[csw-maintainers] Simplify OpenCSW bootstrapping on a server

[James Lee]

On 22/09/10, 17:36:31, Philip Brown <phil at bolthole.com> wrote regarding Re:
[csw-maintainers] Simplify OpenCSW bootstrapping on a server:

> > Pardon me for not recalling the thread but "ARCH=all" means what it
> > says and not "binaries=none", just as ARCH=i386 does not mean it
> > contains only i386 files and no amd64.

> amd64 is in the same architecture "family" as i386, so I dont see any
> relevant conflict in that comparison.

elfdump and nm show sparc and i386 binaries belong to the same family
but I fail to understand the relevance of family likeness as an AMD64
binary has the same chance of running on a pure 386 machine as a Sparc
binary - none.

What are you trying to achieve by saying no ELFs in an all arch
package?  Seems you are only constraining yourself.  You can't mean no
executable that won't run (on the named ARCH=) because amd64 and
sparcv9+vis2 are both examples where this isn't true.



> > Interestingly the man page for pkginfo(4) makes no mention of "all":

> However, it is a clear de-facto standard employed by sun.

This de-facto internal "standard" is not clear to me.

SUNWglow (arbitrarily chosen) is not ARCH="all" but contains no ELFs,
only arch independent files and the contents cksums match on i386 and
sparc.



> For example, the following packages in the solaris distribution:

...

> >                               "Third  party  application software
> >          should restrict itself to ARCH values from the following


"Third  party  application software" ...




> and you'll notice that "ppc" is a valid "ARCH". Therefore, if you're
> going by strict definitions, something cannot claim to support "all"
> ARCH values, unless it supports ppc.
> Does the package in question do that? :-}

Show me an Oracle supported customer copy of Solaris 10u9 for PPC and
we'll talk.



James.