Very Computer

Hi,

I hope this is the right forum to inform about in bug in glibc-2.1.1.
Otherwise please tell me where to go. (I don't know the adress of a
glibc mailing list).

We just ported our very large ERP system to linux. All went well but
the programs were very slow.
After 3 days of bug hunting I observed that our debugging package uses
sleep(0) very often. In all OSes we already used this makes no
problem. But glibc-2.1.1 uses nanosleep() to implement sleep(). A call
of nanosleep to sleep for 0 seconds results in a sleep of at least
10ms (as also described in the manual page).

We think this is a bug. sleep() should check for a argument of 0 and
simply return in this case.

glibc-2.0.7 doesn't have this problem, because this version does not
use nanosleep() to imlement sleep().

This bug is currently no problem for us any longer. We simply check
ourselves for a 0 argument to sleep. But it might happen that other
people also run into this same problem.

Peer.

        No. It's handly to be able to use 'sleep(0)' to get the minimum
possible sleep. It's a good way to implement something similar to
spinlocks or to yield the processor.

Quote:> This bug is currently no problem for us any longer. We simply check
> ourselves for a 0 argument to sleep. But it might happen that other
> people also run into this same problem.

        When your program calls 'sleep(0)', what is it trying to do? What do
you think 'sleep(1)' should do?

        DS

This is what sched_yield() is for.

--
`I'm glad you want me to be happy. I'd like you to be happy too.'
            --- Nat Lanza, while flaming Larry McVoy on linux-kernel

        No, that is not what sched_yield is for. Read the pthreads standard.

        DS