http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46655
--- Comment #23 from Richard Nolde <richard.nolde at cybox dot com> 2011-02-25
00:20:54 UTC ---
On 02/24/2011 01:42 PM, michael.haubenwallner at salomon dot at wrote:
Progress report on building Gcc 4.5.0 on AIX 6.0:
Using your suggestion for gmake bootstrap STAGE1_FLAGS=-0 gets me much
further in the build. The problem has moved to building libgomp, and the
failure appears to be due to the fact that AIX nm needs nm -B to be
compatible with the AIX ld. The top level configure scripts seem to deal
with this fine, but in the libgomp directory, configure builds a
makefile that invokes
/devel/opensource/gcc/gcc-4.5.0/host-powerpc-ibm-aix6.1.0.0/gcc/nm
and bombs.
If NM is defined/exported as NM="/usr/bin/nm -B", or "/usr/bin/nm -B
-X32_64", the script:
errors out a line 83.
[83]: /usr/bin/nm -B -X32_64: not found
I believe the command on line 83: exec "$original" ${1+"$@"} is
looking for a file named "-B" or "-B -X32_64" depending on how you
define NM. It would seem that we need an NM_FLAGS variable comparable
to LD_FLAGS in which to pass the extra flags required to make AIX do the
right thing here. The question is why do the other makefiles built by
configure not have the same problem? If you don't define NM="/usr/bin/nm
-B" at least, the top level configure reports that you cannot build
shared libraries, which is not true on AIX. AIX shared libraries contain
both 32 bit and 64 bit objects and can use .a or .so for a suffix.
It appears that the case statement at the top of the file is going to
fall through without matching nm since basename returns nm -B and we
then fall through the second case statement and invoke /usr/bin/nm such
that it looks for the file -B. I not quite sure on this one.
...
55 case "$original" in
56 ../*)
57 # compute absolute path of the location of this script
58 tdir=`dirname "$0"`
59 scriptdir=`cd "$tdir" && pwd`
60
61 if test -x $scriptdir/../$dir/$prog; then
62 test "$fast_install" = yes || exec
$scriptdir/../$dir/$prog ${1+"$@"}
63
64 # if libtool did everything it needs to do, there's a
fast path
65 lt_prog=$scriptdir/../$dir/$objdir/lt-$prog
66 test -x $lt_prog && exec $lt_prog ${1+"$@"}
67
68 # libtool has not relinked ld-new yet, but we cannot just
use the
69 # previous stage (because then the relinking would just
never happen!).
70 # So we take extra care to use prev-ld/ld-new *on
recursive calls*.
71 test x"$LT_RCU" = x"1" && exec
$scriptdir/../prev-$dir/$prog ${1+"$@"}
72
73 LT_RCU=1; export LT_RCU
74 $scriptdir/../$dir/$prog ${1+"$@"}
75 result=$?
76 exit $result
77
78 else
79 exec $scriptdir/../prev-$dir/$prog ${1+"$@"}
80 fi
81 ;;
82 *)
83 exec "$original" ${1+"$@"}
84 ;;
85 esac
86
Adding a few echo statements to send current values to a log file before
line 55 and at line 83 and then resuming the make from the last
invocation produces the following:
invoked is set to nm
prog is set to %{prog}
dir is set to binutils
original is set to /usr/bin/nm -B -X32_64
Calling exec /usr/bin/nm -B -X32_64 -B -X32_64 -BCpg .libs/alloc.o
.libs/barrier.o .libs/critical.o .libs/env.o .libs/error.o .libs/iter.o
.libs/iter_ull.o .libs/loop.o .libs/loop_ull.o .libs/ordered.o
.libs/parallel.o .libs/sections.o .libs/single.o .libs/task.o
.libs/team.o .libs/work.o .libs/lock.o .libs/mutex.o .libs/proc.o
.libs/sem.o .libs/bar.o .libs/ptrlock.o .libs/time.o .libs/fortran.o
.libs/affinity.o
Let me know if I can provide further debugging information.
Thanks,
Richard Nolde
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46655
>
> --- Comment #20 from Michael Haubenwallner<michael.haubenwallner at salomon
> dot at> 2011-02-24 20:42:46 UTC ---
> (In reply to comment #19)
>
>>> /usr/bin/gcc -c -g -fkeep-inline-functions -DIN_GCC ...
>>>
>> This is a problem in /usr/bin/gcc, not in the GCC sources you're compiling.
>>
> (In reply to comment #1)
>
>> $ gmake bootstrap STAGE1_CFLAGS=-O
>>
> Richard, does this workaround work for you?
>
>