Bug 12471 – XPathResult not invalidated for documents retrieved via XMLHttpRequest

• Status: RESOLVED
• Resolution: FIXED
• Component: XML

[Alexey Proskuryakov]

XPathResult relies on DOM Mutation events for invalidation, but these don't work for documents retrieved via XMLHttpRequest (addEventListener() checks for document()->attached(), and bails out if it's not).

This causes several layout test failures:
dom/svg/level3/xpath/XPathResult_invalidIteratorState_ORDERED_NODE_ITERATOR_TYPE.svg
dom/svg/level3/xpath/XPathResult_invalidIteratorState_UNORDERED_NODE_ITERATOR_TYPE.svg
dom/svg/level3/xpath/XPathResult_invalidIteratorState_ANY_TYPE.svg
dom/svg/level3/xpath/XPathResult_iterateNext_INVALID_STATE_ERR.svg

[Alexey Proskuryakov]

This check for document()->attached() was added in r8557 <http://trac.webkit.org/projects/webkit/changeset/8557> as a part of a fix to memory leaks on eBay.

[Alexey Proskuryakov]

Created attachment 12832 [details]
work in progress

It seems that DOM events don't work for createDocument() and XMLHttpRequest-based documents in Firefox and Opera, too.

I've tried rewriting XPathResult to use domTreeVersion for invalidation - but that also doesn't work in such documents (the DOM version is mostly updated from attach()/detach()).

[Alexey Proskuryakov]

Created attachment 30828 [details]
proposed fix

DOM tree version magically works in ToT.

[Darin Adler]

Comment on attachment 30828 [details]
proposed fix

> + * Copyright (C) 2006, 2009 Apple Computer, Inc.

The preferred form is "<year> Apple Inc. All rights reserved."

I sometimes worry that the single document-wide tree version number is too blunt a tool; glad it works better here than what we were doing before!

r=me

[Alexey Proskuryakov]

Committed r44317.