[Gnu-arch-users] Re: cvs interoperation question

[Miles Bader]

Robert Collins <address@hidden> writes:
> > Just use tla-cvs-sync, then you can synchronize often.
> 
> I've not used tla-cvs-sync (being a cscvs by myself :]) - does it
> preserve changeset boundaries in both directions? one?

It makes no attempt to deduce changeset boundaries from CVS commits,
but of course the other direction is easy -- it just replays arch
changesets one-by-one, and commits each to CVS individually.

For my main purpose (emacs) automatic changeset deduction is pretty much
a non-starter anyway, but I think even in the case where CVS committers
are disciplined enough to make _detection_ of CVS `changesets' possible,
attempting to preserve the boundaries in both directions would make
arch<->CVS synchronization quite a bit more difficult.  The reason is
that both systems require the tree to be `up to date' before committing,
and if you want to replay and commit changesets in _both_ directions,
you end up in a situation where you want to commit from a tree that's
only partially up-to-date.

With `halfway' operation like tla-cvs-sync uses, you can fully update
from CVS first, which gives you a free hand in the other direction (note
that the initial update from CVS is only commited to arch after all the
arch changsets have been committed to CVS).

-Miles
-- 
((lambda (x) (list x x)) (lambda (x) (list x x)))