On Sun, Oct 20, 2013 at 6:05 PM, Gabriel Reid <[email protected]> wrote: > > The only use case I can see where using a different underlying > combiner implementation would be useful is if the instantiation of > U is very costly, and so we don't want to create a new U for every single > V. I can envision any situation where this would actually be the case, so > it doesn't seem worth doing it like this (until this use case comes up). >
And now just after writing this and thinking about it, I realize that this isn't even feasible -- a combiner's operation must be commutative, and converting from V to U isn't commutative, so it's not even technically feasible. So yet more reason to not try to do anything tricky here. - Gabriel
