Bug 6290 – XML documents with "html" in dtd name use XHTML entities when they shouldn't

• Status: RESOLVED
• Resolution: FIXED
• Component: WebKit Misc.

[Ian 'Hixie' Hickson]

Currently any XML file with a DOCTYPE whose "name" part is "html" triggers
"HTML" parsing mode. This is wrong. It should be based on the PUBLIC identifier.

TESTCASE: http://www.hixie.ch/tests/adhoc/xml/parsing/010.xml

This is a regression caused by bug 4301.

[Eric Seidel (no email)]

This is such a trivial edgecase.

[Ian 'Hixie' Hickson]

Not that much of an edgecase, XHTML2 will likely trip this, and even if XHTML2
isn't much in use today, it eventually may be. It would be sad for XHTML2
authors if they were confused by old builds of Safari.

The bug is that externalSubsetHandler() is checking |name|, but should be
checking |externalId|. Here are the IDs to compare against (taken from Mozilla's
list, with the MathML DTDs removed):

   "-//W3C//DTD XHTML 1.0 Transitional//EN"
   "-//W3C//DTD XHTML 1.1//EN"
   "-//W3C//DTD XHTML 1.0 Strict//EN"
   "-//W3C//DTD XHTML 1.0 Frameset//EN"
   "-//W3C//DTD XHTML Basic 1.0//EN"
   "-//WAPFORUM//DTD XHTML Mobile 1.0//EN"

[Eric Seidel (no email)]

Calling this a regression is pretty strong language.  As far as I can tell this test case only "passed" before, 
because html entities were not supported in xhtml at all in Safari.  Perhaps in the expat days this test case 
passed for valid reasons.  Now html entities are supported in anything that looks like an xhtml document.  
This one happens to look enough like xhtml that we treat it as one.  Eventually, we'll need a HashSet, and 
a list of allowed doc types.

[Ian 'Hixie' Hickson]

Why not just a set of six string comparisons in the place of the one comparison
you have now? You'll almost always either match the first one or not enter the
function, so it shouldn't be a perf hit.

It's a regression because before you used to not have this bug, and now you do.
That's what a regression is: the introduction of a new bug.

[Eric Seidel (no email)]

Created attachment 5364 [details]
fix & test

[Ian 'Hixie' Hickson]

You have this line:

+        || (extId == "-//W3C//DTD XHTML Basic 1.0//EN")

...in there twice. (In the Mozilla codebase they have the MathML 2.0 DOCTYPE
where you have the second occurance, but you don't want to put that in because
that isn't XHTML and has different entities. It's probably sort-of ok to have
the XHTML+MathML DOCTYPEs in there, I guess, though. It'll at least work for
some pages, anyway.)

Other than that, looks great. Thanks. :-)

[Maciej Stachowiak]

Comment on attachment 5364 [details]
fix & test

r=me

[Joost de Valk (AlthA)]

Removing keyword(s) since bug is fixed.

[Joost de Valk (AlthA)]

Removing keyword(s) since bug is fixed.

[Eric Seidel (no email)]

Removing Regression keyword from bugs already fixed.