Dotted revno "algebra"

[Stephen J. Turnbull]

Andrew Cowie writes:
 > On Mon, 2010-05-03 at 11:21 +1000, Martin Pool wrote:
 > > I think that users see the dotted revnos basically as near-opaque
 > > tokens for revisions.  They're not inherently useful or meaningful to
 > > anywhere near the degree that mainline revnos are.

By now, it's quite clear that many users don't see them as opaque.

 > I gotta say, after all this time, I can't help but wonder why not just
 > use integers.

Well, for one thing, they can't be stable in distributed operation.
As soon as there are concurrent commits, purely sequential revnos will
conflict, and this will have an impact on mainline numbering for at
least some users.  This is something I detest about Mercurial; on
several occasions, somebody has fingered a particular revision as the
one that injected a bug, and it turns out that there were concurrent
commits and in my repo it has a different revno.

While I myself think in terms of an algebra of patches and strongly
prefer counting back from heads and merge points to revnos of any
kind, it's quite clear that many users have a model of version control
that makes within-branch sequencing important.  Stable sequencing is
especially important on public branches, including the mainline, and
dotted revnos provide workably stable sequencing for most (all?) 
workflows that Bazaar can handle.

 > [Use of dotted revnos] implies there is some meaning to those
 > numbers that is useful to me. Not sure that in practise it is
 > useful at all. It is like Martin says, I think: just a token.
 > 
 > And, as someone said last week, if I need to see the further
 > relationship between what merged where, I use bzr-gtk.

Maybe the revision identification feature could be moved to a hook, so
that (e.g.) dotted revnos could be used or not, etc.