[RFC] branch --bind

[Martin Pool]

Oh, another consideration to add:

bzr has forward-counting revision numbers, which are nice in several
ways but as everyone ought to know local to a particular branch.  (And
yes, specifically local to a branch not to a wt or repository.)

It can cause confusion if it's not clear what context a number is
being looked up in.  Bound branches are an opportunity for this kind
of confusion, and it has caused some debate in things like 'update
-r'.

Simple checkouts don't seem to have this problem because their branch
is always used.

An object that always treats its master branch pointer as canonical is
also safe - this is a heavyweight checkout I believe - but it makes
all operations connect to the master or fail, so is not useful on slow
or intermittent networks.

A bound branch that is constrained to be a proper left-hand ancestor
of the parent shouldn't cause confusion.  This could be ok if there
were no local commits and the master followed append_revisions_only
behaviour.  (There may be other edge cases?)

A bound branch with local commits or with other ways to create
arbitrary divergence from the master has this problem strongly.  At
this point I'd say you might as well just have a separate branch.

-- 
Martin <http://launchpad.net/~mbp/>