[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Debian versioning scheme (r1 vs .1)

On Tue, Nov 05, 2002 at 05:24:31PM +1000, Anthony Towns wrote:
> On Mon, Nov 04, 2002 at 05:37:14PM -0700, Joel Baker wrote:
> > I don't see any *fundamental* reason why we couldn't, either. 
> Try installing sarge right now.

Try taking quotes completely out of context.

"Fundamental reason", as I used it, was meant to indicate something on the
order of a law of nature. Decisions to abandon (problematic) old software
before the replacement even *exists* in any meaningful fashion do not

Again, I assert: the only thing which truly prevents Debian from having an
actual release schedule is the utter lack of interest in changing way in
which we decide when to do a release.

In other circles, "we have six months until we try to release again" means
that if the installer takes 18 months to write, you assume that you do not
have it for 3 release cycles. Other things may happen to cause the cycle to
slip, often unforseen (and exceeding normally accounted for) last-minute
delays, but they don't *assume* that 18 months is a viable release cycle.

And it is proven to function, and work. Debian's method is also proven
to function, and work... up to a point. I say that empirical evidence
indicates that we have reached the point where it is no longer tenable as
it is.

Prior to Woody, I heard the refrain "testing will fix this, but it hasn't
had a chance to sit between major releases to prove it". Okay, valid
argument. Now that it's sitting between major releases, it may well be
true that we could freeze and release much faster (having not yet tried
to do it, I can only say 'it appears likely'). So we... make choices that
guarantee another 2-year release cycle.

Seriously: can you name at least one other significant (let's say, oh, 1000
users as a reasonable minima; certainly we have far more, as do most things
folks would consider major), evolving (as opposed to mature) project which
makes feature-upgrade releases once every 2 years, and is considered at all

I can name only one off the top of my head - TinyFugue. Which is steadily
having people switch away from it to other clients that offer the same
basic function, because it hasn't even remotely kept up with features. I
wouldn't say it is a success any longer, so much as "the only client of
it's capabilities on most Unix boxes". That seems to be much of the staying
power it has, and is steadily becoming less true as other clients advance
to fill the void.

I mean, we're not talking about the window manager flavor of the week, or
SpamAssassin, or the Linux development branch when it's in kernel of the
week mode. Things like BIND, Sendmail, and Apache all have regular feature
upgrades... and due to Debian's policies on stable (which are otherwise
perfectly sane and reasoanble), we end up with badly outdated versions of
even such slow-change tools.

A private conversation had a rather telling comment in it, the other
day; another developer noted that much of the pressure to actually get a
release out the door seems to have vanished, with the advent of 'testing';
most developers run at least that, on any workstation, and there were
significant numbers of servers upgraded to it during the final months of
Woody's preparation, because it was more feasible to track the updates and
manually watch for security fixes than it was to try to update all of the
software that needed to be updated from a potato system with software two
years out of date.
Joel Baker                           System Administrator - lightbearer.com
lucifer@lightbearer.com              http://users.lightbearer.com/lucifer/

Attachment: pgpT07ngRvjcq.pgp
Description: PGP signature

Reply to: