Re: [Monotone-devel] Re: mtn add weirdness

[Nathaniel Smith]

On Mon, Aug 07, 2006 at 11:36:41AM -0400, Jonathan S. Shapiro wrote:
> What is really being committed is a *binding* from a name (a
> project-relative path) to a state (the new content). The current
> behavior is effectively refusing to form the name that must be bound. In
> principle, there exists no sensible formulation of a (name, state) pair
> that omits any portion of the path leading up to the file name.

The name of a file is a (parent directory id, basename string) pair.
Formulating such a name requires that the directory exist, and that
the basename be otherwise unused.  In the general case, the operations
that create such a situation may be arbitrarily complex, e.g.,
requiring arbitrarily long rename chains be included, deletion of
apparently unrelated files (like in the example I gave up-thread),
etc.

That's why we don't do this normally; "when in doubt refuse to guess"
and all that.

I think the question at hand is probably, should there be a special
case exactly for the situation where the dependencies consist only of
directory adds.

-- Nathaniel

-- 
"Of course, the entire effort is to put oneself
 Outside the ordinary range
 Of what are called statistics."
  -- Stephan Spender