From starksb@ebi.ac.uk

Greetings,

I install/maintain perl on Linux, Solaris, Tru64 and IRIX at our site.

It's rather well known that gcc on IRIX64 has problems passing small
structures properly. There is a good description of the problem at
<http​://groups.google.com/groups?oi=djq&selm=an_586981842>.

There is already a something in hints/irix_6.sh for this​:

# Check for which compiler we're using

case "$cc" in
...
*gcc*)
...
case "`uname -s`" in
# Without the -mabi=64 gcc in 64-bit IRIX has problems passing
# and returning small structures. This affects inet_*() and semctl().
# See http​://reality.sgi.com/ariel/freeware/gcc-2.8.1-notes.html
# for more information. Reported by Lionel Cons <lionel.cons@​cern.ch>.
IRIX64) ccflags="$ccflags -mabi=64"
ldflags="$ldflags -mabi=64 -L/usr/lib64"
lddlflags="$lddlflags -mabi=64"
;;
*) ccflags="$ccflags -DIRIX32_SEMUN_BROKEN_BY_GCC"
;;
esac
;;

Perhaps that works if you've built gcc with mabi=64 support. We have
not, so the above hint breaks our build. We can only apply the
hint -DIRIX32_SEMUN_BROKEN_BY_GCC. This presumably solves the problem
with semctl in perl.h​:

# ifdef USE_SEMCTL_SEMUN
# ifdef IRIX32_SEMUN_BROKEN_BY_GCC
union gccbug_semun {
int val;
struct semid_ds *buf;
unsigned short *array;
char __dummy[5];
};
# define semun gccbug_semun
# endif
# define Semctl(id, num, cmd, semun) semctl(id, num, cmd, semun)

But that leaves us with a broken inet_ntoa, reflected by failing tests
3 and 4 in lib/io_multihomed.t.

The fix is very simple, just "roll your own" inet_ntoa, as described
by Stuart Levy in the above-referenced comp.sys.sgi.admin article.
Here's the patch I've used​:

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

--- ext/Socket/Socket.xs.orig   Sun Apr  8 07:09:16 2001
+++ ext/Socket/Socket.xs        Thu Oct  4 14:06:12 2001
@@ -166,6 +166,18 @@
 
 #endif /* ! HAS_INET_ATON */
 
+#if defined(sgi) && defined(__GNUC__)
+char *inet_ntoa(struct in_addr sa)
+{
+  static char addr[20];
+  sprintf(addr, "%d.%d.%d.%d",
+         ((unsigned char *)&sa.s_addr)[0],
+         ((unsigned char *)&sa.s_addr)[1],
+         ((unsigned char *)&sa.s_addr)[2],
+         ((unsigned char *)&sa.s_addr)[3]);
+  return addr;
+}
+#endif
 
 static int
 not_here(char *s)

Could this be incorporated?

Please CC me on any followups, as I am not subscribed to any perl
mailing lists.

Thanks for your consideration.

Kind regards,
David

-------------------------------------------------------------------
  David Starks-Browning | starksb@​ebi.ac.uk
  EMBL Outstation -- |
  The European Bioinformatics Institute |
  Wellcome Trust Genome Campus | tel​: +44 (1223) 494 616
  Hinxton, Cambridge, CB10 1SD, UK | fax​: +44 (1223) 494 468
-------------------------------------------------------------------

Flags​:
  category=core
  severity=high

Site configuration information for perl v5.6.1​:

Configured by starksb at Thu Oct 4 14​:07​:08 BST 2001.

Summary of my perl5 (revision 5.0 version 6 subversion 1) configuration​:
  Platform​:
  osname=irix, osvers=6.5, archname=IP27-irix
  uname='irix64 talisker 6.5 10120104 ip27 '
  config_args='-de -Dprefix=/sw/common/test-patched -Dbin=/sw/arch/bin/test-patched -Dscriptdir=/sw/arch/bin/test-patched -Dprivlib=/sw/common/test-patched/share/perl/5.6.1 -Darchlib=/sw/common/test-patched/share/perl/5.6.1/mips-irix -Dsitebin=/sw/arch/bin/test-patched -Dsitescript=/sw/arch/bin/test-patched -Dsitelib=/sw/common/test-patched/share/perl/site-perl/5.6.1 -Dsitearch=/sw/common/test-patched/share/perl/site-perl/5.6.1/mips-irix -Dinstallusrbinperl=no -Dperlpath=/sw/arch/bin/test-patched/perl5.6.1 -Dstartperl=#!/sw/arch/bin/test-patched/perl5.6.1 -Dcc=gcc -Dld=gcc -Dcf_email=starksb@​ebi.ac.uk -Dloclibpth=/sw/arch/lib -Dlocincpth=/sw/common/include -Dldflags=-Wl,-rpath,/sw/arch/lib'
  hint=recommended, useposix=true, d_sigaction=define
  usethreads=undef use5005threads=undef useithreads=undef usemultiplicity=undef
  useperlio=undef d_sfio=undef uselargefiles=define usesocks=undef
  use64bitint=undef use64bitall=undef uselongdouble=undef
  Compiler​:
  cc='gcc', ccflags ='-D_BSD_TYPES -D_BSD_TIME -D_POSIX_C_SOURCE -DIRIX32_SEMUN_BROKEN_BY_GCC -fno-strict-aliasing -I/sw/common/include -DLANGUAGE_C',
  optimize='-O3',
  cppflags='-D_BSD_TYPES -D_BSD_TIME -D_POSIX_C_SOURCE -DIRIX32_SEMUN_BROKEN_BY_GCC -fno-strict-aliasing -I/sw/common/include -DLANGUAGE_C'
  ccversion='', gccversion='2.95.3 20010315 (release)', gccosandvers='irix6.5'
  intsize=4, longsize=4, ptrsize=4, doublesize=8, byteorder=4321
  d_longlong=define, longlongsize=8, d_longdbl=define, longdblsize=8
  ivtype='long', ivsize=4, nvtype='double', nvsize=8, Off_t='off_t', lseeksize=8
  alignbytes=8, usemymalloc=y, prototype=define
  Linker and Libraries​:
  ld='gcc', ldflags ='-Wl,-rpath,/sw/arch/lib -Wl,-woff,84 -L/sw/arch/lib'
  libpth=/sw/arch/lib /usr/lib /usr/local/lib /lib
  libs=-lgdbm -ldb -lm -lc
  perllibs=-lm -lc
  libc=, so=so, useshrplib=false, libperl=libperl.a
  Dynamic Linking​:
  dlsrc=dl_dlopen.xs, dlext=so, d_dlsymun=undef, ccdlflags=' '
  cccdlflags='-fpic', lddlflags='-shared -L/sw/arch/lib'

Locally applied patches​:
 

@​INC for perl v5.6.1​:
  /sw/common/test-patched/share/perl/5.6.1/mips-irix
  /sw/common/test-patched/share/perl/5.6.1
  /sw/common/test-patched/share/perl/site-perl/5.6.1/mips-irix
  /sw/common/test-patched/share/perl/site-perl/5.6.1
  /sw/common/test-patched/share/perl/site-perl
  .

Environment for perl v5.6.1​:
  HOME=/homes/starksb
  LANG (unset)
  LANGUAGE (unset)
  LD_LIBRARY_PATH (unset)
  LOGDIR (unset)
  PATH=/sw/arch/corba/ORBit-0.5.7/bin​:/sw/arch/pkg/MySQL/3.22.32/bin​:/homes/starksb/bin​:/homes/starksb/bin/IRIX64​:/sw/arch/bin​:/usr/sbin​:/usr/bsd​:/sbin​:/usr/bin​:/bin​:/usr/bin/X11​:/usr/etc​:/usr/local/bin​:/sw/common/share/parallel/pvm3/lib​:/homes/starksb/pvm3/bin/SGI6​:/sw/common/share/parallel/pvm3/bin/SGI6​:/sw/arch/lib/parallel/LAM/bin​:/sw/arch/bin/EMBOSS
  PERL_BADLANG (unset)
  SHELL=/usr/bin/bash

From @jhi

Okay, the -mabi=64 needs obviously be tested for existence.

+#if defined(sgi) && defined(__GNUC__)
+char *inet_ntoa(struct in_addr sa)

I think static char *my_inet_ntoa() + #define inet_ntoa my_inet_ntoa
would be better.

Isn't the rolling-your-own redundant *with* -mabi=64? In other words,
if one has -mabi=64, one doesn't *need* to roll-his-own? But, on the
other, hand, if one had -mabi=64, one *could* use the self-rolled?

I can see two options​:

#1 we add the roll-your-own (though a generalized a bit​:
  #ifdef NEED_OWN_INET_NTOA, and then IRIX hints define that)
  but also make the hints file to test whether -mabi=64 works
  and if it does, the NEED_OWN_INET_NTOA should be undef
#2 we add the roll-your-own (as in #1) but if the roll-your-own
  version works even with -mabi=64, we can drop the -mabi=64
  logic from the hints and just always use the self-rolled

Now, the problem with both of these options is that to test whether
our solution works, we need two gccs​: one built with -mabi=64 and one
without.