240850 - nsContentUtils::GetFirstDifferentAncestors should use nsIContent

Status: RESOLVED INVALID

(Core :: DOM: Core & HTML, defect)

Description

[Boris Zbarsky [:bzbarsky]]

nsContentUtils::GetFirstDifferentAncestors uses nsIDOMNode, which involves
calling GetParentNode a bunch.  GetParentNode has to QI; this makes it
noticeably slower than nsIContent::GetParent.  I suspect that we could win
easily here if we only do two QIs up front and then work with nsIContent
throughout the method....

Thoughts?  Maybe the caller is what should be working with nsIContent, by the
way....

Comment 1

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Actually, the entire compare-tree-position code should be rewritten. The code in
txXPathNodeUtils::comparePosition is a lot faster, and even that could be
improved in the cases where we don't have to take multiple documents into account.

Comment 2

[Boris Zbarsky [:bzbarsky]]

Out of curiousity, if we improve compareTreePosition enough, could
txXPathNodeUtils::comparePosition be removed?  Or does standalone need it?

Comment 3

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

transformiix doesn't live in, or link to, gklayout.dll so that won't work.
Furthermore it needs to be able to compare nodes in different trees, as well as
compare nodes that are in subtrees outside of the main document, which probably
would be a perfhit on the nsContentUtils one.

Comment 4

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Attachment: WIP

This is not done and doesn't even compile yet iirc. Just don't want to loose
the patch

Comment 5

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

All this code is long since rewritten now.