On Sun, Sep 25, 2005 at 08:41:42PM +0200, Maurice van der Pot wrote:
> Getting this every time:
> 
> ~> echo abc | c | c | c | c | wc
>       1       1       4

> If there's anything else I can try, let me know.

Would you try that with, say, 50 or more members in the pipeline?

If it does not ever break even with "huge" pipelines of several hundred
members, it might mean it is working (which would be a great news).

On another machine with 2.6.8 (Debian) I easily trigger problems with
a pipeline of about 250 cats while one of 50 works, most of the time.

> ~> uname -a
> Linux griffon27 2.6.12 #2 SMP Thu Jul 21 18:49:27 CEST 2005 i686
> Intel(R) Pentium(R) 4 CPU 3.00GHz GenuineIntel GNU/Linux

I am badly dependent on pipelines behaving well, as otherwise
configure scripts break or silently miss some dependencies.
All of my programs happen to run from Coda.

I tried twice to switch to 2.6 kernels, but had to back out
to a patched 2.4.

> model name  : Intel(R) Pentium(R) 4 CPU 3.00GHz
> stepping  : 9
> cpu MHz   : 2998.860
> cache size  : 512 KB

Looks like pretty similar to my setup, though SMP vs uniprocessor
may certainly influence timing and the races.

Anyway, if your tests are positive, I'll give 2.6.12 kernel (or later)
a try.

Thanks Maurice!
--
Ivan