On Fri, Nov 11, 2011 at 10:16 PM, case <case.biomaps.rutgers.edu> wrote:
> On Fri, Nov 11, 2011, Ross Walker wrote:
> >
> > I do not want to discourage people, I am just trying to make you all
> > consider the consequences if you mess up.
>
> Maybe we need a new branch, called "cruise-control". This would never be
> updated directly, but would periodically (say weekly for now, more often
> as we approach deadlines or releases) merge in master, then trigger its
> own cruise control test. Things that broke the cruise-control branch
> would incur the wrath of Ross, and maybe an automatic roll-back of that
> branch to the last known good state.
>
I like this approach.
> Effectively, the current master branch would be a buffer for mature
> commits. Cruise control would continue to be available, but there would
> be less pressure on commits -- you'd have a day or so to fix something
> that didn't work like you thought. An update to the master branch would
> imply
> a commitment to follow it on cruise control and to promptly fix any
> problems
> found there.
>
> People who wanted to test out the current code could clone the
> "cruise-control" branch and have (some) added assurance of not having to
> fight trivial bugs along the way.
>
I don't think git has a way of turning off commit access in just one
branch, but we'll just have to enforce it past the software level :).
> Less mature work would continue to be done in "-dev" branches, still with
> the
> hope and expectation that people working in these branches will
> periodically
> merge in master (to keep code divergence somewhat in check).
>
I'll inject a comment or 2 here. I personally think we as a community
should be a bit conservative with development branches we publish to
git.ambermd.org. (I personally don't publish any of my -dev branches on
git.amber, but on a local gitosis repo set up here)*. I think that's a
better alternative to dumping everything on git.amber in many (not all)
cases. (I count 18 branches on git.ambermd.org:/amber.git, many of which
haven't been touched in a long time). The frequent merge advice is good
advice, and I echo it.
For the PIs out there, having a group-wide git machine with gitosis set up
can do worlds for group organization, collaboration, and efficiency. Take
it from Darrin, he has a great system going on in his lab using git and
gitosis (sorry for throwing you under the bus).
All the best,
Jason
* I did accidentally push a branch in recently, but promptly deleted it :)
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Fri Nov 11 2011 - 21:30:03 PST
