Re: [Gnu-arch-users] Re: [BUG] microbranches: prism-merge vs multi-merge

[Stefan Monnier]

> I'd really like something like this, but I'm still nervous about how the
> result would interact with merging commands; for instance, if I was in the
> habit of star-merging from branch A to branch B, and then commited such a
> `continuation' changeset to A, what would happen the next time I star-merged
> from A to B?

If you use patch-based merging, I don't know.  With diff3-based merging, it
should make no difference since the "continuation or not" only influences
the way a given revision can be built, but not the end result
(i.e. it's basically an implementation detail).

One interesting twist:
let's say you have a micro-branch B that tracks branch A and let's say you
only ever commit "continuations" to B.  Then the archive in branch B will
contain a succession of patches against branch A.  It's likely that those
patches will all look very much alike, so it might be worthwhile to store
all those revisions (except the first) as a diff against the
previous patch: i.e. base-0 will be a changeset between revision A-0 and B-0,
then patch-1 will be a changeset between "the changeset base-0" and "the
changeset between A-1 and B-1", ...

OT1H it's only an implementation detail so it's rather pointless, but OTOH
it does match my mental image of the life of my local long-lived branches.
So maybe the archival format doesn't need to be changed, but some tools
could be provided to be able to watch the evolution of a long-lived branch
that way.


        Stefan