caching parent makes firefox faster in 3 times :) but it still incredibly slow. Running browser with http://www.mozilla.org/projects/minefield/ and focusing "search mozilla: " textbox takes 30 secundes before and 10 secundes after. Opera takes nothing.

What if the DOM changes?

Another (related) problem is that you're keeping a weak reference to the parent, so if the object goes away and you later assume is there - boom! If you retain it (add to the reference count) you will create a cyclical reference (because parents retain their children, and now those children would also retain the parent) so then it would be very important that you break that cycle at some point to avoid memory leaks.

Yes, that's just an example. I think we can use weak reference here but make sure we invalidate it in invalidateChildren method. Though actually parent caching doesn't bring a win always. Sometimes only.

I recently did some more thorough testing, and found that the most performance problems happen when VoiceOver hits plain text. As long as I navigate through a series of link accessibles, buttons or text fields, like on www.google.com, performance is not as snappy as with Safari, but response comes back in under 1 second, too. Only if I then continue and get to search results that have plain text nodes, the performance gets bad. As soon as I am past the text on another non-text element, I'm better again. Does this belong here with the other discussion, or should I file a separate bug for this?

Comment on attachment 337439 [details] [diff] [review] cache parent obsoleted by bug 455443

As is is now, the performance does not seem to be so bad anymore. There are a few areas where Firefox appear to be "Busy" for a bit too long.

(In reply to Hub Figuiere [:hub] from comment #6) > As is is now, the performance does not seem to be so bad anymore. great to hear > There are a few areas where Firefox appear to be "Busy" for a bit too long. let's keep the bug open, that shouldn't happen (I bet your machine is not slow).