macros expanding into definitions

["Guillermo J. Rozas" <gjr@martigny.ai.mit.edu>]

   Posted-Date: Tue, 01 Dec 92 13:43:03 EST
   Date: Tue, 01 Dec 92 13:43:03 EST
   From: John D. Ramsdell <ramsdell@linus.mitre.org>

   > Personally, I would favor a semantics for internal definitions that is
   > closer to toplevel definitions:
   > 
   > 1) Definitions are evaluated in order.
   > 2) They can be intermixed with non-definitions.
   > 3) A redefinition of a variable is legal and equivalent to a set!

I don't particularly object to this, but there are cases that have not
been specified, namely

    (define a 23)

    (define (foo)
      (write a)
      (define a 44)
      (write a))

   (foo)

Do you get an error for the first write?
Does it print 23?

A more complex example:

    (define a 23)

    (define (foo)
      (set! a 56)
      (write a)
      (define a 44)
      (write a))

    (begin (foo) (write a))

Is the assignment in error?
Does the last write print 56 or 23?

Personally I prefer that the first example be in error and the second
print 23 at the end, since, to me, the first would be rewritten as

    (define a 23)

    (define (foo)
      (let ((a <unspecified>))
        (write a)
	(set! a 44)
	(write a)))

    (foo)

while the second would be rewritten as

    (define a 23)

    (define (foo)
      (let ((a <unspecified>))
	(set! a 56)
	(write a)
	(set! a 44)
	(write a)))

    (begin (foo) (write a))