Divorce the 3 tiers

Udo Grabowski udo.grabowski at arcor.de
Tue Jul 13 11:48:04 PDT 2004

Hui Zhou wrote:
> <hijacking the thread here, sorry>
Good idea!
> ....
Your post indicates the central problem: Most of the active developers
concentrate on the new stuff. And I think it's the right way to go.
Those who are able and willing and anxious to push the development
should do that, that's what keeps the world running forward...
There's are need for a stable release, but I personally don't feel
that the testing area itself should live on its own right as a separate
sandbox for the not so brave.
So how reaching a stable, solid "release" ? The Mozilla development
shows how it works: develop a while, then, at more ore less regular
intervals, raise the bar for new stuff and let things settle under
the sharp eyes of an old wise guy (or two) to a seemingly working
state, then branch from unstable into a short-living testing
path, which finally ends up in a stable release. After the branch,
unstable can be hijacked by the crowd again, and experienced branch
drivers lead the testing branch into the stable release, kicking out
or taking in whatever is needed to make the branch really stable.

This has advantages: New stuff is regularily incorporated into
a stable release, they are also more often up to date than these days,
and the need to have people working on the testing branch (which
would otherwise do much flashier stuff) is relaxed to the release
phases. As done at mozillla.org, stable branches also could live
long (and even in parallel) with regular fix and update releases,
if somebody is interested keeping them in loose sync. You could even use
the tinderbox stuff to organize such a model, which, btw., also
allows a lot of testing and unstable branches in parallel.
So any developer who is not quite happy with the "main direction"
on the trunk could in principle maintain and test his own variant
(and even release from that branch) and hopefully inject good stuff
from there into the main trunk.

I'm just new to this LfS stuff, but the last 3 weeks showed me
clearly that LfS is really lacking a good development model, and
Mozilla is a good prototype applicable in most of its aspects here.
This would, of course, not satisfy everyone, but at least everyone
then knows some rules, and the unsatisfied can build their own
sandboxes if they wish so, instead of just giving up and leaving.

More information about the lfs-dev mailing list