Re: [gdm-list] at-spi-registryd not starting automatically in gdm [was multiple at-spi...]

[Brian Cameron <Brian Cameron Sun COM>]

Francesco:

> Brian Cameron wrote:
> > I just fixed this bug for the GDM 2.20.0 release.  Now you can specify
> > the location of the at-spi-registryd via the --with-at-spidir configure
> > argument.  Your distro will still need to use this flag when configuring
> > GDM to fix the problem.  I assume you can follow up with them on this?
>  
> Yesterday, the Ubuntu 7.10 Distribution offered an updated versions of 
> GDM. The packager took (among others) the new configuration option into 
> account and at-spi-registryd gets automatically started as it should.
>
> Thanks again for the fix.

Great!  I'm glad to hear that a11y with GDM is finally working better in
Ubuntu land.

> Now, I stumbled on an odd behaviour of at-spi-registryd during GDM. 
> Maybe it is normal and I do not understand what is going on. Here is 
> what happens:
>
> When I remotely (ssh) change a value with an editor in the gesture 
> configuration file of gdm, I enter "sudo /etc/init.d/gdm restart" to 
> make it recognise the change; and an instance (*) of at-spi-registryd is 
> automatically started under the gdm user. (this is probably not the best 
> way to make it recognise the change, but it is not the topic now)
>
> Then, when I connect remotely to gdmlogin, the applications that I try 
> to launch with a gesture tell me that at-spi-registryd is not running.
>
> So I disconnect from the gdm session and afterwards I manually kill the 
> at-spi-registryd instance (*) that was created by the restart of gdm.
>
> I go on and connect remotely to gdm, and a new instance of 
> at-spi-registryd is started which is then recognised by the applications 
> started by a gesture.

I could imagine there could be two problems here:

1) GDM probably needs to be smarter about making sure that the
   at-spi-registryd gets killed when the GUI is killed.  I suppose we
   could hack gdm-restart to look for and kill the process, but this
   seems wrong.  I'd think at-spi-registryd should go down when its
   parent process dies.  I'd recommend filing a bug against
   at-spi-registryd about this issue.

2) You might have some issues if multiple users are wanting to use
   accessibility at the same time.  I believe at-spi-registryd is
   supposed to be smart enough to run different instances as the
   same user if they are on different displays.  If you find that
   at-spi-registryd doesn't work in remote logins, or if you find
   that it fails to work when multiple users try to use a11y at
   the same time, I'd think this is the problem.  Again, I'd say
   this is a bug for at-spi-registryd.

> Please, do not get me wrong: at-spi-registryd is recognised without 
> problems if it was started by the current gdm session; however if it was 
> running before opening the gdm session, it is not found, although it is 
> running under the gdm user.

This makes sense.  The at-spi-registryd should die when its parent
process dies.  If this is not happening, then this should be fixed in
the at-spi-registryd code.  If GDM needs to do something special to kill
any running daemons, then the at-spi-registryd maintainer should let us
know what we need to do.

> Is there any explanation for this behaviour, or is there something wrong 
> going on?

I think you need to take up these issues with the at-spi-registryd
maintainers.

Brian