Precluding hardware-dependent issues

[Michael Ströder <michael@stroeder.com>]

HI!

I have strange seg faults when sending a SASL/DIGEST-MD5 bind with
OpenLDAP 2.4.11 running with BDB 4.6.21p1 on different VMware machines
running with SLES 10. I plan to update both software packages anyway
but, as usual, the customer is a little bit scared of upgrading. ;-)

I'd also like to figure out what the real problem is since on one
machine it simply works and it does not on another.

I've checked pre-installed SuSE packages, all related config files, file
ownership/permissions and dynamic linking of slapd (ldd output).
Everything seems ok. Start over with slapadd-ing to an empty DB didn't
help either. Running slapd -d 32767 on the concole shows that the SASL
user is correctly mapped to the user's entry DN but immediately after
that slapd crashes. (Can't grab a useful backtrace with the installed
build though.)

Both machines are identical except that one VMware machine is running on
an AMD processor where also the build was made and the other VMware
machine is running on an Intel processor where the built files were just
copied to. (This copying worked from one AMD machine to another).

So I currently suspect that there's a processor-dependent optimization
done especially in the BDB build. How can I avoid that? See the current
BDB build script below. Any other suspectible compile time option?

Many thanks in advance.

Ciao, Michael.

------------------------------ snip ------------------------------
export CFLAGS="-O2"
cd build_unix
rm config.status
make clean
../dist/configure \
  --prefix=/appserver/bdb-4.6 \
  --with-uniquenames \
  --enable-diagnostic \
  --enable-shared \
  --enable-static \
  --enable-rpc \
  --disable-java \
  --enable-posixmutexes \
  --enable-largefile \
  --enable-cxx \
  --disable-compat185 \
  i686-suse-linux
make