Re: Reflection on function argument types

[Per Bothner <per at bothner dot com>]

On 12/27/2016 09:34 PM, Duncan Mak wrote:

> I wrote the following code:
>
> (import (srfi 1))
> (define (foo (a :: integer)) (+ a a))
> (define (hello-number (n :: number) (i :: number)) (list "number" n))
> (define (hello-string (s :: string)) (list "string" s))
> (define (print-types (func :: gnu.mapping.MethodProc))
>   (display (procedure-property func 'name)) (newline)
>   (for-each (lambda (t) (display t) (newline))
>             (map (lambda (i) (func:getParameterType i)) (iota
> (func:numParameters))))
>   (newline))
> (print-types foo)
> (print-types hello-number)
> (print-types hello-string)
>
> I'm surprise to see that the output differs between foo and hello-*:
>
> foo
> Type integer
>
> hello-number
> ClassType java.lang.Object
> ClassType java.lang.Object
>
> hello-string
> ClassType java.lang.Object
>
> Why is it that I only see java.lang.Object for the hello-* functions?
> Shouldn't number and string show up, like it did for foo?

This turns out to be a problem with demangling.  'foo' works because
it doesn't have any special characters.

In ModuleMethod.java in resolveParameterTypes it says:
            String mangledName = Mangling.mangleNameIfNeeded(name);
but it should be:
            String mangledName = Mangling.mangleName(name);
 
Unfortunately, fixing this improves the accuracy of type-checking
enough to cause some problems.  For example this code
(simplified from r7rs-tests.scm):

(let ((val 'abc))
  (cond ((and (complex? val)
              (not (nan? val)))
         (display val))))

Kawa complains that the call to nan? is invalid because the argument val
is a symbol.  This is bogus, since the previous test to complex? ensures
nan? is never called.

I created this issue:
https://gitlab.com/kashell/Kawa/issues/1
--
	--Per Bothner
per@bothner.com   http://per.bothner.com/