error in unread-char

Hello all,
I think I've found some behavior in unread-char which
doesn't match the spec (which  is a pity, because in this
case I found the actual behavor rather useful).

Consider the following

(with-input-from-string (s "abcdefg")
  (read-char s)
  (read-char s)
  (unread-char #\b s)
  (unread-char #\a s)
  (read-char s))

In LispWorks:
=> #\a

In CLISP:
=> *** - UNREAD-CHAR from #<INPUT STRING-INPUT-STREAM> without
READ-CHAR before it

and the docs state

"It is an error to invoke unread-char twice consecutively on the same
stream without an intervening call to read-char (or some other input
operation which implicitly reads characters) on that stream."

of which I must admit, this is a case.

Also of note (which I just discovered), it doesn't seem to matter
whether or not the
arg to unread char was actually the last unread char. And, there seem
to be two distinct
behaviors in LW.

(with-input-from-string (s "abcdefg")
  (read-char s)
  (read-char s)
  (unread-char #\x s)   ;; not #\b
  (unread-char #\k s)   ;; not #\a
  (read-char s))
=> #\a ;; magic !

However, in a listener, I seem to be able to unread whatever I want. E.g.

CL-USER 12 > (unread-char #\h)
#\h

CL-USER 13 > h        ;; I didn't insert this h


Any thoughts? I'm on LWM 445, Mac OS X 10.4.2


--
=====================
Joshua Taylor
tayloj@rpi.edu