GNOME Bugzilla – Bug 363697
Gdm 2.17.0 just crashes at startup
Last modified: 2006-10-30 18:54:12 UTC
Steps to reproduce: 1. Start GDM 2.17.1 Stack trace: Other information: When it comes up, it just says "The greeter application crashed. Try a different one". I've searched /var/log, I can't find any useful information there
I suspect that the problem is caused by the new at-spi launching stuff. I'm also seeing it crash. I'm cc:ing Ariel who did that fix to see if he can look into this. I also see this crashing on Solaris. What OS are you using? Look in the following directories for core files: - /var/lib/log/gdm - /var/lib/gdm - the root directory - the gdm user $HOME directory A stack trace would be handy. If you have another login program to start the Xserver you should be able to set DOING_GDM_DEVELOPMENT=1 and run /usr/lib/gdmgreeter from the command line and perhaps get a stack trace that way.
Note bug #345434 for more information.
Does this happen with a11y turned off? ariel
This only happens when a11y is disabled due to the fact that gtk_init() is called after the call to initialize at-spi. If a11y is disabled, then error_dialog() in gdmcommon.c is called. This function makes GTK+ calls which cause the crash since gtk_init() was not called first. However, according to the comments in gdmlogin.c, gtk_init() cannot be called before at-spi is initialized since that would crash the bridge. I guess one thought is to pass the argc and argv info on through so that gtk_init() can be called before the error dialog is displayed. Then, the at-spi initialization function can return a boolean as to whether or not it succeeded. If it did not, then gtk_init() would not need to be called again. Or just call it again since that should not cause a problem. Alternatively, don't show the error, or only show it via syslog or on the console. Or do something else entriely :-).
Stack trace:
+ Trace 78075
Ok I have a patch but I will do more testing before submitting it. Can someone assign this to me and make bug #345434 depend on this one. thanks
I think the error message should just go to the syslog. We shouldn't be crashing because we can't display a GUI. Do we really need a pop-up dialog here?
A patch will be posted to bug #345434 I have eliminated the popup and write to syslog instead more info on that bug.
Patch posted. I checked it on my setup but Alex can you test it please? ariel
In comment #2 Brian Cameron wrote: > I suspect that the problem is caused by the new at-spi launching stuff. Hm, from some time I am collecting core files from at-spi-registryd. There are some samples here: https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=206323 Now I am starting wonder if this is not related. These bombs seem to be correlated with a gdm process terminating/restarting.
If at-spi-registryd is crashing, then this should probably be moved to the at-spi category instead of GDM. Would you move the category so this can be looked into by the at-spi experts. Note that you might need to use the updated version of at-spi since the handshaking between GDM and at-spi is changing in 2.18.
> If at-spi-registryd is crashing ... Well, with the current "rawhide" updates just recently it stopped crashing; or at least looks that way. :-) I am still waiting what will happen on a longer run. In the past there were some moments when at-spi-registryd was crashing, and later not crashing without any changes in that, and later crashing again. I thought that I better mention my observations but I am not really sure what is really going on.
Well, I will close this bug. Seems the GDM issue is fixed. If at-spi-registryd is crashing, then I'm sure that someone will report the bug to that category. Note that older versions of at-spi-registryd have crashing problems. On Solaris, this was related to the fact that it would try to start up the festival driver even though festival wasn't installed on the system. You might try upgrading at-spi-registryd if it crashing becomes a problem for you. If you aren't making use of accessibility, then it might not be an issue at all for your use.
You should really ask Matthias Clasen how he fixed it in rawhide. If he patched something gdm is *not* fixed @gnome * Fri Oct 27 2006 Matthias Clasen <mclasen redhat com> - 1:2.17.0-2 - Fix a crash with launching a11y support * Mon Oct 23 2006 Matthias Clasen <mclasen redhat com> - 1:2.17.0-1 - Update to 2.17.0
For what it's worth, I "fixed" this in FreeBSD by commenting out the contents of the error_dialog() function in anticipation of a better fix upstream once the details of the bug were reported. So the syslog fix would be a step up.