Subject: Re: Tangent: Current-kernel revision naming...
To: Chris G. Demetriou <cgd@netbsd.org>
From: Frederick Bruckman <fb@enteract.com>
List: current-users
Date: 02/17/2000 00:41:06
On 16 Feb 2000, Chris G. Demetriou wrote:

> Andrew Brown <atatat@atatdot.net> writes:
> > that makes less sense to me.  isn't there plenty of stuff in 1.3K that
> > doesn't go into 1.4, but *does* make it into 1.4A?

Some projects seem to have a "development" track containing wild,
unworkable stuff, which is later either merged into the release or
discarded. NetBSD doesn't do that--rather, the releases are branched
off of the main, development, trunk.

> no.  by definition, everything in any 1.3x version will be in 1.4
> (unless it's deleted, or whatever).

Here's my considered opinion, for what it's worth.

1) The versions after "Z" should user two small letters: 1.4Z -> 1.4aa
-> 1.4ab -> ... -> 1.4az -> 1.4ba ... . This sorts correctly, without
appearing too wierd. This scheme permits 702 revisions. If that's not
enough, we'll cross that bridge when we come to it.

2) The "releases", or rather, the non-released interim release track,
should increment the same way. Seeing things like 1.4.1A, 1.4.1B, will
suddenly put 1.4A - 1.4zz in perspective, besides the obvious benefit
of telling the user what degree of binary compatability to expect. I'm
not suggesting that we go back and renumber 1.4.1, but the first patch
to 1.4.2, post release, should be to increment the version to 1.4.2A,
just as 1.4 became 1.4A.

Thus, it'll be clear that polished, coherent, tagged releases have no
letters after them, whereas appended letters imply evolution from the
(base, unlettered) release.