Re: current mozilla or phoenix debs for PPC?
Wouter Verhelst <wouter@grep.be> writes:
> Op vr 28-11-2003, om 06:40 schreef Scott Holder:
> > Chris Tillman wrote:
> >
> > > Well, it's true we owe our elders respect (as I give a quick glance
> > >
> > >towards my Mac IIci). But, OTOH, I think the current philosophy of
> > >all-or-none may be a little too inflexible. Especially as m68k users
> > >get fewer and fewer, and developers appear to be an endangered
> > >species. We need a plan for quiet, benign senility where some
> > >architectures are concerned. This does not involve leaving them out in
> > >the cold to die, just restricting them to given (working) versions
> > >and letting the rest go on.
>
> I didn't see the quoted mail, as I'm not subscribed to -powerpc.
>
> Obviously, this argument has been brought up numerous times, and part of
> it is, most certainly, true (some packages really are useless on m68k).
> However, if we would decide to stop compiling some packages for the m68k
> architecture, there's a serious problem: How do we decide what packages
> we will build for m68k? If we do decide to stop building some packages,
> we have to draw a clear line somewhere; but it's hard to tell which
> packages are useless, and which aren't; there's no clear line to draw,
> only a (very) blurry one. The reason for this is that "useful" is a very
> subjective statement. Some people might find games a useless waste of
> time, for instance; in very much the same way, some people might think
> that running KDE on m68k is a useless waste of time, while others might
> disagree.
>
> Even if we did manage to find a good subset of packages which would be
> 'useful' for our users, I'm sure there will be questions of disappointed
> users inquiring why packages foo and bar are not available.
...
Consider the following:
A good starting point would be the popularity-contest data. Anything
used in the last half year gets build.
Every package thats not compiled is replaced by a dummy package
stating why it isn't autobuild, explaining the problem.
When installed the dummy package (or a locally build version of the
missing deb) would get reported by popularity-contest and autobuilders
would pick it up again. The dummy package should have version 0 so any
deb is (hopefully) newer.
Alternatively or parallel to that there could be a web or mail
interface to get packages added again (which should probably make them
top of the buiild queue for the first build).
At first some packages would be missing and some people would scream
but we can warn before implementing this and hope enough people
install popularity-contest on m68k to make this minimal.
I'm not advocating this but if we take a turn for the worse this would
be an option.
MfG
Goswin
Reply to: