Mark H Weaver <mhw@netris.org> writes:

> calcium <calcium@disroot.org> writes:
>> Here is the bug that I found :
>>
>> ;;; -START- code with the bug -START- ;;;
>>
>> (define (strange lst)
>>   (let loop ((lst lst)
>>              (is-empty '()))
>>     (cond ((nil? lst)
>>            (if (nil? is-empty) 'works
>>              (list 'should-not-occur is-empty)))
>>           (else
>>            (loop (cdr lst)
>>                  is-empty)))))
>>
>> (strange '())
>> => (should-not-occur ())
>>
>> (strange #nil)
>> => (should-not-occur ())
>
> Indeed, this certainly indicates a bug.
>
> I believe the bug is in 'local-type-fold' in (language cps type-fold).
> It contains a local procedure 'scalar-value' which, if I understand
> correctly, seems to incorrectly assume that (nil? x) returns #t if and
> only if X is 'eq?' to #nil.

The bug actually wasn't in 'local-type-fold', but it's true that the
type inferencer assumed in a few places that if (nil? X) returned #t
that X must be #nil, and similarly for 'null?' and ().

I've attached a proposed fix.

Andy, does this fix look right to you?

     Thanks,
       Mark