CTYPE leakage from SETQ

[Paul Khuong]

We splice CTYPEs in directly instead of using a type specifier when ir1 converting setq. Should we try and detect these meta-level leakage issues during ir1-conversion, or do we (want to) actively depend on that being possible?

Test case by Zach Beane:

;;;; foo.lisp

(defun foo ()
  (let ((em 0d0))
    (declare (type double-float em))
    (dotimes (i 42)
      (setf em (float (1+ i))))))

;;;; (compile-file "foo.lisp")

; compiling file "/tmp/foo.lisp" (written 15 NOV 2011 10:59:09 AM):
; compiling (DEFUN FOO ...)

; file: /tmp/foo.lisp
; in: DEFUN FOO
; (SETF EM (FLOAT (1+ I)))
; --> SETQ THE SB-C::%COMPILE-TIME-TYPE-ERROR
; ==>
; '((THE #<SB-KERNEL:NUMERIC-TYPE DOUBLE-FLOAT> (FLOAT (1+ I)))
; SB-KERNEL:%SINGLE-FLOAT SB-C::N)
;
; caught ERROR:
; SB-KERNEL:NUMERIC-TYPE (supertype of SB-KERNEL:NUMERIC-TYPE) is not a DEF!STRUCT-defined type.

; --> SETQ THE
; ==>
; (SB-C::%COMPILE-TIME-TYPE-ERROR 'SB-C::DUMMY 'DOUBLE-FLOAT
; '(VALUES (SINGLE-FLOAT 1.0) &OPTIONAL)
; '((THE #<SB-KERNEL:NUMERIC-TYPE DOUBLE-FLOAT>
; (FLOAT (1+ I)))
; SB-KERNEL:%SINGLE-FLOAT SB-C::N))
;
; note: The fourth argument never returns a value.

; (DOTIMES (I 42) (SETF EM (FLOAT (1+ I))))
; --> DO BLOCK LET TAGBODY PSETQ PSETF LET* MULTIPLE-VALUE-BIND LET 1+ +
; ==>
; I
;
; note: deleting unreachable code
;
; compilation unit finished
; caught 1 ERROR condition
; printed 2 notes

; /tmp/foo.fasl written
; compilation finished in 0:00:00.006

[Paul Khuong (pvk)]

Fix committed in 5efae23 Plug a CTYPE leak into fasls via SETQ.