[Gcl-devel] Feedback on ANSI bugfixes

[Paul F. Dietz]

I *really* should put the partial test suite I'm using up for
ftp, so you can run it yourself.  I'll do that in
a day or so; the glue files need to be tweaked to work around
ANSI CL lacks in gcl.  A comprehensive test suite will be utterly
essential to upgrading gcl (for conformance or performance).

Anyway... here are some problems.


(let* ((n "abcde") (x (list* 'a 'b 'c n))) (ldiff x (copy-seq n)))
  ===> (A B C)
Since (not (eql n (copy-seq n))), it should have returned
(A B C . "abcde").


(TAILP 10203040506070 (LIST* 'CL-TEST::A 'CL-TEST::B
                                  (1- 10203040506071)))
  ==> NIL
when it should be true (the arithmetic in this test prevents
an aggressive readed from making the two numbers EQ.)


(member-if #'listp 1)
   ==> 1
when it should signal a type error.

(member-if-not #'listp 1)
   ==> 1
when it should signal a type error.


(maplist #'identity 'a)
   ==> NIL
when it should signal a type error.


(assoc-if #'evenp '((1 . a) nil (3 . b) (6 . c) (7 . d)))
   ==> infinite loop
(and also for similar cases for assoc-if-not, rassoc-if, and
 rassoc-if-not).

(assoc-if #'null '((a . b) nil (c . d) (nil . e) (f . g)))
   ==> infinite loop
(ditto for the other three functions.)


This is the test case for the set-exclusive-or problem.
I've had arguments with other Lisp implementors as to
whether the standard really calls for this behavior.
IIRC, CMUCL does not enforce this ordering constraint.

(let ((list1 '(a b c d))
          (list2 '(e f g h)))
      (block fail
              (set-exclusive-or
               list1 list2
               :test #'(lambda (s1 s2)
                         (when (or (member s1 list2)
                                   (member s2 list1))
                           (return-from fail 'failed))))))
   ==> failed
when it should return some permutation of (a b c d e f g h).


        Paul Dietz
        address@hidden