On Tuesday 24 June 2008 01:38:29 pm Carlo Marcelo Arenas Belon wrote: > On Tue, Jun 24, 2008 at 09:42:59AM -0400, Jarod Wilson wrote: > > On Tuesday 24 June 2008 06:45:14 am Carlo Marcelo Arenas Belon wrote: > > > attached an implementation which fixes the 2 issues raised before and > > > also does some extra cleanup, but that will need more testing in > > > different Linux distributions and with different autoconf versions > > > before it can be backported. > > > > > > as explained before, I don't think is a good solution though as it is > > > just extending the original hack (which wasn't really needed either) > > > > Just had another thought with respect to setting lib64 if > > host_cpu=powerpc64... The vast majority of the time, a 32-bit userspace > > is actually preferred on 64-bit PowerPC -- 32-bit ppc doesn't have the > > register limitations 32-bit x86 has. The gains you see with 64-bit x86_64 > > over 32-bit x86, outside of addressable memory, don't exist when going to > > 64-bit ppc binaries (since ppc32 has a saner register space than x86), > > and you often get marginally slower code, since there's an expanded > > memory footprint. Need for huge amounts of memory is one of the few > > reasons to run a 64-bit ppc binary on a 64-bit ppc kernel. > > which is why I presume fedora ppc and therefore the fedora ppc package will > be 32bit and should be using /usr/lib, right?, therefore the only builds > that should be supported in fedora will be: i386, x86_64 and ppc.
Nope, we'll have i386, x86_64, ppc and ppc64 builds. But unless you specifically request to install the ppc64 version on a ppc64 kernel, you'll get the ppc (32-bit) build installed. We still build everything ppc64 as well as ppc (when it builds), it just doesn't get installed by default. > > In other words, yes, lib64 is correct if building a ppc64 binary, but not > > for a ppc32 binary, which is actually what most people should be using on > > a ppc64 system. That being the case, if we automatically swap in lib64 on > > ppc64, it should probably be done only if host_cpu=powerpc64 *and* the > > binary being built is 64-bit. > > agree, and that was another deficiency of the current (and previous) hack, > which always assumed based on the type of CPU available which kind of build > was being used. in x86 that means that probably means that if you use > "CFLAGS=-m32" in an x86_64 system you will get 32bit ABI libraries in > the directory where 64bit ABI libraries are meant to be: /usr/lib64 That was my suspicion as well. > and then I presume the hack had just backfired and broke the ppc > package? Most likely. Not sure I actually tried a ppc build until taking in the patch that just uses whatever libdir was passed in from the %configure macro. -- Jarod Wilson [EMAIL PROTECTED] ------------------------------------------------------------------------- Check out the new SourceForge.net Marketplace. It's the best place to buy or sell services for just about anything Open Source. http://sourceforge.net/services/buy/index.php _______________________________________________ Ganglia-developers mailing list Ganglia-developers@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/ganglia-developers
