[bug report] 275.19 makes Gtk apps go into infinite loop when quitting, breaks Gnome

[nE0sIghT]

280.13 doesn't fix this

[rapsure]

I tested 280.13 as well and the bug is still present.

[tetromino]

OP here. Bug is still present in 275.21 and 280.13.

Some additional things I've discovered:
* if an affected application is run from strace, gdb, or valgrind, the bug will not appear;
* if an affected application is run with __GL_NO_DSO_FINALIZER=1, the bug will not appear (but it's definitely not a long-term solution because, if I understand it right, it will lead to memory leaks in opengl applications);
* and setting __GL_SINGLE_THREADED=1 has no effect (the bug still appears).

Also, in addition to gnome-settings-daemon and gnome-session, every single terminal emulator that uses the vte library appears to be affected (gnome-terminal, xfce terminal, etc.), no matter whether they are gtk2-based or gtk3-based.

Some other people have apparently been able to work around the problem by building cairo with --disable-gl (USE="-opengl" in Gentoo parlance). However, for me that has no effect; on my machine, for both cairo-1.10.2 and the latest cairo git master HEAD, the affected gtk+ applications go into an infinite loop on exit no matter whether cairo was built with --disable-gl or --enable-gl.

[tetromino]

OP here. Bug is still present in 285.03 beta...

[cthulhu]

While I truely appreciate the hard work the good (but unfortunately few) Linux folks at NVIDIA put into this driver, this bug is another good example why open sourcing the driver would be an excellent idea. This bug was found over a month ago, but still plagues a lot of users. I'm confident that had the source been opened up to the community, this bug would've been fixed long ago. If more people had had a chance to look into the source this bug might've been found before causing problems like this one.
A rather large percentage of all Linux users are also NVIDIA users, some of thses are highly skilled developers. Just look at how far the talented people working on the nouveau driver has gotten without any help from NVIDIA. Imagine how far they would've gotten, had they received in the form of documentation from NVIDIA.
To this day the binary blob still lacks features like Kernel Mode Setting and proper randr support, while most open source drivers support both.

Sorry for the long rant, but this bug just reminds me of the benefits of open source.

EDIT: Clarification.

[Krellan]

I can confirm this bug happens as well. On Gentoo, I'm now running:

nvidia-drivers-275.21
xorg-server-1.10.4

Josh

[Comrade DOS]

It has been almost two months from the time the bug was found and it was reported. When he finally will be fixed?
At this point I can confirm the bug in all versions of drivers from 275.21 and 285.03 ending.
And regardless of the version of the x-server.

[ppurka]

Originally Posted by cthulhu

While I truely appreciate the hard work the good (but unfortunately few) Linux folks at NVIDIA put into this driver, this bug is another good example why open sourcing the driver would be an excellent idea.
...
Sorry for the long rant, but this bug just reminds me of the benefits of open source.

EDIT: Clarification.

That bugs reported to open source projects get fixed quickly is a myth. Yes, some of them do get fixed quickly, but there are many that go on unfixed for years.

[cthulhu]

Originally Posted by ppurka

That bugs reported to open source projects get fixed quickly is a myth. Yes, some of them do get fixed quickly, but there are many that go on unfixed for years.

While I tend to agree, you're missing a few key points. The NVIDIA blob has so many users, that the possible developer base is so large, that it would be highly unlikely, that bugs like these would live for very long.

Other OSS projects with smaller user bases often have long outstanding bugs, unless they have really dedicated maintainers.

[ppurka]

Originally Posted by cthulhu

While I tend to agree, you're missing a few key points. The NVIDIA blob has so many users, that the possible developer base is so large, that it would be highly unlikely, that bugs like these would live for very long.

Other OSS projects with smaller user bases often have long outstanding bugs, unless they have really dedicated maintainers.

Yes, smaller projects might lag behind, but even big projects like KDE and Xorg also have multiple-year long bugs. Even Firefox. You only have to do a search to find out these longstanding bugs.

I don't want to hijack the thread, so I will stop here.

[tetromino]

Looks like this bug has finally been fixed in driver version 285.05.09.

[nE0sIghT]

Not for gdm 3.2.0