Race condition at session startup sometimes prevents applications from being accessible

I can confirm this. I have two machines, both running a fully current Lucid system as of the time of this comment. One is an Athlon64 3400+ with 1 gb of ram, the other an Asus Eee 1005PE also with 1 gb. This problem happens on both machines, though it seems much more likely to happen on the netbook. Typically I can kill X and log back in to take care of it, but other times I must do this more than once. This did not happen at all in Karmic nor any earlier version of Ubuntu, and is certainly a regression.
If any type of debug output would be useful, I'll be happy to send whatever is needed.

Ok, could you please do the following steps, so we can attempt to gather some more information about the system that is having the problem in question.

1. Add my personal PPA to your sources.list http://launchpad.net/~themuso/+archive
2. Update your system. This will pull in a slightly modified version of at-spi, for testing purposes.
3. Install the following packages: bootchart pybootchartgui
4. Enable automatic login. This can be found in System, Administration, Login Screen. This dialog is completely accessible, even after you unlock it so its fine to use.
5. Reboot your system. You will automatically be logged in.
6. Check in /var/log/bootchart. You should have a png file there, named with your computer's hostname and the date. Please get this file and attach it to this bug. Note that this file won't necessarily be present immediately after you are up and running at the desktop.

One other thing to try, is killing gnome-settings-daemon once you are logged in, and restarting it, and checking to see whether apps become accessible again. To do this, go to a terminal and type: killall gnome-settings-daemon. Once you do this, immediately press Alt + F2 to bring up the run dialog box, and run gnome-settings-daemon. Once done, please check apps that were not accessible before, i.e panel, desktop, etc, and let us know whether this has changed.

I am subscribing Martin Pitt to the bug who will hopefully be able to help us debug this problem further.

Thanks for your cooperation in fixing this.

 subscribe pitti

For the record, I checked the at-spi code, and it seems fine to me. One thing that we could try is to add an XSync (display, False) right after the XChangeProperty() call, just in case the property otherwise lands too late. Otherwise we'd need to add some logging to GTK to check whether panel/nautilus/etc. actually see the root property when they start up. This should tell us whether the race condidtion is in the property, or trying to talk to at-spi-registryd which isn't fully operational yet.

This affects me as well, in ways similar to those described above.

I'll try using the custom PPA later today, as I can't afford to reboot now, but killing and restarting gnome-settings-daemon didn't work for me. Should I have had the new at-spi packages installed when I did that?

Nolan Darilek [2010-04-14 14:23 -0000]:
> but killing and restarting gnome-settings-daemon didn't work for
> me. Should I have had the new at-spi packages installed when I did that?

No, restarting g-s-d was an independent test.

OK, another question. The login settings dialog may be completely accessible, but your instructions include a series of menus to traverse, which at this time are not accessible. What is the command line name for this tool?

OK, answered my own question, gdmadmin.

Interestingly enough, now my desktop appears accessible while the panel, run dialog, etc. are still not. Before, the desktop was also inaccessible.

Nolan Darilek [2010-04-14 15:49 -0000]:
> OK, answered my own question, gdmadmin.

"gdmsetup" actually. :-)

Martin
--
Martin Pitt | http://www.piware.de
Ubuntu Developer (www.ubuntu.com) | Debian Developer (www.debian.org)

One thing I have discovered, and can reproduce 100% of the time, is a race condition, if it can be called that, for gdm with accessible login enabled. Gnome-orca loads and speaks, however I do not get any accessibility feedback from the gdm greeter, or the bottom panel with the various options present there. I say this probably can't be called a race because this occurs 100% of the time for me.

My machine is not letting me reproduce the race any more at the desktop, even when I added the suggested XSync function call to the at-spi code. I have uploaded this modified package to http://launchpad.net/~themuso/+archive for others to test. The modified package adds teh XSync call that Martin suggested.

Please test and report back.

I have also attached a boot chart for at-spi/orca in gdm with accessible login enabled.

Or maybe not. The PPA queue is 3 days long, so please hold off from upgrading for the moment. I'll report back when the packages are built and ready for testing.

Hello Luke,

Luke Yelavich [2010-04-15 3:02 -0000]:
> My machine is not letting me reproduce the race any more at the desktop,
> even when I added the suggested XSync function call to the at-spi code.

Hm, I'm confused about the "even"; can you reproduce the race without
the XSync(), and not any more with it? That would be a great result
indeed.

> The PPA queue is 3 days long

Build score bumped, should be there in ~ 1 hour now.

Seems to now work for me as well. I just tried everything that formerly reported as inaccessible, and everything seems to work just fine. Then again, I haven't tried multiple reboots to rule out simple luck, but considering something always reported as inaccessible before, this would seem to be a good sign.

So, do we still need to run the version from your PPA, or is this something that just magically resolved itself? :)

Hi i am having to killall gnome-panel after startup to get gnome-panel to show up in pyatspi/accerciser.

pyatspi version: 1.30.0-0ubuntu2
gnome-panel version: 1:2.30.0-0ubuntu1

Wow...this is really confusing...

I have two similar desktops at home.

1º Desktop specs (Desktop "A"):
    Ubuntu 10.04 LTS
    ASUS P5QC
    Intel Core2Quad Q8200
    2x1GB SuperTalent W1333UA1G8 memory module (2GB Ram Total)
    Sapphire ATi HD4890 1GB DDR5
    Driver Packaging Version 8.723.1-100408a-098580C-ATI
    2D Driver Version 8.72.11

2º Desktop specs (Desktop "B"):
    Ubuntu 10.04 LTS
    ASUS P5QC
    Intel Core2Quad Q8400
    2x2GB OCZ OCZ3G13332G memory module (4GB Ram Total)
    Sapphire ATi HD4890 1GB DDR5
    Driver Packaging Version 8.723.1-100408a-098580C-ATI
    2D Driver Version 8.72.11

Both machines installed from scratch, so both root partitions where formatted at install time. Both with same ATi propietary driver.

The funny thing is that "Desktop A" boots up as expected, really short delay between boot menu and usable GNOME Desktop. In "Desktop B" boot thime is unnacceptable (60 secs at least), Skype and Empathy login sounds being played, but only mouse pointer with blask background is displayed!

Please also take a look at my previous screenshot, because sometimes is unable to load the default GNOME theme.

Please advise...this is really weird!

Sorry...ignore my last post

I installed with my system the 2010.04-27 daily live CD, and install
updates with all day.
Some time if I using longer time the Orca screen reader and for example
browsing the internet, and try changing for example top/bottom panel
with Ctrl+Alt+Tab key combination, Orca is does'nt spokening the actual
choosed panel before I not release this combination. Another issue if I
try change task with Alt+Tab key combination, I not hear application
name with I jump before I release this key combination. This problem is
related this bug? If I restart Orca Screen Reader, again working fine
this two task. I see this type problem after 20-30 minute use, and this
problem repeat again.

Now, all updates are installed. gnome-orca package version is
2.30.0-0ubuntu3, and at-spi package version is 1.30.0-0ubuntu2.

Anybody see this type problem? Possible help if I reinstall my system
with final Ubuntu CD iso?

Thank you the answers,

Attila

Is this also the cause of Ubuntu 10.10 freezing when using orca? Have we been able to resolve this issue in Ubuntu 10.04?

Are there any updates to this bug?
I would be interested if this bug is fixed in newer Ubuntu versions?

Girish, why have you marked this "confirmed"? Thomas requested more information here. Are you still seeing this bug? If so, please explain which versions of Ubuntu you see it on.

Luke, are you still experiencing this?
If so, are you still working on a fix? (You are assigned to this).

The at-spi package was dropped from the CDs for precise, and removed from the archive in quantal. So if this issue still exists, it's not going to be fixed in the at-spi package now.