Re: [Monotone-devel] Re: user-friendly hash formats, redux

[derek]

On Wed, 8 Dec 2004, Bruce Stephens wrote:

> "John S. Yates, Jr." <address@hidden> writes:
>
> [...]
>
> > I like [developer id, UTC of commit].  Assuming that developers do
> > not spoof their identities (hence cannot be committing into multiple
> > archives at precisely the same instant) this should be globally
> > stable.  It has connotive value and UTC values are ordered.  It is
> > true that given such a [developer id, UTC of commit] pair it is
> > impossible to predict the next pair.
>
> Or the previous pair.  So really it's more or less opaque, so I'm
> going to end up doing cut and paste to actually use it, aren't I?  (In
> which case hashes seem OK, too: they've got no whitespace or
> punctuation, so cut and paste works rather nicely.)

Currently, the selector syntax allows for keyname/timestamp. A very 
slight extension to this might be to allow keyname/timestamp/offset
where offset might be something like -n or +n to indicate the nth
parent or child respectively.

This seems like it should be relatively easy to do. Additionally, for 
commands like diff in particular, perhaps allowing just an offset as one 
of the two revision id arguments would suffice. The implication would be 
that the offset applies to the other, more fully specified, revision. So 
for example, I could do

diff -2

to get a diff of my current working copy against it's grand-parent 
revision. or

diff f345 +2

to diff revision f345 against it's grandchild revision.

This would probably have to fail in the face of forks and merges but that 
may not be so bad. It may also be possible to say something like derek/-2 
to specify the author of the selected parent.

Cheers,
Derek