On Fri, 2007-03-23 at 13:24 -0500, Les Mikesell wrote:
Matt Hyclak wrote:
On Wed, Mar 21, 2007 at 10:41:23AM -0500, Les Mikesell enlightened us:
I'm trying to use opennms to collect data from some CentOS3.x x86_64 boxes but I am getting an error "Integer too large: cannot decode" that http://bugzilla.opennms.org/cgi-bin/bugzilla/show_bug.cgi?id=1316 says is a bug in net-snmp that was fixed in a later version. However, when I try to build the 5.4.1 src.rpm from sourceforge I get "/usr/lib/libpopt.so: could not read symbols: Invalid operation" although it builds OK on a 32 bit system. Has anyone worked around this problem? Or is it really a bug in opennms?
In a time long ago and a land far far away, I tried building a 64-bit package with a 32-bit popt installed on it (I think it was galeon, but I could be wrong). As long as popt.i386 was installed, the build failed with the same error. Removing it (leaving only popt.x86_64) allowed the build to complete.
If you're going to build 64-bit packages, you'll want a system with as few i386 packages as possible. I recall an e-mail earlier this month from Johnny that listed only 3.
Thanks - I had to use --nodeps' to get rid of popt.i386 but after that the 5.4.1 rpm rebuild worked, and it is working now with opennms.
Just for the record on building packages on x86_64 ...
If one is going to routinely build packages for x86_64, it is better to setup a machine (or VM, chroot, mock buildroot, etc.) that is totally x86_64 RPMS with only the glibc.i686 nad glibc-devel.i386 packages installed from the i386 arch.
There are several other packages like popt.i386 that will cause issues AND also since the gcc builds both i386 and x86_64 packages and linker can link against both i386 and x86_64 packages, what you can end up with is packages that build/link against the wrong things on x86_64.