Issue REQUIRE-PATHNAME-DEFAULTS-YET-AGAIN Writeup

Forum:		Public Review

Issue:		REQUIRE-PATHNAME-DEFAULTS-YET-AGAIN

References:	Barrett's public review comment #26

		Loosemore's public review comment #7

		REQUIRE

		Issue REQUIRE-PATHNAME-DEFAULTS

		Issue REQUIRE-PATHNAME-DEFAULTS-AGAIN

Category:	CHANGE

Edit history:	20 Dec 1992, Version 1 by Loosemore

Status:		Proposal RESTORE-ARGUMENT passed 11-0 on letter ballot

                92-302.



Problem description:


  In the base document (CLtL), the REQUIRE function was specified as

  accepting a required module name argument and an optional pathname

  argument.  In the current specification, the optional pathname

  argument has been removed.  The rationale given for this in

  X3J13/91-415 is that it was the second argument that led to

  portability problems.  This claim is incorrect.  It was the case that

  both the one argument and the two argument forms had portability

  problems in the past, but the addition of logical pathnames made the

  two argument version easier to use in a portable way.  Elimination of the

  second argument means that we're left with the same problem the single

  argument form has always had, namely that there are several competing,

  incompatible mechanisms for dealing with the situation where the

  module being required has not already been provided.



Proposal (REQUIRE-PATHNAME-DEFAULTS-YET-AGAIN:RESTORE-ARGUMENT):


  Restore the optional pathname argument to REQUIRE, as specified

  in the base document.


  Replace the examples with ones that better illustrate portable and

  nonportable uses of REQUIRE, such as those suggested below.



Rationale:


  Restoring the second argument makes it at least *possible* to use

  REQUIRE in a portable way.



Examples:


  ;;; This illustrates a nonportable use of REQUIRE, because it

  ;;; depends on the implementation-dependent file-loading mechanism.


  (require "CALCULUS")


  ;;; This use of REQUIRE is nonportable because of the literal 

  ;;; physical pathname.  


  (require "CALCULUS" "/usr/lib/lisp/calculus")


  ;;; One form of portable usage involves supplying a logical pathname,

  ;;; with appropriate translations defined elsewhere.


  (require "CALCULUS" "lib:calculus")


  ;;; Another form of portable usage involves using a variable or

  ;;; table lookup function to determine the pathname, which again

  ;;; must be initialized elsewhere.


  (require "CALCULUS" *calculus-module-pathname*)



Current practice:


  Who knows?



Cost to implementors:


  Minor.



Cost to users:


  None.



Aesthetics:


  Having a feature in the language which is impossible to use portably

  is unaesthetic.



Editorial impact:


  The changes are confined to the dictionary entry for REQUIRE.



Discussion:


  The proposal to restore REQUIRE to the language without its second

  argument was brought up at the Dec 1991 meeting without the benefit of

  a formal writeup or any advance discussion on the mailing list, which

  accounts for the resulting technical blunder.  We don't have a very

  good record when it comes to technical decisions made on the fly at

  meetings.


  Loosemore says: I would still prefer to remove this broken feature

  from the language entirely, but I'd rather not waste any more time

  arguing about that given that some people seem to feel strongly about

  keeping it in.  I'm willing to accept proposal RESTORE-ARGUMENT

  as a compromise, but doing nothing would not be acceptable to me.