(let ((predicate-name (second hairy-spec)))
(declare (type symbol predicate-name)) ; by ANSI spec of SATISFIES
(if (fboundp predicate-name)
- (values (not (null (funcall predicate-name obj))) t)
+ (let* (;; "Is OBJ of the SATISFIES type?" represented
+ ;; as a generalized boolean.
+ ;;
+ ;; (Why IGNORE-ERRORS? This code is used to try to
+ ;; check type relationships at compile time.
+ ;; Passing only-slightly-twisted types like
+ ;; (AND INTEGER (SATISFIES ODDP)) into the
+ ;; rather-significantly-twisted type dispatch
+ ;; system can easily give rise to oddities like
+ ;; calling predicates like ODDP on values they
+ ;; don't like. (E.g. on OBJ=#\NEWLINE when the
+ ;; above type is tested for TYPE= against
+ ;; STANDARD-CHAR, represented as a
+ ;; MEMBER-TYPE.) In such cases, NIL seems to be
+ ;; an appropriate answer to "is OBJ of the
+ ;; SATISFIES type?")
+ (gbool (ignore-errors (funcall predicate-name obj)))
+ ;; RAW coerced to a pure BOOLEAN value
+ (bool (not (not gbool))))
+ (values bool t))
(values nil nil)))))))))
\f
;;; Return the layout for an object. This is the basic operation for
(assert (functionp fun))
(assert (not warnings-p))
(assert (not failure-p))))
+
+;;; a bug in 0.7.4.11
+(dolist (i '(a b 1 2 "x" "y"))
+ ;; In sbcl-0.7.4.11, the compiler tried to source-transform the
+ ;; TYPEP here but got confused and died, doing
+ ;; (ASSOC '(AND INTEGERP (SATISFIES PLUSP)))
+ ;; *BACKEND-TYPE-PREDICATES*
+ ;; :TEST #'TYPE=)
+ ;; and blowing up because TYPE= tried to call PLUSP on the
+ ;; characters of the MEMBER-TYPE representing STANDARD-CHAR.
+ (when (typep i '(and integer (satisfies oddp)))
+ (print i)))
+(dotimes (i 14)
+ (when (typep i '(and integer (satisfies oddp)))
+ (print i)))