[R6RS] eval/environment proposal

[dyb at cs.indiana.edu]

Here is the promised eval/environment proposal.  I've also committed it
to the repository as draft/eval/eval.txt.

Kent

--------

Eval/environment proposal

1. The eval proceedure remains essentially as it is in R5RS:

   (eval x env)

   evaluates x (which must be an s-expression representing a valid
   expression) in env and returns its values.

2. Add a new "eval" library <import-phase>

   <import-phase> -> expand | run | eval

   In the terminology of SRFI 83, importing a library "for eval" is like
   importing it "for run" in that the library is invoked at run time.  It
   differs in that the library is not visited when the library is expanded
   but rather when the expression passed to eval is expanded.  If a library
   is imported both "for run" and "for eval" it is invoked only once at
   run time, and both ordinary code and eval'd code share the same set of
   library bindings.

3. Add a new syntactic form, "the-library-environment":

   (the-library-environment)

   If (the-library-environment) appears within a library form, it
   evaluates to an environment containing the set of bindings imported
   "for eval" into the library.

   If (the-library-environment) appears within an expression passed to
   eval, it evaluates to the same library that was passed to eval.

   Both the environment and its bindings are immutable.

--------

Examples (making assumptions about library syntax):

  (library foo
    (import (for r6rs run eval))
    (write (eval '(let ((x 3)) x))))

writes 3.

  (import (for r6rs expand run)
          (for (add-prefix (only r6rs car cdr cons null?) "eval:") eval))

imports r6rs for expand and run and only car, cdr, cons, and null? for
eval, as eval:car, eval:cdr, eval:cons, and eval:null?.