Issue LOOP-FOR-AS-ON-TYPO Writeup

Status:         Proposal FIX-TYPO passed 12/91

Issue:  	       LOOP-FOR-AS-ON-TYPO

Reference:      X3J13/89-004 page 2-10, CLtL/II page 721, and dict-flow.tex

Category:       Clarify/fix-typo

Edit History:   Version 1  20-Aug-91 by JonL 



Problem Description:


 From the draft proposal .tex file, under the subsection {\bf for-as-on-list},

 the words:


     In the {\it for-as-on-list\/} subclause,

     the {\tt for\/} 

     or {\tt as\/} construct iterates over the contents of a 

     @type[list]. It checks for the

     end of the @type[list] as if by using

       @funref[endp].  


 were erroneously copied verbatim from the subsection {\bf for-as-in-list},

 rather than the intended wording from the Iteration Subcommittee.


 At one level, this prescription doesn't make sense;  by saying that 

 it is iterating over the "contents", rather than down the successive cdrs,

 of the list, there is confusion about the values assumed by the variable

 [which is subsequently said to take on the "tails" of the list.]


 In the spring of 1988, the Iteration Subcommittee suggested a change from

 the then existing practice of terminating both LOOP-FOR-AS-IN and 

 LOOP-FOR-AS-ON by simply a NULL test.  [Indeed, in the MacLisp days before

 CommonLisp, ENDP didn't exist; and the common use of NULL rather than ATOM

 was based on a performance hack for the PDP10 Architecture.]  So ENDP was 

 specified for LOOP-FOR-AS-IN because that style views the list as a sequence 

 having "contents" or elements; but ATOM was specified for LOOP-FOR-AS-ON

 since the only concern was the succession of list tails.  The LOOP-FOR-AS-ON

 case is intended to be more general -- and hence have a less restrictive

 termination test -- so that user-defined iteration over non-standard

 lists can be supported.  But the current draft specification completely

 and accidentally defeats this design.



Proposal (LOOP-FOR-AS-ON-TYPO:FIX-TYPO)


 Restore the original wording as intended by the Iteration Subcommittee

 as follows:


    In the {\it for-as-on-list\/} subclause,

    the {\tt for\/} 

    or {\tt as\/} construct iterates over a

    @type[list]. It checks for the

    end of the @type[list] as if by using

      @funref[atom].  



Editorial Impact:


 Very small.


Rationale:


 The evidence that this is a typo is in the Lucid Product Documentation

 of about the same time that 89-004 was abstracted from it; this product

 documentation has the "correct" wording, and the person responsible for

 doing the abstracting simply copied over the wrong, albeit very

 similarly worded paragraph.  The two paragraphs are close together

 in that text.


 Without this fix, there will surely be gratuitous variations as to

 how the termination testing is done for the LOOP-FOR-AS-ON case.



Current Practice:


 Some existing implementations still adhere to the MacLisp inspired

 version that used to be distributed from MIT, which simply used NULL

 for both the FOR-AS-IN and the FOR-AS-ON cases; some adhere to a

 later Common Lisp version that uses ENDP, at least for the FOR-AS-IN 

 case.  Harlequin has apparently implemented the broken semantics of 

 "X3J13/89-004 page 2-10".



Discussion:


 This typo has gone undetected for over two years, between January 1989 

 until about April 1991, when a fresh implementation of LOOP taken purely 

 "from the spec" produced anomalous results on LOOP-FOR-AS-ON termination 

 tests.  Lawrence G. Mayka <lgm@iexist.att.com> noticed the variation when 

 comparing Harlequin's LOOP implementation with some of the others.



 Jonl 11-Apr-91:  I know this (the use of ATOM for FOR-AS-ON termination) 

    was discussed and agreed upon; and I hope that the forum of discussion 

    was within X3J13, either the iteration subcommittee or the whole 

    committee.  I even recall changing Lucid's implementation after this 

    decision. [And an edit-history record of that change, along with the

    rationale cited above, still exists in the Lucid source code for

    Loop, dated 9-May-88.]