Issue ASSOC-RASSOC-IF-KEY Writeup

Issue:        ASSOC-RASSOC-IF-KEY

References:   ASSOC-IF (p280), ASSOC-IF-NOT (p280), RASSOC-IF (p281),

              RASSOC-IF-NOT (p281)

Category:     ENHANCEMENT

Edit history: 22-Apr-87, Version 1 by Pitman

              20-Nov-87, Versions 2,3 by Masinter

              23-Nov-87, Version 4 by Masinter


Problem Description:


The descriptions of ASSOC-IF, ASSOC-IF-NOT, RASSOC-IF, and RASSOC-IF-NOT do not

mention a :KEY option, although ASSOC and RASSOC have one. 


This is often reported as an inconsistency in Common Lisp.


Proposal (ASSOC-RASSOC-IF-KEY:YES):


Allow a :KEY keyword for ASSOC-IF, ASSOC-IF-NOT, RASSOC-IF, and RASSOC-IF-NOT.

If not supplied, it should default to #'IDENTITY as do the :KEY keywords for

other -IF and -IF-NOT functions. The function, as with the :KEY argument for

ASSOC and RASSOC, is applied to the CAR of the pair in the association list for

ASSOC-IF and ASSOC-IF-NOT and the CDR of the pair for RASSOC-IF and

RASSOC-IF-NOT.


Documentation impact:


A better description of the intent might be to say that the car /contains/ the

key of the association, and by default the car /is/ the key of the association.


Example:


(assoc-if #'zerop pathnames :key #'pathname-version)


could be used to search a list indexed by pathnames finding one with zero

version. 


Rationale:


This is an inconsistency in the language that is simple to fix.


Current Practice:


Symbolics implements :KEY for the -IF and -IF-NOT assoc functions. Others follow

the book and allow :KEY only for ASSOC.


Cost to Common Lisp implementors:


A small amount of additional code is necessary to support this in

implementations not already offering it as an extension.


Cost to Common Lisp users:


The change is essentially upward compatible with user code.


Benefits:


This would make the set of -IF and -IF-NOT functions be more regular in their

calling conventions.


Aesthetics:


All the other -IF and -IF-NOT variations of list operations omit the :TEST and

:TEST-NOT keywords, but allow :KEY. For example, consider the family of MEMBER,

MEMBER-IF, and MEMBER-IF-NOT. Although this introduces additional mechanism, it

does so in a way that probably makes it easier to think about which functions do

what, so it would likely be seen as a simplification.


Discussion:


The omission of :KEY in this situation in CLtL was probably an oversight.


The cleanup committee supports this proposal.