🚀 go-pugleaf

In article <2400046@otter.hpl.hp.com> sfk@otter.hpl.hp.com (Steve Knight) writes:

   I wrote in an earlier message (amongst other topics) about the
   omission of full lexical binding from Algol 68, comparing the
   Algol68 policy with that of languages such as Lisp and Standard ML.

   Charles comments: Yes, but all these languages pay a high run-time
   > overhead for the facility.

   I'd like to understand this claim.  

   From my perspective, as a programmer in 1991, the implementation of
   lexical binding only involves a cost in two circumstances.  ...

   The second situation is relevant.  In the case where a procedure is
   not returned as a result but the compiler is unable to deduce this
   (without inter-procedural reasoning) there is an associated cost.

What's the associated cost? 

   An example would typically be uses of "maplist" which maps a
   procedure of type A -> B across a list of type list(A) to produce a
   result of type list(B).  When "maplist" is used, the compiler may
   not know enough about maplist to 'realise' that the lexical
   variables captured by the procedure can be implemented cheaply.

How would it matter if the compiler could figure out that "maplist"
doesn't return a procedure? I don't see anything about "maplist" that
would allow lexical variables captured by its procedure argument to be
implemented cheaply. 

-Mark
--

Mark Friedman
MIT Artificial Intelligence Lab
545 Technology Sq.
Cambridge, Ma. 02139

markf@zurich.ai.mit.edu