Boost Testing :
From: Christopher Cambly (ccambly_at_[hidden])
Date: 2007-09-24 12:44:42
> On Sep 20, 2007, at 1:09 PM, K. Noel Belcourt wrote:
> > I just checked in the mods to execunix.c and output.c. I also
> > touched option_l.jam because once I added the time out diagnostic,
> > test.sh began reporting errors.
> > Tested on Sun, Ibm, Sgi and Linux variants.
> I just committed a patch to trunk that fixes a bug in the -lx unix
> timeout code. If you run Boost trunk regression tests on a unix
> system, the timeout code may not have killed all processes when an
> action timed out. If you run ps, you may see some Boost regression
> test processes (such as compiles) that are owned by init (process 1)
> or that are zombies. These would be the processes that bjam failed
> to terminate correctly and they should be killed.
> Details if you're interested.
> If, for example, you had a g++ compile that timed out, bjam would
> kill the g++ process but it may not have terminated processes spawned
> by g++, such as cc1plus. The patch I just committed to trunk fixes
> this by making the bjam forked action (such as g++) a session leader
> so that when g++ times out and is killed, any processes started by g+
> + are also killed.
> -- Noel
I noticed this late last week. Our nightly runs were still not completing.
I could see that the parent process was killed, however the subprocess were
still around. The runs completed as soon as I killed the subprocesses.
Since we have results published, I believe everything is working correctly
for us now.
IBM XL C/C++ Compiler Development