[Webkit-unassigned] [Bug 179682] Incorrect bounds inside <mover>/<munder> when a stretchy operator is present

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=179682

--- Comment #46 from Minsheng Liu <lambda at liu.ms> ---
(In reply to zalan from comment #45)
> Comment on attachment 328250 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=328250&action=review
> 
> >> Source/WebCore/rendering/mathml/RenderMathMLUnderOver.cpp:73
> >> +    }
> > 
> > I would write:
> > 
> >     RenderBox* fixingBox = stretchyOperator;
> > 
> > Instead of writing the static_cast.
> > 
> > I also would write this as a for loop:
> > 
> >     for (RenderBox* box = stretchyOperator; box != ancestor; box = box->parentBox()) {
> >         box->setNeedsLayout();
> >         box->layout();
> >     }
> > 
> > But also, it is really strange to do synchronous calls to layout after setNeedsLayout. We need a layout expert to weigh in on this. It seems wrong. Antti? Zalan?
> 
> Normally you set the dirty bit on the renderer soon after you figure its
> position/size are invalid (usually result of a style change), schedule an
> async layout and here you would just call layoutIfNeeded(). 
> However since in this function we are already have a valid layout context 
> (through RenderMathMLUnderOver::layoutBlock), this looks more like a
> multi-pass layout where the descendants need to be readjusted after some
> ancestors/siblings etc layouts are done. I don't know much about MathML so
> if that's not the case, then the setNeedsLayout calls are wrong here and
> they need to be moved somewhere else (see above). 
> Even if it was a type of multi-pass layout, the stretch operation should
> probably dirty the descendants (since that's where the children get invalid)
> and you'd just call layoutIfNeeded here.

So what is the difference between layoutIfNeeded() and layout()? Is it that when layout() is called then a layout is carried out immediately, while when layoutIfNeeded() is called then a layout is not executed until we are doing some actual layout?

Second, this is indeed a multi-pass layout. The case I try to handle is that I am stretching a child deeply nested so after the stretching I need to layout not only the child but also all its ancestors. It is like:

<munder> // we are executing layout here
  <msub>
    <mover>
      <msup>
        <mo>...</mo> // the <mo> being stretched
      </msup>
    </mover>
  </msub>
  ...
</munder>

So if I understand you correctly, I could just set <msub>, <mover>, <msup>, all the way to <mo> dirty, and then call layoutIfNeeded() on <musb>? That does seem better! Thanks!

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.webkit.org/pipermail/webkit-unassigned/attachments/20171204/69591d1e/attachment-0001.html>