[Bioperl-l] can people with HP/Irix/Alpha confirm these reported problems with bioperl-1.0 ?

Mark Dalphin mdalphin@amgen.com
Wed, 24 Apr 2002 16:07:30 -0700


I can report a fix for Irix (IRIX64 6.5).

I consulted with one of our SysAdmins, Richard Stokotelny, and he found
a kernel tuning parameter which alters the ARG_MAX limit.

This change has permitted me to install BioPerl 1.0 without any
problems. (If a number bigger than 64k is entered, Irix has a kernel
panic during the re-boot. I used 50000.)

Richard says to do the following:
=====================================================
as root on your SGI, 
# systune -i
systune-> ncargs <a number LESS than 65535>
<it asks if you are sure> 
systune-> quit

autoconfig -v
/etc/reboot

=======================================================

> > This is a troubling bug report. Can others confirm? I've had zero
> > trouble with OpenBSD and Linux but don't have easy access to Irix, HPUX
> > or Alpha systems. We may have gotten bitten by the huge size of our
> > distribution.
> 
> Serge seems to know what he is talking about. In the past I've had the same
> problem in IRIX but I never bothered to find out what exactly was wrong.
> Thanks Serge!
> 
> Confirmed on IRIX:
> 
> cetus /tmp/bioperl-1.0> make
> make: execvp: /bin/sh: Arg list too long
> make: *** [pm_to_blib] Error 127
> cetus /tmp/bioperl-1.0>
> 
> I got curious and ran Serges program on various platforms:
> 
> IRIX64:  ARG_MAX =   20480
> SUNOS:   ARG_MAX = 1048320
> OSF1:    ARG_MAX =   38912
> LINUX:   ARG_MAX =  131072
> 
> I've no access to HPUX.
> 
> 
> Looks like we have to come up with a 1.01 quite soon.
> 
>         -Heikki
> 
> 
> > Actually, HP just donated a Itanium box to us so I'll have HPUX to work
> > on in a few days.
> > 
> > -Chris
> > 
> >   ----------------------------------------------------------------------------
> > 
> > Subject: [Bioperl-guts-l] Notification: incoming/1133
> > Date: Fri, 22 Mar 2002 13:31:25 -0500
> > From: bioperl-bugs@bioperl.org
> > To: bioperl-guts-l@bioperl.org
> > 
> > JitterBug notification
> > 
> > new message incoming/1133
> > 
> > Message summary for PR#1133
> >         From: batalov@gnf.org
> >         Subject: bioperl-1.0 doesn't build
> >         Date: Fri, 22 Mar 2002 13:31:24 -0500
> >         0 replies       0 followups
> > 
> > ====> ORIGINAL MESSAGE FOLLOWS <====
> > 
> > >From batalov@gnf.org Fri Mar 22 13:31:24 2002
> > Received: from localhost (localhost [127.0.0.1])
> >         by pw600a.bioperl.org (8.12.2/8.12.2) with ESMTP id g2MIVOkO005811
> >         for <bioperl-bugs@pw600a.bioperl.org>; Fri, 22 Mar 2002 13:31:24 -0500
> > Date: Fri, 22 Mar 2002 13:31:24 -0500
> > Message-Id: <200203221831.g2MIVOkO005811@pw600a.bioperl.org>
> > From: batalov@gnf.org
> > To: bioperl-bugs@bioperl.org
> > Subject: bioperl-1.0 doesn't build
> > 
> > Full_Name: Serge Batalov
> > Module: bioperl
> > Version: 1.0
> > PerlVer: 5.6.1
> > OS: IRIX,HP,DEC/Compaq-Alpha
> > Submission from: pat.gnf.org (63.196.132.64)
> > 
> > #> perl -MCPAN -e shell
> > #>     ...install Bundle::...
> > #> install B/BI/BIRNEY/bioperl-1.0.tar.gz
> > fails in 'make' with 'Arg list too long' message
> > 
> > What happens is that a list of things to MakeMake is too long for the
> > command line (in "# --- MakeMaker pm_to_blib section")
> > 
> > For the reference, command-line length limit is not
> > a shell limit; it is built in kernel/libc.
> > It has to do with the execvp()
> > limit ARG_MAX, which is 4096 characters in POSIX,
> > but most systems have it inflated, to different extents.
> > 
> > Test it with
> > --------arg_max.c--------
> > #include <unistd.h>
> > main() {
> >   printf("ARG_MAX = %ld\n  This is the maximum length of the arguments to the
> > exec() family of functions\n",
> >          sysconf(_SC_ARG_MAX));
> >   exit(0);
> > }
> > --------
> > 
> > Linux is the second most generous (most systems tested show 64K).
> > It is topped only by SUN (tried 5.8), where it is 128K.
> > IRIX and DEC/Alpha have 20K limit.
> > HP -- just 8K.
> > 
> > It would be prudent to stick to the POSIX limit, and
> > split the pm_to_blib into pieces.


-- 
Mark Dalphin                          email: mdalphin@amgen.com
Mail Stop: 29-2-A                     phone: +1-805-447-4951 (work)
One Amgen Center Drive                       +1-805-375-0680 (home)
Thousand Oaks, CA 91320                 fax: +1-805-499-9955 (work)