[darcs-users] growing the darcs team

[Stephen J. Turnbull]

Eric Y. Kow writes:

 > Lately, we have been working on ways to make darcs development go
 > faster.  Our goal here is to increase the number of expert darcs
 > hackers with time to work on the project.  To achieve this, we
 > intend to give developers better feedback (growing them into
 > experts) and in more timely fashion (so that they can work more
 > quickly).  Here are the changes we think will help us get there:

Looks good to me.  My project (XEmacs) works in the unstable -> stable
workflow, and it worked quite well (as far as moving the important
patches from unstable to stable) for a long time.  More recently
things have gotten problematic mostly because of lack of manpower
(which you have experienced and are trying to address, so no problem
for Darcs now), but also because of excessive divergence of the stable
and unstable branches.  Watch out for that divergence.

 > 7) Darcs will now be switching to a new time-based release model
 >    with new releases coming out every six months.

I'm not sure 6 months is a good timeline.  It's very easy in a 6-month
cycle to decide that "oh, a couple days' slippage is not a big deal,
and we want it to be '6-months-of-work quality'" and slip six months,
one week at a time.  (Cf. Emacs, which has on several occasions shown
itself capable of slipping 5 years, one week at a time. :-) If you aim
at a one-month cycle, though, suddenly even days count.

This is the cycle that Python (for its beta/RC cycles) and bazaar have
chosen.  Bazaar of course has about 6 fulltime paid developers
courtesy of Canonical, so the comparison may be unfair.  However,
AFAIK, none of the Python maintainers are paid to work on Python at
all (Barry Warsaw is a pure volunteer AFAIK, and Guido van Rossum has
to negotiate his "pure Python" time with Google).  So it's possible.
Both have been reasonably successful in adhering to the monthly
schedule, which is why I suggest thinking about it.

I don't much like the Bazaar model.  They're actually doing feature
releases on that schedule, and I think stuff gets lost and the
discussion on their lists too hectic.  In the Python model, feature
releases are (planned) to take place about once a year from now on,
with a 6 month feature development phase followed by a ~ six month
phase with one release per month (3 betas, followed by 2 or 3 release
candidates).  Perhaps something like the Python model would be
appropriate to Darcs?