1) I definitely agree with the herd. The order function goes first.
   Although I frequently think the convention is misguided, I'd prefer
   having all parametric functions passed in the same way.

2) can the API (and ref impl) be factored to provide a lower-level
   interface to the vector sorts? I've been finding that an interface
   that provides swap indices to a parametric function would be really
   useful. I've been working on a vector version of SRFI-1 for a while
   and I keep finding that I frequently want both interfaces: one that
   provides indices (very useful when working on parallel vectors and
   other cleverness) and one that operates directly. Since the direct
   operation version is implementable in terms of the indexed version,
   it's certainly been an LOC win anyway.

Otherwise this is very nice work. Since it is a Shivers production, I
have no qualms about late changes. Heck it's not even two months
overdue, yet ;)

david rush
--
It seems that the folks who designed the C++ template system ...
reinvented the wheel. The C++ lug nuts have been praised, but the
squarish shape of the C++ wheel is often criticized.
	-- Dr. William Clinger (on comp.lang.scheme)