[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Issue: PACKAGE-FUNCTION-CONSISTENCY (Version 1)
- To: CL-Cleanup@SAIL.Stanford.EDU
- Subject: Issue: PACKAGE-FUNCTION-CONSISTENCY (Version 1)
- From: Kent M Pitman <KMP@STONY-BROOK.SCRC.Symbolics.COM>
- Date: Fri, 21 Oct 88 18:20 EDT
From my list of "pending issues" distributed at the last X3J13 meeting...
References: 11.7 Package System Functions and Variables (pp182-188)
Edit history: 21-Oct-88, Version 1 by Pitman
Status: For Internal Discussion
CLtL is vague about whether either or both of package or package name
are permissible in some cases.
Clarify that it is permissible to pass either a package object
or a package name (symbol or string) in the following situations:
- the :USE argument to MAKE-PACKAGE or IN-PACKAGE
- the first argument to IN-PACKAGE, FIND-PACKAGE, and RENAME-PACKAGE
- the second argument to INTERN, FIND-SYMBOL, UNINTERN
- the second argument to EXPORT, UNEXPORT, IMPORT,
SHADOWING-IMPORT, and SHADOW
- the first argument (or a member of the list which is the first
argument) to USE-PACKAGE or UNUSE-PACKAGE.
- the PACKAGE argument to DO-SYMBOLS.
- the PACKAGE argument to DO-EXTERNAL-SYMBOLS.
- the PACKAGE argument to DO-ALL-SYMBOLS.
If FIND-PACKAGE is given a package object as an argument, it simply
If IN-PACKAGE is given a package object as an argument, that package
is selected. It is an error if, when a package object is given as a
first argument to IN-PACKAGE, a :USE list is explicitly specified and
does not exactly match the package or :NICKNAMES is explicitly supplied
and does not exactly match the package.
Clarify that the functions PACKAGE-NAME, PACKAGE-NICKNAMES,
PACKAGE-USE-LIST, and PACKAGE-USED-BY-LIST are (at least conceptually)
accessors to package data structures and permit only package objects
Clarify that the function MAKE-PACKAGE permits only a package name
as an argument since it does not make sense to create an existing
Clarify that package nicknames must always be expressed as package
names (symbols or strings) and may never be actual package objects.
(INTERN "FOO" "KEYWORD") => :FOO
(DEFVAR *FOO-PACKAGE* (MAKE-PACKAGE "FOO"))
(RENAME-PACKAGE "FOO" "FOO0")
(PACKAGE-NAME *FOO-PACKAGE*) => "FOO0"
This makes things more consistent.
Symbolics Genera permits strings as package names.
Symbolics Cloe does not permit strings as package names.
Cost to Implementors:
Cost to Users:
None. This change is upward compatible.
Cost of Non-Adoption:
Implementations would continue to vary gratuitously, leaving a potential
for portability problems.
The cost of non-adoption is avoided.
This makes things more regular, and so presumably more aesthetic.
Pitman ran across this problem while trying to port Macsyma to various
implementations. Discussion with other maintainers of portable programs
shows this is a common source of aggravation in portable code.
Since the pathname accessors all take namestrings or streams, one might
easily argue that it would be more consistent for PACKAGE-NAME,
PACKAGE-NICKNAMES, etc. to also take arguments of package names. After
(PACKAGE-NAME "FOO") => "FOO"
is no stranger than
(NAMESTRING "JOE:fred.lisp") => "JOE:fred.lisp".
If anyone wants to see a redraft of this to include even these
functions, they should speak up.
It would be possible to say that MAKE-PACKAGE took package objects as
arguments and just returned that package. That might have limited
usefulness on rare occassions, but mostly seemed too far out in left
field to bother suggesting it.