Issue MULTIPLE-VALUE-SETQ-ORDER Writeup

Issue:            MULTIPLE-VALUE-SETQ-ORDER

References:       MULTIPLE-VALUE-SETQ, SYMBOL-MACROLET

Related issues:   SYMBOL-MACROLET-SEMANTICS

                  SETF-SUB-METHODS

                  SETF-OF-VALUES

Category:         CLARIFICATION

Edit history:     V1,  1 Jun 1990, Sandra Loosemore

                  V2,  6 Jun 1990, Sandra Loosemore

                  V3, 12 Feb 1991, Sandra Loosemore


Problem description:


Proposal SYMBOL-MACROLET-SEMANTICS:SPECIAL-FORM extended

MULTIPLE-VALUE-SETQ to allow symbols naming symbol-macros to appear in

the "variables" argument to MULTIPLE-VALUE-SETQ.  The intent was that 

this would be treated the same way as with SETQ or SETF.  However, the

proposal did not address order-of-evaluation issues.


When dealing only with variables, the order in which

MULTIPLE-VALUE-SETQ performs the assignments is not relevant.  On the

other hand, the storing form that results from applying the

SETF method of the macroexpansion of a symbol-macro can be fully

general.  It might include references to other variables or "places"

named by other symbol-macros in the variable list.  The question is,

which other variables or "places" in the list have already been set

when the storing form is evaluated?


For example, consider something like


  (defun foo (v i)

    (symbol-macrolet ((element  (svref v i)))

      (multiple-value-setq (v element i) ...)))


Does the value corresponding to "element" get stored in the new

or old value of the vector "v", at the new or old value of the index

"i"?



Proposal (MULTIPLE-VALUE-SETQ-ORDER:LIKE-SETF-OF-VALUES):


  Clarify that


      (multiple-value-setq (<place1> ... <placen>) <value-producing-form>)


  is exactly equivalent to


      (values (setf (values <place1> ... <placen>) <value-producing-form>))


  under proposal SETF-OF-VALUES:ADD, with the additional constraint that 

  each of the <placei> must be a symbol.


  Rationale:


    Assuming VALUES is added as a SETF place, it is obvious to define

    of MULTIPLE-VALUE-SETQ (or MULTIPLE-VALUE-SETF) in terms of it.


Current Practice:


  Symbolics Genera 8.0 does both assignments and evaluation of place

  subforms in left-to-right order.

  Utah Common Lisp appears to do the assignments (and evaluation of 

  place subforms) right-to-left.

  Chestnut's Lisp-to-C translator implements proposal SETF-OF-VALUES.

 


Cost to Implementors:


  Assuming proposal SETF-OF-VALUES:ADD is adopted, minor.


Cost to Users:


  Probably nobody depends on any particular behavior here, since

  symbol-macros are a relatively new addition to the language.  This

  definition of MULTIPLE-VALUE-SETQ is fully upward-compatible with its

  definition in CLtL.


Cost of non-adoption:


  This part of the language specification remains vague.


Performance impact:


  The situation probably doesn't crop up enough for it to have a 

  significant impact.


Benefits:


  This part of the language specification becomes less vague.


Esthetics:


  Any code that depends on any particular order is likely to be

  pretty nasty.  However, defining MULTIPLE-VALUE-SETQ in terms of

  SETF of VALUES seems like the right approach to me.


Discussion:


  An earlier version of this issue (with three different proposals) was

  discussed at the June 90 meeting.

-------