From: Jeff King <peff@peff.net>
Date: Tue, 03 Jul 2012 20:24:52 GMT
Subject: Re: [PATCH v4 00/18] Extending the shelf-life of "git describe" output
Message-ID: <20120703202452.GA24984@sigill.intra.peff.net>
In-Reply-To: <7vtxxotx2n.fsf@alter.siamese.dyndns.org>

On Tue, Jul 03, 2012 at 01:21:36PM -0700, Junio C Hamano wrote:

> Jeff King <peff@peff.net> writes:
> 
> > What happens if I set multiple flags? One or more of them will be
> > ignored (you _could_ try to establish a hierarchy, for example that
> > TREEISH is a superset of COMMITISH, but that could not handle the *_only
> > cases, which really are mutually exclusive). IOW, I think these are not
> > really flags but rather enum elements.
> 
> Yes, I was aware of that.  I counted five useful ones (see the new
> ones near the tip of 'pu') but there may be others, so you cannot
> fit in 2 bits and would need 3 bits.

I am not worried about the bits, only that adding a new 'enum
disambiguation_type' parameter would be a pain, as most places would
just be passing 0. But it would enforce the mutual exclusion at compile
time.

> I actually am thinking to move these bit assignments backto
> sha1_name.c and make them private, as get_sha1_tree() and friends
> are easier to understand public interface functions than
> get_sha1_typish(... I_WANT_COMMIT|I_WANT_COMMITTISH|...).

Yeah, I think that might be sane. If you really wanted to be flexible
anyway, you would let the caller pass a custom disambiguation function
(which I suspect describe would want for handling the "must come at this
point in the history" rule).

-Peff