Bug 25895 – [Gtk] Segfault when deleting the last/only character in an entry if a11y is enabled

• Status: VERIFIED
• Resolution: FIXED
• Component: Accessibility

[Joanmarie Diggs]

Steps to reproduce:

1. Launch GtkLauncher

2. In the Google search entry, type a character and then press Backspace.

Expected results: GtkLauncher would not segfault.

Actual results: GtkLauncher segfaults with this error:

~~~~
ASSERTION FAILED: anchorType() == PositionIsOffsetInAnchor
(./WebCore/dom/Position.h:85 int WebCore::Position::offsetInContainerNode() const)
~~~~

This only seems to occur when a11y is enabled.

I'm using rev. 43932 from svn trunk.

[Xan Lopez]

I don't seem to be able to reproduce this now, can you?

[Joanmarie Diggs]

(In reply to comment #1)
> I don't seem to be able to reproduce this now, can you?
> 
I'm afraid so. I just built the latest WebKit from trunk and the latest Epiphany from master.

* If a11y is enabled (log out and back in if it wasn't already enabled), all I have to do is navigate to Google type a letter in the search entry, and press Backspace. This is true both in GtkLauncher and Epiphany.

* If a11y is not enabled, the problem does not occur.

* This is true both with and without the patches for get_text_at_offset.

[Xan Lopez]

Weird, I have a11y enabled (everything works OK under accerciser), but I get no crash. I don't have *session* a11y enabled, but I don't think that can make any difference? Could you paste a full backtrace?

[Xan Lopez]

Well, since you seem to have session a11y and I don't I guess it does make a difference, heh. Wonder how that works.

[Joanmarie Diggs]

(In reply to comment #4)
> Well, since you seem to have session a11y and I don't I guess it does make a
> difference, heh. Wonder how that works.
> 
Heh. Mid-air collision. :-) Does that mean you can repro it now? If not, I'll get a trace in a little while.

[Xan Lopez]

(In reply to comment #5)
> (In reply to comment #4)
> > Well, since you seem to have session a11y and I don't I guess it does make a
> > difference, heh. Wonder how that works.
> > 
> Heh. Mid-air collision. :-) Does that mean you can repro it now? If not, I'll
> get a trace in a little while.
> 

No, haven't tried with session a11y, but I believe that it crashes :)

If you can provide a trace that would be useful yeah, thanks! /me goes to bed now

[Joanmarie Diggs]

Created attachment 31110 [details]
backtrace

Here you go. Thanks!

[Xan Lopez]

For the record, I wasn't seeing this because it's crashing in an ASSERT, so it only happens in debug builds. It seems to work fine in release builds. And yes, you need a11y, but not session-wide, just for the app will do.

[Xan Lopez]

Created attachment 31289 [details]
offsetincontainer.patch

Fix for the crasher.

[Holger Freyther]

Comment on attachment 31289 [details]
offsetincontainer.patch

looks sensible.

[Xan Lopez]

Landed as r44683, closing the bug.

[Joanmarie Diggs]

Verifying. Thanks!