Header problems when building on AIX

[Walter Landry <wlandry@ucsd.edu>]

Greetings,

I'm trying to build gcc on an AIX 5.2 box which I don't have
administrative rights to.  There are two versions of gcc already
installed: 3.2 in /usr/local/bin and 3.3 in /usr/local/gnu/bin.
However, when I try to build things with MPI (using the -mpe switch),
using either gcc I get linker warnings and the code segfaults.  I've
also tried the binary at

  http://aixpdslib.seas.ucla.edu/packages/gcc.html

but it is built for AIX 5.1 and gives the same errors.  I also tried
the binary for AIX 5.2 at

  http://www.bullfreeware.com/

but it seems to require root to install.

So I am trying to build my own version of gcc, since IBM's compiler
has caused me problems in the past.  Whenever I try building gcc with
either installed version of gcc, I get internal compiler errors.  So
I've been trying to use IBM's compilers.

However, if I run

  /tmp/work/wlandry/gcc-3.3.4/configure --enable-languages=c,c++ --enable-threads=aix --prefix=/tmp/work/wlandry/gcc-bin
  /usr/local/bin/gmake bootstrap

then I get an error involving md5.h and conflicting symbols.  It turns
out that the compiler (cc) is looking in /usr/local/include for some
reason, and there is an md5.h there.  I've managed to get past that by
putting md5.h and md5.c in the same directory and compiling it
manually.

However, I then get similar errors involving /usr/include/errno.h
being included by libiberty/strerror.c.  It is again confused by
conflicting symbols.  If I manually work through this error the same
way as before, I get a similar error later on.  And another if I
continue.  And another.  Then I gave up.

This happens whether I'm using gcc 3.3.4 or 3.4.1.

So my question is, is my setup so broken that I'm just hosed?  The
required APAR fixes have been installed, I'm using gmake, and I'm
configuring with absolute pathnames.  The only success report for AIX
on the buildstat page was building from an older version of gcc, and
they had to play around with some standard includes.  Has anyone been
able to build it with IBM's compiler?

Thanks,
Walter Landry
wlandry@ucsd.edu