Source Control system results and discussion
Peter Avalos
pavalos at theshell.com
Mon Oct 27 19:01:26 PDT 2008
On Mon, Oct 27, 2008 at 08:49:26PM -0400, Joe Talbott wrote:
> On Mon, Oct 27, 2008 at 05:15:33PM -0700, Matthew Dillon wrote:
> >
> > I propose first that the DragonFly repository be officially accessible
> > via both Git and Mercurial. Pick your favorite, for read access. I
> > do not think there is any argument here :-)
> >
> > The question now devolves down to those people who commit into the
> > system. Do we use Git as the master repository and Mercurial as the
> > slave, or do we use Mercurial as the master repository and Git as the
> > slave?
> >
Git as the master.
> > I propose second that we use Git as the master repository and Mercurial
> > as the slave, but allow committers to commit to either and auto-merge
> > in both directions.
> >
> > I believe this can be done by using Git's superior branch management
> > to create a staging branch in Git that is mirrored from Mercurial,
> > and then merge that branch into Git's master branch. All automated,
> > of course. Several git users, such as corecode and I, would have no
> > trouble writing such a script.
> >
> > * I am worried that Mercurial's branch management is not good enough
> > to cross-merge from git if mercurial is the master. I am also
> > worried about our older release branches. Even though we may not
> > do a clean initial load into git for non-HEAD branches it should be
> > fairly easy to clean things up.
> >
> > * We would use commit scripts to trigger the merge operation immediately
> > upon a commit to either repository, plus a cron job once an hour to
> > catch any lost triggers.
> >
> > Only clean merges will auto-commit. A failure will require manual
> > intervention using Git, which should be trivial to handle as all the
> > data will be in the Git staging and master branches. If the
> > git->mercurial merge fails (only possible if a conflicting commit
> > is made to git and mercurial simultaniously), then the act of
> > resolving the conflict in the git + staging_branch_from_hg will
> > auto-correct on the next mirror operation from git to mercurial.
> > So no surgery within mercurial will be needed.
>
> It seems to me that allowing commits via both Git and Mercurial is
> going to be a ton of administrative work if problems arise. Being
> that we are a small group do we want to maintain both?
Ugh...Yeh I threw up in my mouth a little reading what Matt wrote. I
think we need to pick one and go with it.
There's administrative and policy decisions that need to be made when we
do decide on a system. Here's a quick brainstorm:
-Distribution of src
-contrib/crypto
-branching policy
-import into base src or use a package
I'm sure more things will surface, but I think we need to solidify
which one it's going to be first. I'm still suggesting we go with git
and make a hg repo available for read-only access.
I guess what I'm really trying to say is...Let's decide on a system so
we can move on and make the policy decisions we'll need to make with a
new system.
--Peter
Attachment:
pgp00002.pgp
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pgp00002.pgp
Type: application/octet-stream
Size: 197 bytes
Desc: "Description: PGP signature"
URL: <http://lists.dragonflybsd.org/pipermail/kernel/attachments/20081027/26adffb6/attachment-0020.obj>
More information about the Kernel
mailing list