[lsb-discuss] LSB conf call notes for 2008-06-04

Joseph Kowalski jek3 at sun.com
Fri Jun 6 10:28:09 PDT 2008

Theodore Tso wrote:
> On Thu, Jun 05, 2008 at 08:41:49PM -1000, Joseph Kowalski wrote:
>> I understand why an distro might want to provided both.  I'm a bit
>> lost as to why this is an LSB issue.  Does the LSB generally require
>> support for 32-bit objects on 64-bit platforms?
> The LSB doesn't require it, no.  I'm not sure we're even 100% set up
> for a distribution which wants to claim simultaneous support for x86
> and x86_64 LSB 4.0 conformance for their 64-bit distribution.
> If we *do* want to allow that sort of thing, though, at some point in
> the future, it becomes problematic to require a fixed pathname such as
> /usr/bin/java, since at the point where you start the JVM, it might
> not be possible to determine in advance whether the 64-bit or 32-bit
> JRE was desired.  (You might be able to search a provided CLASSPATH
> and try to find some .so files, but #1, I'm not 100% user the .so's
> for JNI's are found in the CLASSPATH, and #2, thats still a hueristic
> which could be fooled.)  BTW, This isn't so much a problem for other
> hard-coded paths, such as /usr/lib/sendmail since if a process is
> trying to send an email by running /usr/lib/sendmail, it doesn't
> really care whether it is 32-bit or 64-bit.
> This is a minor issue, since presuambly for a primarily 64-bit distro
> will probably have 64-bit binaries in /usr/bin, it is something to
> consider if we want to support multiarch.  (Which by the way will be
> more common on PPC, doesn't have the problem with x86's pathetically
> small register file which means that x86_64 binaries do have a real
> performance advantage, even though pointers are doubling in size.)
> 	    	       	    	   	    	- Ted

As I noted, there is support on Solaris to support this choice via the 
command line.  Its there for linux also, its just a bit cryptic (this 
discussion should make this "first class").

On SPARC. the performance of 64-bit vs 32-bits is application dependent; 
some are faster, others are not. If Intel had not taken advantage the 
opportunity to double the size of the register file, it would be in the 
same situation.

- jek3

More information about the lsb-discuss mailing list