subtypep giving inconsistent results on certain cons types

[Paul F. Dietz]

If A is a subtype of B, then (NOT B) should be a subtype of (NOT A). If subtypep can determine the subtype relationship in both cases then they must be consistent. However:

(SUBTYPEP '(CONS INTEGER (CONS (REAL -3.3 4.4) (EQL 1)))
          '(NOT (CONS T (CONS (NOT INTEGER) (EQL 0))))) => T, T
(SUBTYPEP '(NOT (NOT (CONS T (CONS (NOT INTEGER) (EQL 0)))))
          '(NOT (CONS INTEGER (CONS (REAL -3.3 4.4) (EQL 1))))) => NIL, T
(SUBTYPEP '(CONS T (CONS (NOT INTEGER) (EQL 0)))
          '(NOT (CONS INTEGER (CONS (REAL -3.3 4.4) (EQL 1))))) => NIL, T

This is in sbcl 1.3.18.

[Paul F. Dietz (paul-f-dietz)]

This is still occurring in 1.4.3.

[Michał "phoe" Herda (phoe-krk)]

(let* ((t1 '(not (cons float t)))
       (t2 `(or (not (cons (eql 0)
                           (real -3.926510009989861d7 -3.926510009989861d7)))
                (not (cons t (eql 0))))))
  (subtypep t1 t2))

;=> NIL, T

This happens even though T2 is equivalent to T.

[Paul F. Dietz (paul-f-dietz)]

That was from test SUBTYPEP.CONS.43 in ansi-test, which was added back in SBCL 0.9.6.57.

[Paul F. Dietz (paul-f-dietz)]

Still occurring in 2.2.11.38-5afe2f165.

[Paul F. Dietz (paul-f-dietz)]

#2 is good now, but the original bug is still there. Also:

(defun bug112 ()
  (let ((t1 '(cons t
              (cons (not (eql 0))
               cons)))
        (t2 '(not (cons null
                   (cons (real -2 -1)
                    null)))))
    (values
     (multiple-value-list (subtypep t1 t2))
     (multiple-value-list (subtypep `(not ,t2) `(not ,t1)))
     (multiple-value-list (subtypep (cadr t2) `(not ,t1))))))

(defun bug112a ()
  (let ((t1 '(cons (not (eql 0)) cons))
        (t2 '(not (cons (real -2 -1) null))))
    (values
     (multiple-value-list (subtypep t1 t2))
     (multiple-value-list (subtypep `(not ,t2) `(not ,t1)))
     (multiple-value-list (subtypep (cadr t2) `(not ,t1))))))

(bug112) ==> (NIL T), (T T), (T T)
(bug112a) ==> (NIL NIL), (T T), (T T) ;; which is ok

x86-64, "2.3.0.112-21b0a66b1"