Re: Branches in CVS repository?

[Richard Earnshaw <rearnsha at gcc dot gnu dot org>]

On Wed, 2005-01-19 at 05:51, Mark Mitchell wrote:
> Nick Clifton wrote:
> > Hi Guys,
> > 
> >>>> The informal GCC convention has been "<org>-<name>" where "org" is the 
> > 
> > 
> >>> RH's internal convention includes a YYMMDD somewhere, which is
> > 
> > 
> >> I'd meant to allow arbitrary characters in "name", which would 
> >> certainly permit what you suggest.
> > 
> > 
> > But it is best to be explicit.  How about:
> > 
> >   <module(s)>-<org>-<reason>-<date>
> > 
> > eg:
> > 
> >   binutils-csl-ARM_development-050118
> > 
> > which may be a little bit wordy, but should avoid all possibility if 
> > name space collisions and should make the purpose of the branch obvious.
> 
> I don't quite understand the date part.  Is this because we might merge 
> our changes forward to a new basepoint (like the 050221 mainline 
> version) later?  Our usual practice has been to merge the delta between 
> 050118 and 050221 onto our branch, rather than create a new branch.  May 
> we make the date optional?

I suspect it's a consequence of a couple of things.

1) CVS gets progressively less efficient the more revisions there are
between the head of the trunk and the head of your branch via the
branch-point.

2) For experimental branches, it's easier to walk away from them and cut
a new branch with a unique tag if it includes the date.

Personally, I find the use of dates awkward.  At any time after the
initial creation point the date is not much better than a random number,
and remembering what set of changes is associate with what date is
nigh-on impossible.

So I think that beyond the basic rules of <module>-<org> at the start of
tags (I'd add a class -- branchpoint, branch, release etc --  there as
well, but that's me) it really ought to be up to the owner of the
branch, since the prefix will always be sufficient to keep the tags from
interfering.

R.