[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[no subject]

We should like the implementation of our subprocedures to be reasonably
transparent. In large systems (eg, Macsyma) there may be a large number
of function-like programs which are in fact implemented as macros.
Having to remember which are subrlike macros and which are true subrs
(and therefore MAPable) detracts from the elegance of the system. The
coding style resulting from such an environment is one which encourages
having two functions around which do the same thing but each of which
has [seemingly] arbitrary conditions on when it can be used.

The `obvious' reason why we can't APPLY or MAP a macro is that its functional
component expects to operate on a pointer to the whole macro form. eg,
SECOND's definition acts on (SECOND x) as a unit. (MAPCAR #'SECOND x) doesn't
work because there is no whole macro form anywhere. If SECOND is a displacing
macro, it may want to displace that toplevel cons with (CADR x). The trouble
in the MAPCAR case is that there is no toplevel cons to displace. The problem
is that this obvious reason is an artifact of the evaluation method chosen
and has nothing to do with high-level issues like getting problems solved.
It is, as RICH points out, frustrating to have to worry about such trivial 
details when you are trying to solve some real problem.

DEFSUBST is the wrong answer because its semantics are ugly and weak. 
Program-writing programs cannot reliably generate DEFSUBST forms because of
the problems of variable naming conflicts (nested lambdas with similar names,
quoted symbols, etc). A real effort needs to be made toward having a general
purpose code-walker/meta-evaluator which can correctly analyze the code
to be SUBSTed for. If we could make

	  (LIST 'X X ((LAMBDA (X) (LIST 'X X)) X)))


	(FOO A) => (LIST 'X A ((LAMBDA (X) (LIST 'X X)) A))

we'd be in much better shape than we are now. That would allow us to define
real open-coding techniques. We just don't have such support yet.

As an interim solution, we might want to look into what it takes to make
something like a (DECLARE (SUBRLIKEMACRO (name number-of-`args') ...)) work
so that #' could have help in deciding what to do. The compiler could in
theory turn 
		(DEFUN G0001 (X) (CDADR X))	
		(DEFUN BAR (X) (MAPCAR #'G0001 X))
		(DEFUN BAZ (L) (APPLY #'G0001 L))

I'm not sure if this is a good idea. I think really that the meta-evaluator
is the right thing to have. GJC has suggested for Macsyma that perhaps a single
meta-evaluator should exist which can drive both the compiler and the 
interpreter so that the semantics of code in the two are uniform...

In any case, I'd be interested in hearing others opinions on these ideas.