Note that we may also have to have some conventions for canonical order for
cases where patches just commute.

If I follow Ganesh correctly, he was working on this last and that the sticking
point is Darcs 2 allowing stupid unpull tricks whereby a patch can have
different dependencies due to duplicates.

There are two problems, in fact.

The first is that calculating the minimal context for a single patch added to
the end of a repo (which already has the hashes of minimal contexts stored for
existing patches) is O(n^2) and this hurts in practice. I have some ideas for
speeding this up but I think it's a hard problem.

The second is the unpull tricks you can play using duplicate patches in darcs-2
semantics, as you mention. That problem can be solved in a reasonable manner,
but at the expense of only including primitive patch dependencies in the hash of
a patch.

I bump the release to 2.6 because I presume this is not going to happen
before darcs 2.5. Please adjust the milestone if you have a good reason
to do so.

More elaboration from Ganesh for what makes minimal context expensive: 
http://lists.osuosl.org/pipermail/darcs-users/2012-March/026406.html

Another issue perhaps: for current darcs send, you need not just a 
minimal context for a single patch, but for a single minimal context for 
a set of potentially unrelated patches.

Wouldn't that be a bit tricky? Naive algorithm might be: commute patches 
backwards to compute minimal context separately for each one of the 
patches, take union of the contexts, then commute each patch forwards 
until it fits into the union-context again.  :-/

(Minimal context still has lots of uses outside of send, if I remember 
correctly, just thinking about this over breakfast)

Ignore me, I was talking nonsense. I forgot that once you have the minimal 
context for the bottom-most patch, you're sorted.

The following patch sent by Guillaume Hoffmann <guillaumh@gmail.com> updated issue issue1514 with
status=resolved;resolvedin=2.10.0 HEAD

* resolve issue1514: send --minimize-context flag for send 
Ignore-this: 486b70607488643e092bb8f46cd046d4