[darcs-users] RE: darcs'ing fptools

[Simon Marlow]

On 13 October 2005 14:18, John Goerzen wrote:

> On Thu, Oct 13, 2005 at 08:17:14AM -0400, David Roundy wrote:
>> On Thu, Oct 13, 2005 at 10:08:27AM +0100, Simon Marlow wrote:
>>> immediately.  We would disable commits to the appropriate parts of
>>> the CVS repo.
>> 
>> Actually, I think you might consider starting with a one-way sync
>> from CVS to darcs, you could then work out some of the bugs in the
>> darcs configuration and get some of the users used to pulling with
>> darcs, and later switch the direction of the one-way sync to allow
>> modification on the darcs side of things.  I imagine that way you
>> could pretty easily (and quickly) switch back to syncing the other
>> direction (and switch to a backup) if things go wrong somehow.
> 
> Yes, I would echo that sentiment.  It is also easier to accomplish,
> since once the repo exists, a re-sync is as easy as just re-running
> tailor.

Ok, sounds good to me.

> Simon, one other question: what about branches?  Normally, tailor will
> just convert over everything on HEAD.  When CVS branches get merged to
> HEAD, the history from the other branches doesn't come along to HEAD,
> so some detailed history could be lost.  I'm assuming that's OK with
> you? 

The branches that were briefly forks of the HEAD and were merged back
are fine, the history is always in CVS, it wouldn't be much of an
improvement to have a darcs repository for these branches.

There are branches that are active forks of HEAD, and for these I think
we'll want to have darcs repositories.  These include:

   pluggable-1-branch
   ghc-assoc-branch
   arity-anal-branch
   CgCmmBranch

If anyone else has an active branch of any part of the GHC sources,
please speak up (these are hard to find).  Also shout if any of the
above branches are not active any more.

Additionally, we want a repository for the STABLE branch, which is
currently

   ghc-6-4-branch

> Is there more than one branch of fptools in active use today?
> 
> If so, I'm not quite sure what the best way to copy them over is.
> Obviously with darcs, one wants to have shared history as much as
> possible, but if I tailor the two separate branches, there will be no
> shared history (to darcs) at all, right?

That would be unfortunate.  I wonder if it can be done, though: if you
can create a darcs repository representing the CVS repo at the branch
point, by unpulling all the relevant patches, then get tailor to pull
over everything on the branch, you'd have a proper branch repository.
Probably the difficult bit is creating the branch-point repo.

Cheers,
	Simon