[csw-maintainers] ARCH=all packages

[Dagobert Michelsen]

Hi,

Am 11.11.2008 um 16:05 schrieb Philip Brown:
> Our standards pages say,
>
> http://www.opencsw.org/standards/build
> "{ARCH} is usually the output of `uname -p`. But for certain special
> packages that run on all solaris hardware, or are otherwise
> architecture-neutral, it may be preferable to have ARCH=all "

Well, the standard is not clear in this point. It could mean

(a) the package doesn't contain data which is interpreted
     differently on different platforms

(b) the package can be used on any platform, irregardless of
     how this is achieved.

> I say that this means that archtecture specific binaries, should not  
> be in
> "ARCH=all" packages.

Complying to a standard is not a virtue on its own. When
you wrote this you had an idea what you invented it for.
The only thing I can think of is to save disk space.
We may discuss here "cleanlyness of implementation"
or "pureness of packages", but does it really get to the
point? IMHO the point is "Does it work?" And providing
multiple binaries in one package does work.

There was the argument of diskspace when installed. This
can be easily solved by putting the different binaries in
separate install classes (like 'sparc' and 'i386') and
install only the binary needed.

> Even if you do not agree that the words clearly state
> that, it was certainly my INTENTION when I wrote the standard.

That's the nice thing about standards, isn't it? ;-)

For myself I maintain the SE Toolkit (RICHPse), which traditionally
is a package that contains binaries for sparc and x86 for
maximum ease of use. However, I made two packages because
having one large package is not easier for the users. But
converting the build-infrastructure from the package was
hell :-P




...and now please forget all of what I wrote above. Because
those were technical arguments. Including binaries for sparc
and i386 does work, and having two packages works also. The
whole discussion is not technical. So gentlemen, please:

Peter: Is it really that important to have one fat package
instead of two? An admin usually does know on which platform
he is on. Will the package be installed on NFS and would it
be run from sparc and i386 at the same time? I guess not.
So two packages won't hurt and would fully comply with the
OpenCSW standard without further discussion.

Phil: The standard is not clear here, regardless of what you
intended with it. The package does work and it is usually
the responsibility of the maintainer to build the package
in the best way he thinks. The release manager (you :-) can
give advice on improvements, but if the maintainer is
reluctant to implement them the package must be released.
Unless, of course, it violates the standard...

So, Peter, would you please deliver two packages and
you, Phil, then release the combined package? ;-)


Best regards

   -- Dago