issue LOAD-TIME-EVAL, version 9

[sandra%defun@cs.utah.edu (Sandra J Loosemore)]

This version incorporates the amendments made at the Hawaii meeting.

Forum:		Compiler
Issue:          LOAD-TIME-EVAL
References:     #, (p. 356),  (EVAL-WHEN (LOAD) ...) (p. 69-70)
		issue SHARP-COMMA-CONFUSION
Category:       ADDITION
Edit history:   06-Jun-87, Version 1 by James Kempf
                17-Jul-87, Version 2 by James Kempf
                12-Nov-87, Version 3 by Pitman (alternate direction)
                01-Feb-88, Version 4 by Moon
                  (from version 2 w/ edits suggested by Masinter)
                06-Jun-88, Version 5 by Pitman
                  (fairly major overhaul, merging versions 3 and 4)
                21-Sep-88, Version 6 by Moon (stripped down)
		17-Oct-88, Version 7 by Loosemore (change direction again)
		30-Dec-88, Version 8 by Loosemore (tweaks)
		23-Jan-89, Version 9 by Loosemore (amendments)


Problem description:

 Common Lisp provides reader syntax (#,) which allows the programmer
 to designate that a particular expression within a program is to be
 evaluated early (at load time) but to later be treated as a constant.
 Unfortunately, no access to this capability is available to programs
 which construct other programs without going through the reader.
    
 Some computations can be deferred until load time by use of EVAL-WHEN,
 but since EVAL-WHEN must occur only at toplevel, and since the nesting
 behavior of EVAL-WHEN is quite unintuitive, EVAL-WHEN is not a general
 solution to the problem of load-time computation of program constants.


Proposal (LOAD-TIME-EVAL:R**2-NEW-SPECIAL-FORM):
    
 Add a new special form, LOAD-TIME-VALUE, which has the following
 contract:

   LOAD-TIME-VALUE form &optional read-only-p	[Special Form]


   LOAD-TIME-VALUE provides a mechanism for delaying evaluation of <form>
   until the expression is in the "runtime" environment.  

   If a LOAD-TIME-VALUE expression is seen by COMPILE-FILE, the compiler
   performs normal semantic processing such as macro expansion but
   arranges for the evaluation of <form> to occur at load time in a null
   lexical environment, with the result of this evaluation then being
   treated as an immediate quantity at run time.  It is guaranteed that 
   the evaluation of <form> will take place only once when the file is 
   loaded, but the order of evaluation with respect to the "evaluation" 
   of top-level forms in the file is unspecified.

   If a LOAD-TIME-VALUE expression appears within a function compiled
   with COMPILE, the <form> is evaluated at compile time in a null lexical
   environment.  The result of this compile-time evaluation is treated as 
   an immediate quantity in the compiled code.  

   In interpreted code, <form> is evaluated (by EVAL) in a null
   lexical environment, and one value is returned.  Implementations which
   implicitly compile (or partially compile) expressions passed to
   EVAL may evaluate the <form> only once, at the time this
   compilation is performed.  This is intentionally similar to the
   freedom which implementations are given for the time of expanding
   macros in interpreted code.

   Note that, in interpreted code, there is no guarantee as to when
   evaluation of <form> will take place, or the number of times the
   evaluation will be performed.  Since successive evaluations of the
   same LOAD-TIME-VALUE expression may or may not result in an evaluation
   which returns a "fresh" object, destructive side-effects to the
   resulting object may or may not persist from one evaluation to the
   next.  It is safest to explicitly initialize the object returned by
   LOAD-TIME-VALUE, if it is later modified destructively.

   Implementations must guarantee that each reference to a
   LOAD-TIME-VALUE expression results in at least one evaluation of its
   nested <form>.  For example,
     (DEFMACRO CONS-SELF (X)
         `(CONS ,X ,X))
     (CONS-SELF (LOAD-TIME-VALUE (COMPUTE-IT)))
   must perform two calls to COMPUTE-IT; although there is only one
   unique LOAD-TIME-VALUE expression, there are two distinct references
   to it.

   In the case of a LOAD-TIME-VALUE form appearing in a quoted expression 
   passed to EVAL, each call to EVAL must result in a new evaluation of 
   <form>.  For example,
     (DEFVAR X 0)
     (DEFUN FOO () (EVAL '(LOAD-TIME-VALUE (INCF X))))
   is guaranteed to increment X each time FOO is called, while
     (DEFUN FOO () (LOAD-TIME-VALUE (INCF X)))
   may cause X to be evaluated only once.

   The READ-ONLY-P argument designates whether the result can be considered
   read-only constant. If NIL (the default), the result must be considered 
   ordinary, modifiable data. If T, the result is a read-only quantity
   which may, as appropriate, be copied into read-only space and/or shared
   with other programs. (Because this is a special form, this argument is
   -not- evaluated and only the literal symbols T and NIL are permitted.)


Rationale:

   LOAD-TIME-VALUE is a special form rather than a function or macro 
   because it requires special handling by the compiler.

   Requiring the compiler to perform semantic processing such as macro
   expansion on the nested <form>, rather than delaying all such processing
   until load time, has the advantages that fewer macro libraries may need
   to be available at load time, and that loading may be faster and result
   in less consing due to macroexpansion.  If users really want to delay
   macroexpansion to load time, this can be done with an explicit call to
   EVAL, e.g.
  
    (LOAD-TIME-VALUE (EVAL '(MY-MACRO)))
    
   Allowing the same LOAD-TIME-VALUE to cause its nested <form> to be
   evaluated more than once makes simplifies its implementation in
   interpreters which do not perform a preprocessing code walk.  It also
   makes the rules for the time of its processing analogous to those
   for macro expansion.

   This proposal explicitly does -not- tie LOAD-TIME-VALUE to the #,
   read macro.  Doing so would be an incompatible change to the definition
   of #, (which is reliably useful only -inside- quoted structure,
   while LOAD-TIME-VALUE must appear -outside- quoted structure in a
   for-evaluation position).

   The requirement that LOAD-TIME-VALUE expressions be evaluated once per
   reference (rather than once per unique expression) prevents problems 
   that could result by performing destructive side-effects on a value 
   that is unexpectedly referenced in more than one place.


Current Practice:

   This is an addition to the language and has not yet been implemented.


Cost to Implementors:

   In compiled code, (LOAD-TIME-VALUE <form>) is similar to 
   '#,<form>.  Most implementations can probably make use of the same 
   mechanism they use to handle #, to handle LOAD-TIME-VALUE.  Note that
   #, does not currently provide a mechanism for dealing with 
   non-read-only-ness.

   Implementing LOAD-TIME-VALUE in the interpreter should be fairly
   straightforward, since one simply needs to evaluate the <form> in the
   null lexical environment.  Implementations that use a preprocessing
   code walk in the interpreter to perform macro expansion could process
   LOAD-TIME-VALUE forms at that time.

   Some code-walkers would have to be taught about this new
   special form. Such changes would likely be trivial.


Cost to Users:

   Some code-walkers would have to be taught about this new
   special form. Such changes would likely be trivial.


Benefits:

   Users are given a mechanism that to force evaluation to be delayed 
   until load time that does not rely on a feature of the reader.


Discussion:

   Earlier versions (up to version 7) of this proposal stated that
   all semantic processing of the LOAD-TIME-VALUE form should be postponed
   until load time.  

   The semantics of LOAD-TIME-VALUE would be simplified considerably if
   the READ-ONLY-P argument were removed and destructive operations on
   the result of evaluating <form> prohibited.  However, some people feel
   that the ability to destructively modify the value is an essential
   feature to include.

   "Collapsing" of multiple references to the same LOAD-TIME-VALUE 
   expression could be allowed for read-only situations, but it seems 
   like it would be more confusing to make it legal in some situations 
   and not in others.

   A number of other alternatives have been considered on this issue, 
   including:

   - A proposal for a new special form that would force evaluation of
     the <form> to happen only once.  This was rejected because of
     implementation difficulties.

   - A proposal to add a function making the "magic cookie" used by #,
     available to user code.  The current proposal does not prevent such
     a function from being added, but this approach appeared to have
     less support than making the hook available as a new special form.

   - A proposal to remove #, entirely (issue SHARP-COMMA-CONFUSION).

   - A suggestion to change the behavior of (EVAL-WHEN (LOAD) ...).


Kent Pitman says:
   Although I'm willing to take multiple evaluation in the interpreter
   as a compromise position, I would like it mentioned in the discussion
   that this was only an expedient to getting this issue accepted at all,
   and that I'm not really happy about it. I have said that I think a
   number of our lingering problems (with EVAL-WHEN, COMPILER-LET, and
   this -- for example) are due to the presence of interpreters which do
   not do a semantic-prepass at a known time. If I had my way, we would
   require a semantic pre-pass and we would then be able to forbid
   multiple evaluations even in the interpreter.
   
-------