Mark shadow root layout tree for rebuilding from style recalc.

Mark shadow root layout tree for rebuilding from style recalc.

The shadow root is added with NeedsReattachStyleChange. Let style
recalc translate that into a setNeedsReattachLayoutTree() instead of
doing it as part of the DOM operation.

This is a step towards marking nodes for layout tree rebuilding from
style recalc only. If we can make this an invariant, we can take
advantage of the fact that the flat-tree cannot change between marking
and the actual rebuild.

R=nainar@chromium.org,esprehn@chromium.org
BUG=595137

[rune, 2017-03-20 22:10:48]

The CQ bit was checked by rune@opera.com to run a CQ dry run

[commit-bot: I haz the power, 2017-03-20 22:11:41]

Dry run: CQ is trying da patch. Follow status at
 
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[esprehn, 2017-03-20 23:12:08]

Can you explain how having the flat tree not be able to change between then
marking the bits and when using them is useful?

[rune, 2017-03-20 23:20:08]

On 2017/03/20 23:12:08, esprehn wrote:
> Can you explain how having the flat tree not be able to change between then
> marking the bits and when using them is useful?

We could mark setChildNeedsReattachLayoutTree on flat tree ancestors and do
layout tree rebuilding in the flat-tree order, which would be beneficial for
whitespace re-attachment.

[esprehn, 2017-03-20 23:31:59]

On 2017/03/20 at 23:20:08, rune wrote:
> On 2017/03/20 23:12:08, esprehn wrote:
> > Can you explain how having the flat tree not be able to change between then
> > marking the bits and when using them is useful?
> 
> We could mark setChildNeedsReattachLayoutTree on flat tree ancestors and do
layout tree rebuilding in the flat-tree order, which would be beneficial for
whitespace re-attachment.

We could still do that with the current system though? Walk down to bits until
we hit a child where needsRebuildLayoutTree(), then begin at the parent and
traverse through the composed tree child list for that node. That should be
faster than resolving the slotting for every sibling for every attachment.

Specifically if you appendChild a child to a slotted list, today we walk down to
that node, find the next sibling in the flat tree, and insert into the layout
tree. We never need to resolve all the slotted siblings to traverse through the
DOM.

Simple might be better though, do you have a prototype patch/doc of the
whitespace algorithm you're talking about?

[commit-bot: I haz the power, 2017-03-21 00:56:33]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2017-03-21 00:56:34]

Dry run: This issue passed the CQ dry run.

[rune, 2017-03-21 01:22:11]

On 2017/03/20 23:31:59, esprehn wrote:
> On 2017/03/20 at 23:20:08, rune wrote:
> > On 2017/03/20 23:12:08, esprehn wrote:
> > > Can you explain how having the flat tree not be able to change between
then
> > > marking the bits and when using them is useful?
> > 
> > We could mark setChildNeedsReattachLayoutTree on flat tree ancestors and do
> layout tree rebuilding in the flat-tree order, which would be beneficial for
> whitespace re-attachment.
> 
> We could still do that with the current system though? Walk down to bits until
> we hit a child where needsRebuildLayoutTree(), then begin at the parent and
> traverse through the composed tree child list for that node. That should be
> faster than resolving the slotting for every sibling for every attachment.

By resolving slotting, you mean looking up assignedSlot() when marking ancestors
for childNeedsReattachLayoutTree()?

> Specifically if you appendChild a child to a slotted list, today we walk down
to
> that node, find the next sibling in the flat tree, and insert into the layout
> tree. We never need to resolve all the slotted siblings to traverse through
the
> DOM.

When the distribution is up-to-date (which it has to be before recalc/rebuild),
you have an up-to-date m_distributedNodes vector you can walk. I'm not sure I
understand the difference with walking the light-tree children. One difference
though, is that you'd need to walk the whole shadow tree since you don't know
how to find the path down to the correct slot if you haven't marked the
flat-tree ancestor chain. E.g.:

<div id="host">
  <:shadowroot>
    <div>
     <div></div>
     <div></div>
     <div></div>
     <div></div>
    </div>
    <div>
      <slot></slot> <span>B</span>
    </div>
  </:shadowroot>
  <div id="slotted">A</div>
</div>

Marking in tree-of-trees means marking #slotted for re-attachment will mark
#host for childNeeds*, and we'd have to traverse all the divs in addition to the
subtree with the slot. Marking the flat-tree ancestors through assignedSlot for
slotted nodes, means we can do a normal childNeeds* based traversal down to the
slot and its distributed nodes.

> Simple might be better though, do you have a prototype patch/doc of the
> whitespace algorithm you're talking about?

Very much work in progress:

https://docs.google.com/document/d/1CFk5_3Gx8bPSsKdJ2twZhceUIrtySasdsij9t1niN...

I do have code. I'll try to upload something tomorrow.

[rune, 2017-03-21 22:41:48]

On 2017/03/21 01:22:11, rune wrote:

> Very much work in progress:
> 
>
https://docs.google.com/document/d/1CFk5_3Gx8bPSsKdJ2twZhceUIrtySasdsij9t1niN...
> 
> I do have code. I'll try to upload something tomorrow.

I was not able to do that today ...

[rune, 2017-03-22 08:18:20]

CL included in https://codereview.chromium.org/2760233004/, closing.

[rune, 2017-03-22 08:18:28]

Description was changed from

==========
Mark shadow root layout tree for rebuilding from style recalc.

The shadow root is added with NeedsReattachStyleChange. Let style
recalc translate that into a setNeedsReattachLayoutTree() instead of
doing it as part of the DOM operation.

This is a step towards marking nodes for layout tree rebuilding from
style recalc only. If we can make this an invariant, we can take
advantage of the fact that the flat-tree cannot change between marking
and the actual rebuild.

R=nainar@chromium.org,esprehn@chromium.org
BUG=595137
==========

to

==========
Mark shadow root layout tree for rebuilding from style recalc.

The shadow root is added with NeedsReattachStyleChange. Let style
recalc translate that into a setNeedsReattachLayoutTree() instead of
doing it as part of the DOM operation.

This is a step towards marking nodes for layout tree rebuilding from
style recalc only. If we can make this an invariant, we can take
advantage of the fact that the flat-tree cannot change between marking
and the actual rebuild.

R=nainar@chromium.org,esprehn@chromium.org
BUG=595137
==========