Stonewall, Thanks for the traces and your efforts. The following trace is really what makes me wonder. Inside Iterate, my patch should not expand m-v-b anymore and call walk-m-v-b instead. ((special-form? (car form)) ; handle known special operators first (walk-special-form form)) ((macro-function (car form) *env*) (walk (macroexpand form *env*))) ((special-operator-p (car form)) Yet both macroexpand and macroexpand-1 show that they are called upon m-v-b. So far, I don't understand where this comes from. 0 MACROEXPAND > ...

...

SYSTEM::FORM : (MULTIPLE-VALUE-BIND (#:G831 #:G832) (FOO) (LET* NIL (VALUES (SETQ A #:G831) (SETQ B #:G832)))) SYSTEM::ENV : #<Environment venv (#<Venv 269101724 B> #<Venv 269101828 A>) fenv NIL benv ((TEST-ITER . #<Block TEST-ITER>)) tenv NIL> 0 MACROEXPAND < ... << VALUE-0 : (SYSTEM::MULTIPLE-VALUE-BIND-CALL (FUNCTION (LAMBDA (&OPTIONAL #:G831 #:G832 &REST #:EXTRA-VALS833) (DECLARE (IGNORE #:EXTRA-VALS833) (COMPILER::WEAK-LAMBDA-NAME COMPILER::DUMMY-LAMBDA)) (LET* NIL (VALUES # #)))) (FOO)) << VALUE-1 : T

There are 3 things worth trying (declaim (notinline walk)) before definition or use (declare (optimize debug)) inside walk replace macroexpand in walk with macroexpand-1

It's interesting that this expansion turns M-V-BIND into M-V-CALL, while the first two traces above produced a M-V-BIND-CALL. Indeed.

I'm also a little suspicious of your revised definition:

...

(or (assoc symbol *special-form-alist*))) since it has a superfluous OR. I dropped the OR. I'm sorry that my saving a little time and typing caused even more loss of time and typing to you and me. I should have simplified the form after dropping the second or clause.

You can get the package from <http://www.lispworks.com/downloads/lw-personal-edition.html>. I imagine that the same error occurs on all the supported platforms. Installing packages cost me a lot of work. I can make no promise as to when I'll be able to do that. But it would be reasonable to install a collection of Lisps here so I can play with comparisons and code that does not work with either CLISP or Corman, when time becomes available.

Regards, Jorg.