[csw-maintainers] ARCH=all packages

[Ben Walton]

Excerpts from Philip Brown's message of Tue Nov 11 10:05:59 -0500 2008:
> Peter B. and I are having a disagreement about the concept of ARCH=all
> packages, so I thought I would bring the discussion here.
> 
> The issue is that he wishes to bundle specific binaries for wget in his
> pkgutil package, AND call it ARCH=all.
> (he wants to bundle one for sparc and one for x86)
> 
> I suggested that he either remove the binaries and use pure perl,
> (either with a full wget replacement, or to use it to download
>  "the appropriate binary")
> or split it up into two separate packages, with appropriate ARCH settings
> for the binary he includes in each one.

A few ideas:

1. Does the standard perl as shipped on sol8 include Net::HTTP and a
checksumming module sufficient for pulling down a wget binary and
verifying it?  If so, I think that would ultimately be the best
solution, as you don't want these 'spare' wget binaries after the system
is 'csw bootstrapped' anyway.
2. Alternately, the tool could bootstrap the installation of Net::HTTP
from cpan?  This treads into territory that could cause problems too,
since it will affect things outside of /opt/csw.
3. Include ascii-armored(/clear-signed?) versions of the applicable wget
binaries in the share/doc/pkgutil/bootstrap part of the packages' file
tree?

> I say that this means that architecture specific binaries, should not be in
> "ARCH=all" packages. Even if you do not agree that the words clearly state
> that, it was certainly my INTENTION when I wrote the standard. 
> I wanted "ARCH=all" to clearly indicate, "this package will work
> EVERYWHERE, it will never need recompiling, and it is safe to be shared in
> some kind of NFS /opt/csw for any and all solaris machines to use."

I agree with this and think it's best to adhere to it if at all
possible.  If any package were to be made an exception though, this
would be a good candidate in my estimation since it has a pivotal role
in getting a system boot-strapped to csw in the first place.

> In other words, ARCH=all should be reserved for things like raw
> arch-neutral data files, docs, java, and shellscripts/perl/python.

If the wget binaries were no longer in binary form (ascii armored),
would that skirt the definition enough to appease all?  pkgutil would
simply treat them as data and use the appropriate one to pull down the
CSWwget package initially...

The definition of ARCH=all could also be taken to mean that the package
will provide equal _functionality_ on all supported machines, which in
this case it would.  I'm not suggesting that this _should_ be how the
definition is read, just that it _could_ be read that way.

-Ben
-- 
Ben Walton
Systems Programmer - CHASS
University of Toronto
C:416.407.5610 | W:416.978.4302

GPG Key Id: 8E89F6D2; Key Server: pgp.mit.edu
Contact me to arrange for a CAcert assurance meeting.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL: <http://lists.opencsw.org/pipermail/maintainers/attachments/20081111/604a6518/attachment.asc>