on it until we figure that out...

Comment 11

[Cameron McCormack (:heycam)]

OK let's leave it for now.

Comment 13

[Jonathan Watt [:jwatt]]

Per bug 1345702 comment 2 we're removing scoped style sheets.