screen freezes up, turns zombie

Could it possibly be the same bug, or at least related to https://savannah.gnu.org/bugs/?26742 ?

The fact that the backend crashes strongly suggests to me that it's not screen-profiles's/byobu's fault (though it may have generated the config that triggers the bug in screen). Even if screen-profiles's /usr/bin/screen wrapper does something a little screwy with spawning the "real" screen, that should not affect the forked background process.

Marking invalid against byobu. I have never been able to reproduce this.

:-Dustin

I had only been hit occasionally by this bug over the last several months, and I've just been ignoring it. But, by chance or by special circumstances, I've triggered the bug three times today... before lunch. So it's really getting on my nerves.

I do think that the bug is in screen (not byobu), and it seems likely that it's the same as https://savannah.gnu.org/bugs/?26742, as Micah suggested. I just looked it up, and that bug has a patch which is believed to work (and will be included in screen 4.1). Since this bug can potentially result in data-loss, and there are a lot of people affected by it, is there any chance we could get a new package built using the patch, either in karmic-updates or at least in a PPA? (If this keeps annoying me, and someone else doesn't have time/inclination to post a package, my next weekend project might need to be learning how to build deb packages and create a PPA.)

Thanks!

I haven't upgraded to maverick yet, but I did recently switch to use the "awesome" window manager, and that has dramatically increased the frequency that I trigger this bug (triggering the bug seems to be a matter of switching back and forth between a high and low column count while displaying vim or a similar program). I noticed several things:

http://savannah.gnu.org/bugs/?26742 was fixed in September of 2009, but a major release hasn't been made since then and the patches don't seem to have ever made their way into a minor release.

From another LP bug report, in Maverick, matching my details almost exactly (https://bugs.launchpad.net/ubuntu/+source/screen/+bug/656396) I gather that this patch hasn't yet made it into Debian or Ubuntu. I suspect that https://bugs.launchpad.net/ubuntu/+source/screen/+bug/369184 is also related.

From my recent /var/log/kern.log*:

Oct 6 11:32:50 arrakis kernel: [22213.433002] screen[5598]: segfault at 18 ip 00007f3cb06481c3 sp 00007fffad6107c0 error 4 in libc-2.11.1.so[7f3cb05ce000+17a000]
Oct 6 21:23:59 arrakis kernel: [57682.092950] screen[782]: segfault at 18 ip 00007ff2fff3f1c3 sp 00007fff16ffd460 error 4 in libc-2.11.1.so[7ff2ffec5000+17a000]
Oct 7 11:20:36 arrakis kernel: [103639.433370] screen[3829]: segfault at 18 ip 00007f73d794b1c3 sp 00007fff267f6bd0 error 4 in libc-2.11.1.so[7f73d78d1000+17a000]
Oct 7 15:21:35 arrakis kernel: [118098.062706] screen[30186]: segfault at 18 ip 00007faf9e4c21c3 sp 00007fff079d5830 error 4 in libc-2.11.1.so[7faf9e448000+17a000]
Oct 11 13:19:09 arrakis kernel: [284508.558063] screen[9936]: segfault at 18 ip 00007f5f72ba01c3 sp 00007fff0ed43750 error 4 in libc-2.11.1.so[7f5f72b26000+17a000]
Oct 11 14:00:50 arrakis kernel: [287009.793219] screen[4099]: segfault at 18 ip 00007fa5ee7da1c3 sp 00007fff058cc710 error 4 in libc-2.11.1.so[7fa5ee760000+17a000]

On second look, I don't think this bug necessarily shares anything with #656395.

Thanks for pointing to the upstream bug. I'll try to cherry pick that patch and get a test package applied. I, myself, haven't been able to reproduce this yet. If I build a test package for you, can you try it out and confirm/deny the fix?

I can reproduce the bug very reliably:

* open up a terminal window (unmaximized),
* start screen,
* start vim,
* run :grep (or something similar) and verify that it works
* maximize terminal window,
* drop the font size a couple of times,
* rerun the exact same :grep (etc)

and now the entire screen session has crashed.

This never fails to reproduce the bug for me.

I would be very happy to try out out a test package for you.

See also bug #727741 - this possibly-buggy version of screen may be left behind when screen-profiles is removed.

I can also reliably reproduce this crash following the steps Nicholas posted, as long as the resulting terminal is over 255 columns. Seems to be related to the fix already posted.

This may be a root cause of: https://bugs.launchpad.net/byobu/+bug/644825

I do not have the issue described in bug #727741.

I would also be happy to try out a test package.

This bug was fixed in the package screen - 4.0.3-14ubuntu8

---------------
screen (4.0.3-14ubuntu8) oneiric; urgency=low

  * Cherry-pick two upstream crash fixes (LP: #375625).
    - 09CHERRY_7cb17d54.dpatch
    - 09CHERRY_45d0c395.dpatch
 -- Kees Cook <email address hidden> Mon, 06 Jun 2011 10:26:45 -0700

Any chance of getting this backported to 10.04?

Backported packages are available in my PPA: https://launchpad.net/~malcscott/+archive/screen

No warranty, etc.; I've tested the natty backport and it seems to work for me (although since the crash is infrequent here I can't yet 100% confirm that the bug is fixed). Backports for other versions will be available in a few hours and are not tested. Use at your own risk.