Startup/exit race condition

[Michael Renner]

Hi!

I'm reporting this here since there seems to be no official/public form of
bugtracking (bugzilla or something the like):

There seems to be a race condition when starting (a detached) screen from
within a screen which is about to exit, causing the newly started screen
to terminate.

This was verified with a simple three-liner shellscript:

---

#!/bin/bash
sleep 5
screen -d -m $0

---

after a while (between 2 and 30 loops) of happy screen-respawning the
about-to-start screen fails. I haven't found a good way to debug this,
since strace doesn't seem to be able to follow the exec'ed processes when
the parent process exits. Putting a "sleep 1" after the screen call
(giving the new process enough time to start up?) fixes this problem.

This was tested on an "old" (Athlon 1Ghz) computer with a fairly loaded fs
(software raid5, vmstat showing a constant read of 1-3mb/sec), so if this
bug is related to the creation/deletion of socket files it could be hard
to reproduce on other computers. Version used is the one which is
currently in debian-tesing, reporting itself as "4.00.02 (FAU) 5-Dec-03",
OS/kernel is linux 2.6.4.

best regards,
michael