Bug 13508

Summary: Compiling WebKit with libxml 2.6.27 or greater causes some layout tests to fail
Product: WebKit Reporter: Matt Perry <mpComplete>
Component: XMLAssignee: Nobody <webkit-unassigned>
Status: RESOLVED FIXED    
Severity: Normal CC: mrowe
Priority: P2    
Version: 523.x (Safari 3)   
Hardware: All   
OS: All   
Attachments:
Description Flags
patch mjs: review+

Matt Perry
Reported 2007-04-26 12:02:07 PDT
There was a bug in libxml prior to 2.6.27 that caused entities to be inserted twice. See http://bugzilla.gnome.org/show_bug.cgi?id=159219 . WebKit has a workaround for this: hackAroundLibXMLEntityBug. However, in libxml 2.6.27, this bug was fixed, so the workaround causes incorrect behavior with that version. A simple #if/#else can fix this.
Attachments
patch (1.32 KB, patch)
2007-04-26 12:06 PDT, Matt Perry 		mjs: review+
DetailsFormatted DiffDiff
View All Add attachment proposed patch, testcase, etc.
Matt Perry
Comment 1 2007-04-26 12:06:54 PDT
Created attachment 14209 [details] patch No layout test is needed. Existing layout tests already fail when using libxml 2.6.27 or greater. For example, dom/xhtml/level3/core/nodegettextcontent14 fails.
Dave Hyatt
Comment 2 2007-04-26 12:16:43 PDT
Comment on attachment 14209 [details] patch This is defined by libxml in its headers?
Matt Perry
Comment 3 2007-04-26 12:41:07 PDT
Yes. LIBXML_VERSION is defined in libxml/xmlversion.h.
Darin Adler
Comment 4 2007-04-26 12:44:36 PDT
Comment on attachment 14209 [details] patch Ideally this should be done as a runtime check, not compile time, since libxml is often dynamically linked in. But looking at xmlversion.h I don't see a way to make the a check at runtime.
Maciej Stachowiak
Comment 5 2007-04-26 22:35:13 PDT
Comment on attachment 14209 [details] patch I agree that a runtime check would be better, but since that apparently can't be done, this seems fine. r=me
Mark Rowe (bdash)
Comment 6 2007-04-26 23:08:43 PDT
Landed in r21136.
Note You need to log in before you can comment on or make changes to this bug.