Re: merge-pathnames

[Bruno Haible <haible@ilog.fr>]

Roger Kehr <kehr@iti.informatik.th-darmstadt.de> asks why in CLISP

> (merge-pathnames (make-pathname :directory '(:relative "x"))
>                  (make-pathname :directory '(:relative "y")))
> => #"x/"

where he expects to get #"y/x/".

There are two reasons for this behaviour:

1. An informal one: I found the latter behaviour confusing and changed
   CLISP to do it the former way. It seems to work better this way.

2. A formal one: MERGE-PATHNAMES is used to specify default components
   for pathnames, so there is some analogy between (MERGE-PATHNAMES a b)
   and (or a b). Obviously putting in the same default a second time
   should do the same as putting it in once:

     (or a b b) is the same as (or a b), so

     (MERGE-PATHNAMES (MERGE-PATHNAMES a b) b) should be the same as
     (MERGE-PATHNAMES a b).

   (This question actually matters because in Common Lisp there is
   no distinction between "pathnames with defaults merged-in" and
   "pathnames with defaults not yet applied". For example, you don't
   know whether COMPILE-FILE will merge in some defaults.)

   Now, (MERGE-PATHNAMES (MERGE-PATHNAMES '#"x/" '#"y/") '#"y/")
   and  (MERGE-PATHNAMES '#"x/" '#"y/")
   are equal in CLISP's implementation, but not in implementations
   that strictly follow the Common Lisp spec. In fact, the above
   twice-default = once-default rule holds for all pathnames in CLISP.


Bruno Haible                                     email: <haible@ilog.fr>
Software Engineer                                phone: +33-1-49083585