[Haskell-cafe] Wumpus World

[Richard A. O'Keefe]

On 27 Mar 2008, at 8:25 pm, Robert Wills wrote:

> This might also be relevant:
> http://web.engr.oregonstate.edu/~erwig/zurg/

But note that the Prolog code that they compared against was, um,
let's put this kindly, seriously naive.  For example,
(a) it has 36 SLOC.        You can do it naturally in 20.
(b) it has  8 predicates.  You can do it naturally in  4.
(c) it does lots of list munching, and that inefficiently.
     You can do it much more naturally with the only list being the  
solution.
     Indeed, a very minor rewrite from the natural code gives you a  
Prolog
     program where NO heap storage is allocated except the list of  
move names.
(d) It generates candidate solutions and then rejects ones that take too
     long.  It is easier and more natural to reject over-time paths  
before
     extending them to solutions, so the Prolog code they used for
     comparison is *structurally* inefficient.

30 years ago people were writing papers showing that Lisp was better  
than
very badly written Prolog.  Now they are writing papers showing that  
Haskell
is better than very badly written Prolog.  How things have changed!   
NOT.

Also note that the paper says
	"The most important feature of Haskell that supports
	 [the impression that Haskell is good at this kind of
	  thing]
	 is the availability of multi-parameter type classes..."
and that Haskell 98 had no multi-parameter type classes, which are
a pretty advanced part of the language for beginners to understand.