[rfc] creating clean diffs

[Martin Pool]

On  3 Jul 2006, John Arbash Meinel <john at arbash-meinel.com> wrote:
> Anyway, just something I wanted to see if anyone had any ideas on how to
> solve.

It depends a bit on whether you want to keep all the intermediate
commits of your work.  In general I like to, but sometimes (if I've
followed a convoluted path) I want to throw things away and just have
the final diff.  (And you can have tools like mq to help you do that.)

This relates to the "don't commit broken code" thread - I might commit
half-working stuff as a checkpoint planning to go and uncommit or throw
away the branch later.  I think it also shows how good it is that you
can just move away or discard branches without poluting your ongoing
namespace, etc.

Sometime post 1.0 I'd like to look at relations between revisions
indicating that they're rewritings of the same logical change.

To focus on the review aspect, rather than how you record the history:
it seems to me we ought to be able to produce a composed diff for all
the revisions other than the merge from bzr.dev -- perhaps directly from
a knit, or by getting all the diffs and composing them.  This might not
be something you could apply to any particular revision, but perhaps you
could use it with a fuzzy patch, and at any rate it might be good for
human reviewers.  We should try that sometime.

-- 
Martin