Re: An old idea, brought back to life
On Thu, Dec 17, 1998 at 12:18:25PM -0500, Peter S Galbraith wrote:
>
> Joseph Carter wrote:
>
> > Ben, did you hear a proposal made by someone, I think Anthony Towns was
> > part of the discussion if not the person who was suggesting it (someone
> > will correct me if my memory is in error), that unstable become a never
> > released distribution?
> >
> > The idea was that unstable is always unstable. As packages became stable
> > they would be moved to a release distribution which is always kept (or at
> > least there is a reasonable attempt to keep it) as stable and as close to
> > release as possible.
>
> How does it handle changes that affect all packages (likelibc5->libc6) ?
Indeed.
Even without the libc5->libc6 migration, I don't think the proposal
has hands and feet. If we model each release as a 'product', then the
idea of migrating stable packages produces two debug phases. First in
the unstable and then in the stable. Just because a package works
among other unstable ones doesn't mean it will work in the stable
environment.
I think that the current release model is a more accurate
representation of the process. We accumulate new toys on the
playground until the chief calls freeze. Then, we bang away at the
mess until we're satisfied that it works--this is a big problem now
because we don't have formalized testing nor do have we defined
release criteria aside from bug-count. When the bell rings to return
to our desks, we drop packages that fail the bug-count test.
In the incremental model, we risk upsetting what we believe is
'stable' by importing packages that we can never know enough about.
Stable means that we believe the whole kit-and-kaboodle will work
together. What sort of criteria (aside from limited bug fixes) could
we *ever* describe that would justify updating stable. And bug fixes,
too, are dangerous. Look at the kernel as it moved along the 2.0.x
track. Many folks thought that each step was 'just bug fixes and
*very* important driver additions'. The nfs client is broken in
2.0.35 and there are reports of security holes, blah, blah, blah.
I'll suggest again that we inject some of the published wisdom about
software release into Debian before we contemplate changing the
existing process. I think there will be some valid changes to make.
The incremental update one is not one of them.
Cheers.
Reply to: