Bug 43586 – Use the HTML5 TreeBuilder for Fragment Parsing

• Component: New Bugs
• Status: RESOLVED FIXED

[Eric Seidel (no email)]

Use the HTML5 TreeBuilder for Fragment Parsing

[Eric Seidel (no email)]

Created attachment 63656 [details]
Patch

[Eric Seidel (no email)]

Created attachment 63788 [details]
crashes due to DocumentParser use after free?

[Eric Seidel (no email)]

Created attachment 63837 [details]
works

[Eric Seidel (no email)]

Created attachment 63903 [details]
Patch

[Darin Adler]

Comment on attachment 63903 [details]
Patch

>      // FIXME: parse error
> +    // fragment case handled by callers.
>      mergeAttributesFromTokenIntoElement(token, m_openElements.bodyElement());

This is a confusing comment. Exactly what is handled by the callers? Is there some way to assert or check that?

It's also strange that the comment has a period but does not start with a capital letter.

> +// This is a direct transcription of step 4 from:
> +// http://www.whatwg.org/specs/web-apps/current-work/multipage/the-end.html#fragment-case
> +HTMLTokenizer::State tokenizerStateForContextElement(Element* contextElement, bool reportErrors)
> +{
> +    if (!contextElement)
> +        return HTMLTokenizer::DataState;
> +
> +    const QualifiedName& contextTag = contextElement->tagQName();
> +
> +    if (contextTag.matches(titleTag) || contextTag.matches(textareaTag))
> +        return HTMLTokenizer::RCDATAState;
> +    if (contextTag.matches(styleTag)
> +        || contextTag.matches(xmpTag)
> +        || contextTag.matches(iframeTag)
> +        || (contextTag.matches(noembedTag) && HTMLTreeBuilder::pluginsEnabled(contextElement->document()->frame()))
> +        || (contextTag.matches(noscriptTag) && HTMLTreeBuilder::scriptEnabled(contextElement->document()->frame()))
> +        || contextTag.matches(noframesTag))
> +        return reportErrors ? HTMLTokenizer::RAWTEXTState : HTMLTokenizer::PLAINTEXTState;
> +    if (contextTag.matches(scriptTag))
> +        return reportErrors ? HTMLTokenizer::ScriptDataState : HTMLTokenizer::PLAINTEXTState;
> +    if (contextTag.matches(plaintextTag))
> +        return HTMLTokenizer::PLAINTEXTState;
> +    return HTMLTokenizer::DataState;
> +}

It is better to write this code out like this rather than using functions in the element classes?

Have to run, no time to read the rest of the patch right now.

[Eric Seidel (no email)]

(In reply to comment #5)
> (From update of attachment 63903 [details])
> >      // FIXME: parse error
> > +    // fragment case handled by callers.
> >      mergeAttributesFromTokenIntoElement(token, m_openElements.bodyElement());
> 
> This is a confusing comment. Exactly what is handled by the callers? Is there some way to assert or check that?

Agreed.  Basically that function should be killed or renamed.  I should kill it or rename it.  It was a victim of our first effort to split the original HTMLTreeBuilder.cpp into multiple data structures.  That function ended up on the wrong side of the split IMO.

> It's also strange that the comment has a period but does not start with a capital letter.

I'm just going to remove the comment, it's not helpful.

> It is better to write this code out like this rather than using functions in the element classes?

Interesting thought.  I think in this case, the if-cascade using tag names is better, since it more closely matches the spec.  But I agree with you, many times it's better to shove this stuff into the DOM implementations.

> Have to run, no time to read the rest of the patch right now.

Thanks for the look!

[Adam Barth]

Comment on attachment 63903 [details]
Patch

WebCore/html/HTMLTreeBuilder.cpp:824
 +  bool isFragmentCaseWithoutBody(const HTMLElementStack* elementStack)
This is wrong.  You probably want to cache this information.  Basically, you're not covering the case when there are more than three elements on the stack.

WebCore/html/HTMLTreeBuilder.cpp:2902
 +          m_document->finishedParsing();
I would use a return here instead of an else.  The else is uglies.

WebCore/html/HTMLTreeBuilder.cpp:2900
 +      // Warning, this may delete the parser, so don't try to do anything else after this.
Please move this comment closer to m_document->finishedParsing().

WebCore/html/HTMLTreeBuilder.cpp:2914
 +              RefPtr<Node> newChild = fragment->document()->adoptNode(child, ec);
This is dangerous.  We need a version of adoptNode that doesn't run JavaScript.  Also, you should grab references to all the children first and then move them to their new home.

WebCore/html/HTMLTreeBuilder.h:212
 +          RefPtr<Document> m_dummyDocumentForFragmentParse;
m_dummyDocumentForFragmentParse -> m_dummyDocumentForFragmentParsing ?

WebCore/html/HTMLTreeBuilder.h:221
 +      FragmentParsingContext m_fragmentContext;
Insert blank line above.

[Eric Seidel (no email)]

Created attachment 63915 [details]
Patch

[Adam Barth]

Comment on attachment 63915 [details]
Patch

WebCore/html/HTMLElementStack.cpp:155
 +      return m_bodyElement;
!!m_bodyElement  ?

[Eric Seidel (no email)]

Created attachment 63916 [details]
Patch for landing

[Eric Seidel (no email)]

Comment on attachment 63916 [details]
Patch for landing

Failed runner.html on the commit-bot, will need one more tweak.

[Eric Seidel (no email)]

Committed r65006: <http://trac.webkit.org/changeset/65006>