Hello,
Jeff Johnson wrote:
> On 3/1/07, Frank Fegert <fra.nospam.nk gmx.de> wrote:
>> Frank Cusack wrote:
>> > As seen on Solaris with multiple CPUs,
scriptlets get unlinked before
>> > they can be exec'd. This is because of
illegal use of a mutex
>> <whistle>,
>> > 10 yard penalty!
>> >
>> > [...]
>> >
>> > My patch simply disables the use of the mutex
altogether. I don't see
>> > much advantage to it anyway ...
>>
>> just for the record: This also solves the problem
of *never* finishing
>> *any* scriptlet on AIX [1] (except
"--noscripts" and being truss'ed,
>> that is ). So
i'd too be interested in a proper fix, or at least
>> being able to disable at configure time.
>
> This is the first rport of a problem.
no offense, but that's not entirely true, see:
https://lists.dulug.duke.edu/pipermail/rpm
-devel/2006-July/001274.html
and following. Although the problem is now reproducable
even
when rpm is truss'ed.
> Details will get you a fix. What is needed is
strace/truss (or
> whatever AIX) uses so that SIGCHLD delivery can be pin
pointed.
Please see the attached truss output. The scriptlet contains
only:
# cat /opt/afw/var/tmp/rpm-tmp.13059
/opt/afw/bin/install-info /opt/afw/share/info/grep.info
--dir=/opt/afw/share/info/dir;
and is still sitting in $VAR. Building rpm with Frank
Cusaks
patch (rpm-4.4.8-scriptlet-mutex.patch) seems to resolve
the
issue. But i'm not able to tell what else might break by
only
disabling threading with 'int nothreads = 1;'.
> And -- no matter what -- there is a pathway through the
existing
> rpmio/rpmsq.c that simply uses waitpid(3), which
presumably
> works as well on AIX as on any other uglix platform.
>
> See rpm-devel archives for the waitpid patch, I've
posted at least twice.
I tried this (see thread mentioned above), but problem
remained
the same.
Thanks & Regards,
Frank
_______________________________________________
Rpm-devel mailing list
Rpm-devel lists.dulug.duke.edu
https://lists.dulug.duke.edu/mailman/listinfo/rpm-devel
a>
|