453832 - nsIAccessNode objects in cache aren't invalidated when they are shown

Status: RESOLVED WORKSFORME

(Core :: Disability Access APIs, defect)

Description

[alexander :surkov (:asurkov)]

scenario is:

we navigate accessible by nsISimpleDOMNode and store nsIAccessNode for hidden elements. When these elements become visible and if we still navigate the tree by nsISimpleDOMNode then we get accessible from the cache and of course these accessible don't implement nsIAccessible.

So either
1) we should have something similar to RefreshNodes() on show events
2) do not cache "pure" nsIAccessNode accessibles in nsIAccessNode interface navigation methods

Aaron, what do you think?

Comment 1

[alexander :surkov (:asurkov)]

Actually Aaron's patches from bug 434464 should address this bug. So if

1) we navigate the accessible tree by nsIAccessNode
2) some DOM node is shown
3) we continue a11y tree nagivation by nsIAccessNode
4) try to query nsIAccessible 

then we fail. I'm not sure is this scenario possible in real applications but we need some testing.

Marco, can you see any problems with chatzilla and jaws?

Comment 2

[Marco Zehe (:MarcoZ)]

Nope, when new test comes in, it is shown properly. This is even true for JAWS 10, only that the text is being run together. But it *appears* as soon as it domes in.

Comment 3

[alexander :surkov (:asurkov)]

So bug 434464 fixed chatzilla issue you watched previously?

Comment 4

[Marco Zehe (:MarcoZ)]

Well not exactly. Bug 434002 did, shortly before the release of Firefox 3.0. As far as I know, fixing bug 434464 did not have any sort of impact on JAWS performance with ChatZilla, either 9 or 10 public beta.

Comment 5

[alexander :surkov (:asurkov)]

Ok, since we haven't any known problems with nsIAccessNode so let's close this bug.

Comment 6

[alexander :surkov (:asurkov)]

reopen, since we need invalidate the cache even if we don't know any existing problems. Also Aaron suggested to use reorder event on parent accessible to notify AT clients if access nodes are changed.

Comment 7

[Aaron Leventhal]

We need to fix this one for JAWS 10 support.

Comment 8

[alexander :surkov (:asurkov)]

accessnode aren't cached anymore, nothing to invalidate, close as worksforme