Re: Kawa regressions since 1.9.90?

[Jamison Hope <jrh at theptrgroup dot com>]

On Oct 1, 2010, at 6:00 PM, Jamison Hope wrote:

> Hi Per,
>
> I just tried rebuilding some code with the latest Kawa from SVN  
> which I had previously compiled using 1.9.90, and I now see a few  
> warnings/errors which I didn't see before:
>
> One is of the form "warning - type integer is incompatible with  
> required type double" in statements such as (javax.vecmath.Vector3d  
> -1.5 0 0). Apparently Kawa forgot how to promote an int to a double?
>
> The other is "warning - no known slot 'Environment' in  
> java.lang.Object" / "invalid type spec (must be "type" or 'type or  
> <type>)"
> in a file which looks like this:
>
> (define-alias Environment foo.Environment) ;; i.e. not  
> gnu.mapping.Environment
>
> (define-simple-class EnvUser (baz.Superclass)
>  ((main (args :: String[])) allocation: 'static :: void
>   ((EnvUser):bar (Environment)))
>
>  ((bar (e :: Environment)) ;; this is where the compiler complains
>   (format #t "~A ~A~%" (this) e)))
> ;; EOF
>
> If I rename my alias to "Env" instead of "Environment", then it  
> compiles cleanly. Now here's the extra weird part: in trying to make  
> a simple test case to demonstrate the problem, I discovered that it  
> also depends upon which class I inherit from. If I make EnvUser a  
> subclass of java.lang.Object, for instance, then it compiles cleanly  
> even if the alias is called Environment. It's only with one  
> particular superclass that I see the errors. Unfortunately it's not  
> a class whose source I'm at liberty to publish right now, but as far  
> as I can tell there isn't anything special about it; it's just a  
> public abstract class written in Java. I'll keep digging, but in the  
> mean time do you have any ideas?

I found what triggers the second problem: the superclass in question  
has a method called getEnvironment(). If I comment it out, then the  
Kawa subclass compiles. It doesn't even matter what the method's  
return type is. So this'll trigger it:

$ ls test
Base.java         Environment.java  Foo.scm
$ cat test/Environment.java
package test;

public class Environment {}
$ cat test/Base.java
package test;

public class Base {
  int getEnvironment() { return 42; }
}
$ cat test/Foo.scm
(define-alias Environment test.Environment)
(define-alias Base test.Base)

(define-simple-class Foo (Base)
  ((main (args :: String[])) allocation: 'static :: void
   ((Foo):bar (Environment)))

  ((bar (e :: Environment))
   (format #t "~A ~A~%" (this) e)))
$ javac test/*.java
$ java kawa.repl -C test/Foo.scm
(compiling test/Foo.scm to test.Foo)
test/Foo.scm:8:15: warning - no known slot 'Environment' in  
java.lang.Object
test/Foo.scm:8:15: invalid type spec (must be "type" or 'type or <type>)


And now that I know what the pattern is, I see that it isn't anything  
special about the name "Environment", it's defining and using an alias  
Xxx when the superclass has a method getXxx(). Here's an even simpler  
test case:

$ ls test2
Base.java   Sub.scm
$ cat test2/Base.java
package test2;
public class Base {
  void getAbc() {}
}
$ cat test2/Sub.scm
(define-alias Abc java.lang.String)
(define-simple-class Sub (test2.Base)
  ((foo (arg :: Abc)) arg))
$ javac test2/Base.java
$ java kawa.repl -C test2/Sub.scm
(compiling test2/Sub.scm to test2.Sub)
test2/Sub.scm:3:17: warning - no known slot 'Abc' in java.lang.Object
test2/Sub.scm:3:17: invalid type spec (must be "type" or 'type or  
<type>)


Like I said before, this bug was introduced in 1.10. If I edit my  
classpath so that it finds kawa-1.9.90.jar instead, then the  
compilation succeeds:

$ java kawa.repl -C test2/Sub.scm
(compiling test2/Sub.scm to test2.Sub)
$

-Jamie

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