Adam Megacz writes:

 > So, I guess the conclusion here is that in its current form, darcs is
 > going to suffer from a lot of false dependencies

I don't see why you come to that conclusion.  You're assuming that
Darcs just gives up if a linearly ordered attempt to "commute out" the
patch that you want to unpull fails.  That simply doesn't conform to
my experience.  Darcs can work quite hard for only a couple dozen
patches.  (I'm not claiming that Darcs doesn't miss any.)

 > Anecdotally, I've run into this quite a few times now; at least now I
 > understand why darcs' notion of "depends on" is so much more
 > conservative than one would hope.

It would help if you would present (some of) the anecdotes.  I'm not
convinced that you have found cases where Darcs fails to commute
independent patches.  There are a number of places where Darcs's
decisions about dependency and conflict are quite logical, but run
counter to naive intuition (mine, at least).

OTOH, and perhaps more likely, you have found such an example that you
can work out in your head, and it could be added as a heuristic to
Darcs's commute search algorithm.

 > Are there any plans for darcs' future evolution which might change
 > this?

Yes.  Search the archives for "conflictors" for some of the
discussions that Darcs users would be likely to understand offhand.
There's also a separate list called darcs-conflicts, but that list
requires careful study (hint: there are a couple dozen posts devoted
to refinements of notation).

 > Perhaps darcs could offer a command which would let the user
 > explicitly ask it to work very hard (ie O(n^2) or worse)

Is O(n!) "hard enough" for you? ;-)  I don't know if anybody has a
formal proof, but a lot of people have conjectured that in some
situations Darcs already does "go exponential".

_______________________________________________
darcs-users mailing list
[email protected]
http://lists.osuosl.org/mailman/listinfo/darcs-users

Reply via email to