[darcs-users] Git v. Darcs: megapatches and 'darcs shunt'

[Isaac Dupree]

On 01/13/11 08:42, Eric Kow wrote:
> On Wed, Jan 12, 2011 at 15:49:31 -0500, Isaac Dupree wrote:
>> What if "pull" could (maybe given a flag that you'd use with your
>> upstream) re-order all patches from upstream, new or old, as far
>> back in your history as possible?
>
> I'm always nervous about introducing new flags [1], but I like
> Isaac's variant of Alexander's idea:
>
> The hypothetical
>
>    darcs pull --reorder
>
> would then pull all patches in such a way that the ones which are
> local to the current repo go on top.

I'm nervous about it too.  What if you mix uses with and without --reorder?

Does --reorder always move all patches found in the remote repo to the 
bottom?  Even the older ones that were, of course, already in the local 
repo?  (And never reorder these patches with each other?  Or perhaps it 
should reorder them to match the remote repo's order?  Is that always 
possible if we have them adjacent, consistent, but in a different order?)

What about patches that originated in the local repo, but were 
eventually pulled into the remote repo -- will the user be confused that 
they're no longer kept on top?

Should reordering against another repo be an operation separable from 
pulling, since you can do it usefully without pulling any patches?

What about a user who habitually pulls from two different repos, not 
just one* -- the behavior is probably clear enough, just might be 
surprising.

*(say, a HEAD and a long-term branch, and pushes to the long-term branch 
after successfully merging the two or adding features.  There's probably 
an even more symmetric local workflow too: I'm working on a little 
Haskell game/simulation toy and I sometimes have a branch to hack on the 
backend model and a branch to hack on the fancy user interface and 
sometimes and another branch where I pull both changes and see if they 
work together, not just separately.  Workflows similar to that anyway.)

Is the behavior even possible to implement?  Can it be implemented in 
faster than O(n) in all patches in the history?

-Isaac