RFC, poke at 'shallow' branches

[Ian Clatworthy]

John Arbash Meinel wrote:

> At a minimum we would have to fix the 'cannot commit to a stacked
> branch' bug.

And that's a good thing.

Either way, +1 on the idea of trying out "shallow branches" early in 2.2.

> If we wanted it to work nicely, then we would want to figure out how to
> not connect on every action (like lightweight checkouts currently do).
> Even for stuff like 'status' that has all the needed information locally
> (and likely 'diff' could also be purely local under this situation).
> 
> I believe Robert has always advocated that lazy connecting will lead to
> late-failures. (Better to know as soon as I can't connect, rather than
> once I really need it... Or something to that effect)

I feel connecting when the data is already local violates the Principle
of Least Surprise. Many users report bugs and/or grumble on IRC when
they realise this is happening.

I suspect part of the issue is the bound-branch vs heavyweight-checkout
conceptual model debate. When *I* think about a bound branch, I expect
95% of operations to be local and for a small number (commit, tag,
others?) to contact the remote site. When I think about a checkout,
contacting the associated branch is a reasonable choice more often. Even
so, I'd err against doing it if the data was already local.

Ian C.