[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Issue: DEFPACKAGE (version 7)

Here's a "cleanup" of the cleanup, incorporating nits etc from
    sandra%defun@cs.utah.edu on Date: Thu, 13 Oct 88 14:34:04 MDT
    peck@Sun.COM on Mon, 17 Oct 88 17:19:14 -0700
    Moon@STONY-BROOK.SCRC.Symbolics.COM on Wed, 26 Oct 88 22:58 EDT
    masinter.pa@Xerox.COM on 31 Oct 88 10:37 PST
as well as some further discussion and re-wordings by me.  In particular,
the discussion contains a new suggestion to replace the mnemonic of the 
"7 Extremely Randoms".

-- JonL --


Issue:         DEFPACKAGE

References:    CLtL section 11.7.
               Issue: IN-PACKAGE-FUNCTIONALITY
               Issue: MAKE-PACKAGE-USE-DEFAULT
               Issue: PACKAGE-DELETION

Category:      ADDITION

Edit history:  Version 1, 12-Mar-88, Moon
               Version 2, 23-Mar-88, Moon, changes based on discussion
               Version 3, 27-Sep-88, JonL 
                (remove :import, :shadowing-import; allow :export to work on
                 imported and inherited; update references to in-package, etc.)
               Version 4,  1-Oct-88, Masinter
               Version 5, 6-Oct-88, Moon
               Version 6, 6-Oct-88, JonL (little nits)
               Version 7, 2-Nov-88, JonL 
		 (incorporate further discussion; simplify handling of
		  syntactic errors; specify ordering constraints).

Problem description:

Many users incorrectly think that package operations can be performed
in any order.  CLtL (p.192) contributes to this misconception.
Programmers need direction on the ordering constraint, especially for
creating packages, since doing things out of order can lead to
confusing or even intractable problems.

If the definition of a package is scattered throughout a program as a 
number of individual forms, it is very easy to read a symbol before the 
package setup needed to read that symbol correctly has been accomplished. 
Three examples: an inherited symbol that should have been shadowed might 
be accessed; a single-colon prefix might be used for a symbol that hasn't
yet been exported; an internal symbol might be created afresh where a 
symbol that will later be imported or inherited was intended.  These 
problems can be difficult to understand or even to recognize; in some 
cases it is difficult or impossible to correct the situation in the
same Lisp image.

Add a DEFPACKAGE macro to the language.  In the description below,
'package-name' and 'symbol-name' can be a symbol or a string; if a 
symbol, only its name matters, not what package it is in.

The syntax of DEFPACKAGE is

  (DEFPACKAGE package-name {option}*)

where each option is a list of a keyword and arguments.  Nothing in a
DEFPACKAGE form is evaluated.

Standard options for DEFPACKAGE are listed below.   Except for :SIZE, 
options may appear more than once (this is useful primarily for 

(:NICKNAMES {package-name}*)
        Set the package's nicknames to the specified names.

(:USE {package-name}*)
        The package is to "use" the other designated packages; that is,
        it will inherit from them.  The default value for this option 
        should be the same as it is for MAKE-PACKAGE (also see the issue

(:SHADOW {symbol-name}*)
        Create the specified symbols in the package being defined, 
        making them shadows, just as the function SHADOW would do.

(:SHADOWING-IMPORT-FROM package-name {symbol-name}*)
        Find the specified symbols in the specified package, import
        them into the package being defined, and place them on the 
        shadowing symbols list.  In no case will symbols be created in 
        any package other than the one being defined; a continuable error 
        is signaled if no symbol is accessible in the package named 
        package-name for one of the symbol-names.

(:IMPORT-FROM package-name {symbol-name}*)
        Find the specified symbols in the specified package and import
        them into the package being defined.  In no case will symbols be 
        created in a package other than the one being defined; a continuable
        error is signaled if no symbol is accessible in the package named 
        package-name for one of the symbol-names.

(:EXPORT {symbol-name}*)
        Find or create symbols with the specified names and export them.
        Note an interaction with the :USE option, since inherited symbols 
        can be used rather than new ones created;  note also an interaction 
        with the :IMPORT-FROM and :SHADOWING-IMPORT-FROM options, since 
	imported symbols can be used rather than new ones created.

(:INTERN {symbol-name}*)
        Find or create symbols with the specified names.  Note an 
        interaction with the :USE option, since inherited symbols 
        can be used rather than new ones created.  This option is useful if 
        an :IMPORT-FROM or a :SHADOWING-IMPORT-FROM option in a subsequent 
        call to DEFPACKAGE (for some other package) expects to find these 
        symbols accessible but not necessarily external.

(:SIZE integer)
        Declare the approximate number of symbols expected in the package.
        This is an efficiency hint only, so that the package's table will
        not have to be frequently re-expanded when new symbols are added
        to it (e.g., by reading in a large file "in" that package).

The order in which the options occur in a DEFPACKAGE form is irrelevant;
but since the effects of the entry-making options are context-sensitive, 
the order in which they will be executed is as follows:
  (2) :USE 
  (4) :EXPORT
Shadows are established first, since they may be necessary to block 
spurious name conflicts when the use link is established.  Use links are 
established next so that :intern and :export may refer to normally 
inherited symbols.  The :export is done last so that it may refer to 
symbols created by any of the other options; in particular, shadows and 
imported symbols can be made external.  Note also the prescription on CLtL 
p.178 to cover the case of calling EXPORT on an inherited symbol.

DEFPACKAGE creates the package as specified and returns it as its value.
It has no other side effects; e.g., it does not alter the value of *PACKAGE*.
The function COMPILE-FILE should treat top-level DEFPACKAGE forms the
same way it treats the other package-effecting functions (see CLtL p.182).

If the specified name already refers to an existing package, then the 
name-to-package mapping for that name is not changed.   At most, the 
existing package will be modified to reflect the new definition;  it is 
undefined what happens if the new definition is at variance with the 
current state of that package.  If one of the specified nicknames already
refers to an existing package, then an error is signaled just the same
as MAKE-PACKAGE would.  See the issue PACKAGE-DELETION for undoing the
name-to-package mapping.

Some DEFPACKAGE errors are, however,  purely syntactic.
  (1) An error should be signaled if :SIZE appears than once.
  (2) Since extended options might be allowed by other implementations, 
      an error should be signaled if an option is present that is not 
      actually supported in this implementation.
  (3) The collection of symbol-name arguments given to the options 
      all be disjoint; additionally, the symbol-name arguments given to 
      :EXPORT and :INTERN must be disjoint. If either condition is 
      violated, an error should be signaled.
Name conflict errors will, of course, be handled by the underlying calls to 


;;; An example that "plays it super-safe" by using only strings as names; 
;;;  does not even assume that the package it is read in to "uses" LISP; 
;;   *never* creates any symbols whatsoever in the package that it is read 
;;     in to.

  (:USE "LISP")

;;; A similar call, mostly using symbols rather than strings as names.
;;; Expects to be read in to a package that "uses" LISP and *may* create
;;;  random internal symbols in that package (such as MY-PACKAGE etc).

(defpackage my-package
  (:nicknames mypkg :MY-PKG)		;remember CL conventions for case
  (:use lisp)				; conversion on symbols
  (:shadow CAR :cdr #:cons)
  (:export "CONS")			;yes, this is the shadowed one.


The availability of DEFPACKAGE encourages putting the entire package 
definition in a single place.  It also encourages putting all the package 
definitions of a program in a single file, which can be loaded before 
loading or compiling anything else that depends on those packages; such a
file can be read in the USER package, avoiding any initial state issues.

In addition, DEFPACKAGE allows a programming environment to process
the whole package setup as a unit, providing better error-checking and
more assistance with package problems, by dint of global knowledge of
the package setup.

Current practice:

Symbolics Common Lisp (SCL) has always had a DEFPACKAGE, and users
prefer it to individual calls to EXPORT, IMPORT, SHADOW, etc.  The SCL
version of DEFPACKAGE has quite a few additional options, but none of
them appears to be necessary to propose for Common Lisp at this time.
This proposal is incompatible with Symbolics DEFPACKAGE in some ways
that will probably not cause major problems.

Cost to Implementors:

Small--DEFPACKAGE can be implemented simply as a bunch of
calls to existing functions.

Cost to Users:

Small, this is upward compatible.

Cost of non-adoption:

Packages continue to be difficult to use correctly.


Guide users away from using packages in ways that get them into trouble.




It has been suggested that the "Put IN Seven EXtremely Random USEr
Interface COmmands" mnemonic described in CLtL p.191 could be removed;
and with possibly a few exceptions, the special handling of them by
COMPILE-FILE could be removed.  As this would be an incompatible change, 
it is not part of this proposal.  However, a new mnemonic can be offered, 
to help remember the ordering constraints mentioned above:
          I REmember Six USEr Interface Expressions
Each word in the sentence corresponds to one operation listed below:
   I				IN-PACKAGE	;"foot" to stand on
   REmember			REQUIRE		;ensure pre-requisite packages
   Six				SHADOW		;block multiple-inheritances
   USEr				USE-PACKAGE	;go for it!
   Interface			IMPORT		;bring in "foreign" symbols
   EXpressions			EXPORT		;a "face" to show to others.

It is noted that DEFPACKAGE cannot be used to create two "mutually
recursive" packages, such as:
    (defpackage my-package
      (:use lisp your-package)	        ;requires 'your-package' to exist first
      (:export "MY-FUN"))
    (defpackage your-package
      (:use lisp)
      (:import-from my-package "MY-FUN") ;requires 'my-package' to exist first
      (:export "MY-FUN"))
However, nothing prevents one from using the package-effecting functions 
such as USE-PACKAGE, IMPORT, and EXPORT to establish such links (which
ought to be very rare) after a more standard use of DEFPACKAGE.

The macroexpansion of DEFPACKAGE could usefully canonicalize the names
into strings, so that even if a source file has random symbols in the
DEFPACKAGE form, the compiled file would only contain strings.

Frequently additional implementation-dependent options take the
form of a keyword standing by itself as an abbreviation for a list
(keyword T); this syntax should be properly reported as an unrecognized
option in implementations that do not support it.

Definition forms in Common Lisp usually just establish a name-to-object
mapping; there is little precedent for them to modify other global-context
state.  For this reason, we didn't want DEFPACKAGE also to "go" into the 
new package.  If it did so, like IN-PACKAGE, then the following reasonable 
file would become confused, because it wouldn't all be in one package:
   (in-package "USER")
   (defpackage "WATER"      (:use "LISP")             (:export "FISH"))
   (defpackage "ALCHEMY"    (:use "LISP" "PHLOGISTON) (:export gold))
Should the token 'gold' be read while in the USER package, or in the
the WATER package?

The issue IN-PACKAGE-FUNCTIONALITY recommends that IN-PACKAGE  be 
incompatibly changed to recognize only existing packages, not to create 
them.  IN-PACKAGE would then not accept any keyword arguments.

The function MAKE-PACKAGE might also be extended to take all the keywords
that DEFPACKAGE does. This could be the subject of a separate cleanup.