Issue CLOS-CONDITIONS-AGAIN Writeup

Date: Fri, 8 Jun 90 19:55 EDT

From: David A. Moon <Moon@STONY-BROOK.SCRC.Symbolics.COM>

Subject: Issue: CLOS-CONDITIONS-AGAIN (Version 6)

To: X3J13@mcc.com

Message-ID: <19900608235525.3.MOON@KENNETH-WILLIAMS.SCRC.Symbolics.COM>


This version includes the amendment that was passed at the X3J13 meeting today.


Issue:         CLOS-CONDITIONS-AGAIN


References:    ANSI CL draft specification, page 5-4


Related issues: CLOS-CONDITIONS


Category:      ADDITION and CHANGE


Edit history:  30-Apr-90, Version 1 by Moon (for Pitman)

               30-Apr-90, Version 2 by Moon (deal with slots)

	        2-May-90, Version 3 by Pitman (minor wording clarifications)

                2-May-90, Version 4 by Moon (final check for typos)

                5-Jun-90, Version 5 by Moon (mention multiple inheritance)

                8-Jun-90, Version 6, by Moon (as amended at X3J13 meeting)


Problem description:


  The condition system should not be too tightly integrated into CLOS, for

  two reasons:  Some implementations already have a native condition system

  that is not based on CLOS, and it should be possible to integrate the

  native conditions and the ANSI CL conditions.  Some people would like to

  define an ANSI Common Lisp subset that does not contain CLOS but does

  contain conditions.


  The problem areas are the use of DEFCLASS, MAKE-INSTANCE, and DEFMETHOD

  to define and create conditions, rather than using more abstract macros

  that conceal the implementation of conditions in terms of CLOS, and

  exposure of the implementation of condition slots as CLOS slots.  If user

  code was written in a more abstract way, it could run in a subset language

  that did not contain CLOS.


  This is Symbolics issue #7.


Proposal (CLOS-CONDITIONS-AGAIN:ALLOW-SUBSET):

	  

  1. Specify that conforming code must use DEFINE-CONDITION (not DEFCLASS)

  to define conditions, and MAKE-CONDITION (not MAKE-INSTANCE) to make

  conditions.  (These two operators already exist in ANSI Common Lisp.)


  2. Specify that conforming code must use the :REPORT option of

  DEFINE-CONDITION (not DEFMETHOD for PRINT-OBJECT) to define a condition

  reporter.


  3. Specify that conforming code must not use SLOT-VALUE, SLOT-BOUNDP,

  SLOT-MAKUNBOUND, or WITH-SLOTS on condition objects.  Instead it must

  call the accessor functions defined by DEFINE-CONDITION.


  4. Clarify that this proposal does not rule out the use of multiple

  parent-types in DEFINE-CONDITION.


Rationale:


  1. The reasons are two:

     - permit flexibility in making the native class system compatible

       with the ANSI CL one

     - permit dialects that don't want CLOS to be available to run most

       common condition-related code.


  2. The reasons are three:

     - isolate the report part without forcing the user to deal with the

       case of *PRINT-ESCAPE* being true.

     - keep a CL subset from having to implement DEFMETHOD to get this

       important functionality.

     - don't define two ways to do the same thing.


  3. Using the existing accessor functions seems more conservative than

     requiring non-CLOS subsets to implement an ersatz SLOT-VALUE function.


     Note also that the requirement here (in #3) is on conforming -code-.

     An -implementation- can permit the use of SLOT-VALUE, SLOT-BOUNDP,

     SLOT-MAKUNBOUND, and/or WITH-SLOTS as an extension and still be 

     conforming.  Programs which used such an extension would not be

     conforming and might not be portable to all implementations.


  4. There is no substitute for multiple inheritance, and given the

     limited set of operations that can be performed on conditions, it

     is easy to fake it.


Current practice:


  1. DEFINE-CONDITION and MAKE-CONDITION are already in the language.


  2. The :REPORT option to DEFINE-CONDITION exists.


  3. Some implementations support use of WITH-SLOTS and some do not.

     Programs that use WITH-SLOTS are not yet portable.


  4. DEFINE-CONDITION is already specified to support multiple inheritance.


Cost to Implementors:


  Trivial.


Cost to Users:


  They have to write their programs in terms of the condition abstractions

  if they want them to be portable.


Cost of non-adoption:


  Condition-using programs will not be portable to Common Lisp subsets that

  don't have CLOS.


Performance impact:


  None.


Benefits:


  See rationale.


Esthetics:


  Abstraction is more esthetic than revealing the implementation.


Discussion:


  Pitman and Moon support this proposal.