Re: FreeBSD 6.0 and onwards

On 7 Nov 2004, Samuel Tardieu wrote:

> >>>>> "PHK" == Poul-Henning Kamp <phk_at_phk.freebsd.dk> writes:
> 
> PHK> For better or worse, we're stuck with CVS for the forseeable
> PHK> future and there are so many other ways to better use the huge
> PHK> amount of developer time it would cost to replace it.
> 
> I've never proposed to replace CVS with anything else. I proposed to
> replace Perforce by GNU Arch. 

It's worth pointing out that it's taken several years for Perforce to get
a useful level of acceptance by FreeBSD developers, who are frequently
quite set in their ways ("if it's not broke, let me get on with my
coding") :-).  However, it's fairly easy to do some useful experimentation
to demonstrate that it's a useful substitute.  A first thing to try is to
use it for ones own work: set up a regular import from CVS bringing in
FreeBSD.org changes as vendor branches, then maintain your own work
relative to it (perhaps you are already doing this?).  This will make sure
that the arch mechanisms are up to the non-trivial load of tracking
FreeBSD.

After that, it would be useful to post about the useful results -- i.e.,
"this worked really well -- the following were very easy to do, I used <n>
branches and merged between them, tracking FreeBSD daily on <x> branches",
etc.  You might already be doing this -- if so, you should say so :-). 
Then we should see about whether there are FreeBSD developers who
recognize the benefit of arch sufficiently to give it a spin using a
repository and see how well it works for them.  I.e., host some
sub-project out of arch with a few developers and make sure all is well --
see what rough edges annoy, and which don't.  The noticeable annoyances in
Perforce are things like the lack of offline operation and non-standard
patch format, for example.

I think the FreeBSD world is open to a new SCM, or things like Perforce,
Subversion, and Arch wouldn't keep coming up.  But I think there's yet to
be shown a real alternative to CVS that's worth investing the thousands of
man hours (yes, thousands) necessary to move over.  It's obvious Perforce
hasn't provided sufficient motivation to move to it as a primary
mechanism, but that it has proven useful enough that a lot of the
mainstream developers use it as a tool to manage works-in-progress.  The
path there was pretty hard, so anticipating similar resistence to
alternatives is probably realistic :-).

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert_at_fledge.watson.org      Principal Research Scientist, McAfee Research