REGRESSION(r167870): Crash in simple line layout code with :after

[Antti Koivisto]

<style>
a { display: none; }
a:after { content: "bar"; }
div:hover a { display: inline-block }
</style>
<div>foo<a></a></div>

[Antti Koivisto]

<rdar://problem/16977696>

[Antti Koivisto]

Created attachment 231834 [details]
patch

[Darin Adler]

Comment on attachment 231834 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=231834&action=review

Looks great.

> Source/WebCore/rendering/RenderBlockFlow.cpp:3423
>  const SimpleLineLayout::Layout* RenderBlockFlow::simpleLineLayout() const
>  {
> -    if (m_lineLayoutPath == UndeterminedPath)
> -        const_cast<RenderBlockFlow&>(*this).m_lineLayoutPath = SimpleLineLayout::canUseFor(*this) ? SimpleLinesPath : LineBoxesPath;
> -
> -    if (m_lineLayoutPath == SimpleLinesPath)
> -        return m_simpleLineLayout.get();
> -
> -    const_cast<RenderBlockFlow&>(*this).createLineBoxes();
> -    return nullptr;
> +    ASSERT(m_lineLayoutPath <= SimpleLinesPath || !m_simpleLineLayout);
> +    return m_simpleLineLayout.get();
>  }

Maybe we should move this back to the header file and make it an inline function again.

I don’t think the assertion is strong enough. It’s seems like it’s not OK to get a non-nullptr value if m_lineLayoutPath is UndeterminedPath. But maybe we can’t put the stronger assertion in because we still need to get at m_simpleLineLayout in the time window before we determine whether to use it or not.

[Antti Koivisto]

Created attachment 231839 [details]
patch 2

[Antti Koivisto]

> I don’t think the assertion is strong enough. It’s seems like it’s not OK to get a non-nullptr value if m_lineLayoutPath is UndeterminedPath. But maybe we can’t put the stronger assertion in because we still need to get at m_simpleLineLayout in the time window before we determine whether to use it or not.

You are right. I tightened the assert and now drop the simple line layout on path invalidation.

[Antti Koivisto]

https://trac.webkit.org/r169189