>> The kernel version is probably not relevant here; what's more
>> interesting is the compiler version, compiler optimization level,
>> and perhaps libc version.
> cshobe@gabrielle:/www/htdocs$ gcc -v
> Reading specs from /usr/lib/gcc-lib/i686-pc-linux-gnu/2.95.3/specs
> gcc version 2.95.3 20010315 (release)
> cshobe@gabrielle:/www/htdocs$ echo $CFLAGS
> -O3 -march=i686 -mcpu=i686 -funroll-loops -ffast-math
Bingo: it's the -ffast-math option that's causing the problem.
See, eg,
http://fts.postgresql.org/db/mw/msg.html?mid=1277557
If you rebuild without that, I think you'll find the problem goes away.
ISTM that at one time we had considered actively discriminating against
-ffast-math in CFLAGS in configure ... but I don't see any sign in
current sources that we make any attempt to remove -ffast-math from
environment-supplied CFLAGS. Peter, do you think that would be a
reasonable thing to do?
regards, tom lane