Hi,
I have fixed some more bugs or problem areas in my copy of iterate, and I can send that to anybody who wishes.
You may want to try the following test in your original version: (deftest or.1 (iter (generate x in '(a nil nil 1)) (generate y in-vector '#(2 #\c #\d)) (collect (or (next x) (next y)))) (a 2 #\c 1))
From my changes file:
o (if (next x) ...) works -- the test form is walked too + allow better compiler optimizations by keeping macros like and, or, prog1 etc. in the expanded code. The expansion also becomes more readable (closer to the source). + #+clisp support ((setf foo) new-value args...) function syntax + #+clisp simple support for handler-bind (not walking handlers) + free-vars correctly handles locally, symbol-macrolet and declare + use macroexpand-1 only to ensure all special handlers get used
and still more test cases
Marco, I don't know how you modified your UCW and CPS converter for use with CLISP, but I recommend you do not use sys::setf-... (clisp internals) -- I'm not even sure your patch works in all conceivable cases. I don't know what the requirements for ucw/cps are, but here's what I use for iterate: #+clisp ; some macros expand into ((setf foo) value other-args...) ;; reported by Marco Baringer on 24 Jan 2005 ((typep form '(cons (cons (eql setf) *) *)) (apply #'walk-cdr form)) You can see that the setf form is treated like any other function call, and there's no need to make use of (possibly changing) internals.
Things I changed previously are: o (DSETQ (values x nil y)) works o restore v 1.0/2 special/macro order, e.g. change special-form? etc. o in-vector/in-sequence downwards by 2 did not start with last element (iter (for i index-of-vector #(0 1 2 3 4) downto 0 by 2) (collect i)) yielded (3 1), expect (4 2 0) by similarity with (iter (for i downfrom 4 by 2) (collect i)) and (iter (for i from 4 downto 0 by 2) (collect i)) + locally accepts [0..N] declarations + constant? and self-evaluating? accept all self-evaluating forms, including keywords and T/NIL, but no other symbols. + avoid ~< pretty-printer (not in Cormanlisp) in error messages, leave that to the implementation + if-1st-time & special-clause ELSE: type 'boolean instead of symbol + #+clisp "explicit declarations are always placed in the generated code" as the manual says. + #+lispworks multiple-value-bind is recognized (again) + make-next-code N <> 1 generates correct code
Regards, Jorg Hohle