RFD: Architecture field being retarded? [was: How to specify architectures *not* to be built?]

[Andreas Rottmann <a.rottmann@gmx.at>]

>>>>> "Russell" == Russell Coker <russell@coker.com.au> writes:

    Russell> On Sun, 11 Aug 2002 16:35, Geert Stappers wrote:
    >> When the cause of the buildproblem is in the package, fix the
    >> problem there. The package maintainer hasn't to do it by
    >> himself, he can/must/should cooperate with people of other
    >> architectures.  A sign like "!hurd-i386" looks to me like "No
    >> niggers allowed", it is not an invitation to cooperation.

    Russell> So you think I should keep my selinux packages as
    Russell> architecture any, even though they will never run on on
    Russell> HURD or BSD?

Thanks, Russell, you are making my point. It is similiar with radvd,
which was designed for Linux/BSD and won't work on the HURD, since it
simply isn't supported upstream. I am not in the position to port
radvd to the HURD, altough this would be the ideal way to go.

In any case, I think the current architecture format/scheme has
serious deficiencies. I guess there are a bunch of packages around
that are aimed at a particuliar OS Kernel/Platform (or at several of
them). radvd and selinux are two examples. IMO, it should be possible
to tell the autobuilders and people trying to build the package "hey,
watch out, this is a linux/BSD/whatever sepecific package", or,
e.g. (as is the case with radvd): "this package will build only on BSD
and linux based operating systems".

I guess we should therefore discuss how we can make such build
requirements clear without stating every platform (Linux, BSD, Hurd,
...) and processor combination (i386, hppa, arm, ...).

I can see two different reasons for making a package neither
'Architecture: all' nor 'Architecture: any'.

1) The package is tailored towards a specific (set of)
   platform(s). This is the case with selinux, AFAICT: it is designed
   for linux, and will never be ported to another kernel/platform (i
   guess ;-). radvd is a bit different in that it supports more than
   one platform (linux + BSD) and might be ported to more platforms in
   the future.

2) The package is somehow dependent on processor architecture. cpuid
   and x86info come to mind. They are only useful (and buildable) for
   i386 derivates.


What I want is a way to specify something like:

Platform: any
CPU: i386

for example. (This would fit cpuid and x86info)

radvd would then get

Platform: linux, bsd      # As soon as we have an official bsd port
CPU: any

This was of course just a suggestion of a possible solution.

I think in the light of upcoming *BSD ports, the HURD port and the
(hopefully not too distant ;-)) possibility of getting the HURD ported
to other CPU families, we really need improvement here.

Regards, Andy
-- 
Andreas Rottmann         | Dru@ICQ        | 118634484@ICQ | a.rottmann@gmx.at
http://www.8ung.at/rotty | GnuPG Key: http://www.8ung.at/rotty/gpg.asc
Fingerprint              | DFB4 4EB4 78A4 5EEE 6219  F228 F92F CFC5 01FD 5B62