Predict freezes, not releases.

[Chad Miller <cmiller@surfsouth.com>]

Hasty freezes cause missed deadlines.  Three running trees of Stable, 
Frozen, and Unstable would solve our release-length woes.

My reasoning follows:

We can never set a release date, and we don't have terribly specific
goals, we know.  The answer of ``when it's ready'' might ring hollow to
the infidels, so I wonder if we can be more specific about the antecedent
of ``it''.  

We probably mean ``when Woody is ready'', but Woody's just a collection
of pieces of software, and software is _never_ finished.  Instead, I
think it might be a good idea to mean ``when the bugs are out of the
current cool stuff''.  I suggest that after some well-defined period of
working with the latest stuff, we freeze Unstable and define a new
Unstable.  Then, we forbid nonbugfix changes to the new Frozen (no
changes in the major-version numbers!  This means you!) and work towards
the Stableification of Frozen.  We release when _that_ is finished.

As I see it, the problem of our release rate (if one sees it as a
problem), is that our planning of a freeze is too immediate.  That is, 
some majority of decide to freeze in a week, e.g., and developers of
software who know that their upstreams will be releasing a version with
Twice as Many New Widgets! in two weeks resist.  The short-range freeze 
seems arbitrary to everyone, and it's ignored when convenient.

If we decided to freeze on some well-defined-in-advance point (like the
equinoxes), and fork off a new Unstable, we developers won't be as likely 
to moan and break the freeze, because there's a place to put our new
toys.

There's the question of developer's attention being diluted by having two
sources of interest until Frozen is released.  I don't think it's too much
of a burden, and will make developers' lives easier, most of the time.
No more problem of being torn between resisting/breaking the freeze in
order to put in the most recent version, and following the schedule.

Also, with more releases, there's no shame in holding off from the most
recent version, because the Stable will be replaced in (using the
equinoxes as freeze-points) six months -- not more than a year!  There
need never be ``exceptions'' and delays again.

So, am I nuts?  Overzealous?  Has this been discussed before? 

						- chad

--
Chad Miller <cmiller@surfsouth.com>   URL: http://web.chad.org/   (GPG)
"Any technology distinguishable from magic is insufficiently advanced".
First corollary to Clarke's Third Law (Jargon File, v4.2.0, 'magic')