Yes, but a transform is not necessarily a projection.
typedef adapted_iterator<C::iterator> iterator; make_pair(iterator(C.begin()), iterator(C.end())) == make_pair(iterator(C.begin()), iterator(C.end()))
That was my initial instinct, too. but Tim is pointing out that the algorithms can all be raised to a higher level of generality without loss of efficiency, and I am now inclined to agree that there's no reason they need to reply on this property.
Good point. I concede :) I think this property is still valid in many cases - and I think the expected behavior of range-generating functions. I suppose this could be defined as a concept, but I'm not sure how many templates take range-generating functions as parameters. Probably none in the BGL. That's not the patch I (or Tim, I think) had in mind. I'm talking
about changing the algorithms so they no longer rely on out_edges(u,g) == out_edges(u,g) and updating the documentation to make it completely clear that the property just stated is not part of the concept requirements.
I know, but it was mentioned. It might also be better applied in cases where an iterator from the range is discarded. Andrew Sutton andrew.n.sutton@gmail.com