This bug makes firefox crash. Steps to reproduce: 1. Launch orca and firefox. 2. Open firefox Preferences->Main tab, press Tab to browse the settings in this tab. Expected result: User could browse the settings in current tab. Actual result: After pressing Tab twice, firefox core dumped. Additional info: This bug can be found with Mozilla/5.0 (X11; U; SunOS i86pc; en-US; rv:1.9a1) Gecko/20060911 Minefield/3.0a1 and orca1.0.0 on gnome2.14 (solaris nevada build 46). It will not be reproduced if the orca is not running.

In nsXULSelectableAccessible::RefSelection accService->GetAccessibleInWeakShell(tempDOMNode, mWeakShell, getter_AddRefs(tempAccess)); fails ************************************************************ * Call to xpconnect wrapped JSObject produced this error: * [Exception... "'Permission denied to get property XULElement.ownerDocument' when calling method: [nsIDOMXULSelectControlItemElement::ownerDocument]" nsresult: "0x8057001e (NS_ERROR_XPC_JS_THREW_STRING)" location: "<unknown>" data: no] ************************************************************

[14:38] bz ginn: oh. [14:38] bz ginn: I guess that always happens... [14:38] bz ginn: if you make a call into a JS-implemented component while no JS is on the stack [14:38] bz ginn: you will end up with that component running on the safe context [14:39] bz ginn: if you know you're calling into a JS-implemented thing, you may want to push the "right" JSContext yourself [14:39] ginn how to push? [14:40] biesi why does this call JS code? [14:40] bz ginn: see http://lxr.mozilla.org/seamonkey/source/content/base/src/nsContentUtils.cpp#2394 [14:40] bz ginn: for what you want to do [14:40] bz biesi: it's calling an XBL-implemented method [14:40] bz biesi: the XBL implements an accessibility interface, see [14:41] bz ginn: ideally xpconnect would do this for you automagically; if pushing the cx helps, file a bug on that and cc me, jst, shaver, mrbkap? Yes, pushing helps.

Aaron, do you think we should check-in this patch to avoid crash first? Tim could not test Firefox with Orca if it's too easy to crash.

er... this bug was about an accessibility issue, right? We should just have a separate bug on the xpconnect issue, one not confused by extraneous patches...

Talking to Mark Pilgrim on IRC, he says that the problem is that the "new-ATK" is exposing things that it shouldn't. In this case, it's returning high-bit unicode characters when "TEXT::getText()" is called to get the input's text. He said that the location bar had the same issue, but now it's fixed. So maybe this isn't the right patch?

I /think/ that this might be fixed by bug 352195, but I'm new to this stuff. (Where you're tabbing to is, after all, a autocomplete textbox).

I don't think so. This bug talks about nsIDOMXULSelectControlItemElement, I don't think it's autocomplete. This bug will be gone by fixing Bug 352791. But before that, i want to avoid null pointer to make the testing smooth.

Commited, still need to fix Bug 352791 to get nsXULSelectableAccessible::RefSelection work.