Re: CL not is not not?

[Jamison Hope <jrh at theptrgroup dot com>]

OK, I figured this out.

The problem line is
> boolean val = PrimType.booleanValue(value);

in Compilation#compileConstant. PrimType#booleanValue is implemented as:
>     return ! (value instanceof Boolean) || ((Boolean)  
> value).booleanValue();
which (of course!) returns the wrong value when passed Lisp2.FALSE
(which is a LList, not a Boolean) in Common Lisp.

Replacing that line in Compilation.java with
boolean val = getLanguage().isTrue(value);
does the right thing:

$ kawa --scheme
> #|kawa:1|# (not #t)
> #f
> #|kawa:2|# (not #f)
> #t
$ kawa --lisp
> #|kawa:1|# (not t)
> nil
> #|kawa:2|# (not ())
> t

Patch attached.

-Jamie

On Apr 24, 2012, at 4:25 AM, Per Bothner wrote:

> On 04/24/2012 01:03 AM, Jamison Hope wrote:
> > Speaking of Common Lisp, I find this interaction to be very  
> > curious...
> >
> > $ kawa --lisp
> > #|kawa:1|# (not t)
> > t
>
> Trying with --debug-print-final-expr I see:
>
> #|kawa:1|# (not t)
> [Compiling final atInteractiveLevel$1 to atInteractiveLevel$1:
> (Module/atInteractiveLevel$1/1/ () (Quote () ::boolean))]
> t
>
> So constant-folding works, correctly yielding ().
> Then that gets cast to boolean.  My guess is the problem is
> that cast is to the *Scheme* version of the boolean type,
> so () is treated as true.
>
> Using --debug-dump-zip shows the result is passed to writeBoolean
> in the actual run method.
> --
> 	--Per Bothner
> per@bothner.com   http://per.bothner.com/

--
Jamison Hope
The PTR Group
www.theptrgroup.com

Attachment: not-fix.patch
Description: Binary data