313278 - <script xmlns="http://www.w3.org/1999/xhtml"> leaks namespace to younger siblings

Status: VERIFIED FIXED

(Core :: XML, defect)

Description

[Benjamin Smedberg]

See attached testcase, basically

<element xmlns="1">
  <script xmlns="xhtml" src="foo" />
  <foo />
</element>

Foo has the wrong namespace. Which ends up screwing some significant number of
published SVG examples and a lot of mixed XHTML+SVG content.

Comment 1

[Benjamin Smedberg]

Attachment: Testcase

Comment 2

[Boris Zbarsky [:bzbarsky]]

This regressed between 2004-12-15-06 and 2004-12-16-08, so looks like we somehow
fail to pop off the default namespace after the expat landing...

Comment 3

[Boris Zbarsky [:bzbarsky]]

At a guess, we're adding the namespace binding, then bailing out when we block,
and never removing the binding...

Perhaps this is xmlparse.c line 2324?  We call the endElementHandler in line
2321, but we want to make it down to line 2341 before returning the error, no? 
Or do we not want to clear the tempPool here?  We definitely want that loop over
the bindings, I'd think.

Comment 4

[Peter Van der Beken [:peterv]]

Attachment: v1

It's going to be hard for me to test this soon, so if anyone could verify that
this fixes the problem I'd be grateful.

Comment 5

[Boris Zbarsky [:bzbarsky]]

Attachment: Nonempty tag testcase

Comment 6

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 200359 [details] [diff] [review]
v1

Yeah, that fixes both testcases. r+sr=bzbarsky.

Can you check this in, or do you want me to?

Comment 7

[Asa Dotzler [:asa]]

please land and verify on the trunk. 

Comment 8

[Peter Van der Beken [:peterv]]

If you could, I'm off for the night.

Comment 9

[Boris Zbarsky [:bzbarsky]]

Fixed on trunk.

Comment 10

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 200359 [details] [diff] [review]
v1

Requesting 1.8 branch approval.  This is a pretty serious regression in our XML
parsing.  The fix is pretty safe -- just move our early returns to be not quite
so early to allow expat to pop the namespaces associated with the just-closed
element.

Comment 11

[Benjamin Smedberg]

Verified fixed on trunk, we should definitely take this on branch.

Comment 12

[Peter Van der Beken [:peterv]]

Checked in on branch. Thanks bz for your help.