Re: [gentoo-releng] 2004.2 planning

[Nathaniel McCallum]

On Sun, 2004-05-02 at 12:04 -0500, Lance Albertson wrote:
> On Sun, 2004-05-02 at 11:45, Donnie Berkholz wrote:
> > On Sun, 2004-05-02 at 09:32, Nathaniel McCallum wrote:
> > > On Sun, 2004-05-02 at 07:29 -0400, Kurt Lieber wrote:
> > > > * Releases are largely meaningless, comprised of nothing more than package
> > > >   updates for GRP.
> > >
> > > Lets use 2004.x quarterly releases to perfect our GRP proceedure and get
> > > GRP online.  Once we are at that point we can divorce GRP from liveCD
> > > releases if we need to.
> >
> > Exactly what I was thinking when reading this thread. Just provide a
> > package+stages CD (or two) quarterly, and let them use a LiveCD from the
> > (biannual?) LiveCD release. From what I see, it appears that much of the
> > QA time goes into getting the LiveCDs working properly.
>
> Funny, I was actually just thinking of that last night before I went to
> bed. I see that quarterly releases for stages/grp package sets are
> needed, but I really don't see a need for quarterly livecd releases.
> There needs to be enough time to work out bugs/implement new features,
> etc. The only thing that I can see for quarterly releases is driver
> upgrades on the kernels used on the livecds. So, that may be debatable.
>
> Thoughts?

The real problem here is not a problem with releng so much as it is a
problem with portage.  I'm not knocking portage at all, it has served us
great thus far.  However, having a seperate GRP release is really
inconvienient as compared to a binary tree.  Most enterprise users and
most personal users who don't have a ton of time on their hands, want
the ability to build from source if they need too, but to use binaries
when they don't.  Portage does GREAT at the first, but not so great at
the second.  If we had a binary tree(s), we could do with more
infrequent liveCDs (and a lot less stress on releng).  GRP should be a
temporary provision until portage has better binary support (notice the
emphasis on TEMPORARY).  However, there are even some devs talking about
canning GRP.  This is simply not an option, but neither is making GRP
perminant.  We need to, once and for all, engineer portage for the
greatest possible flexability.  Anyway, I know I'll probably get flamed,
but as I see it we only have two options kill off GRP or make portage
support a binary tree.  I've said enough...

don_fireproof_suit()
Nathaniel


--
gentoo-releng@g.o mailing list