Re: warnings when calling methods from Java interfaces, not classes

[Per Bothner <per at bothner dot com>]

Hoehle, Joerg-Cyril wrote:

> The problem is that these methods come from interfaces, not classes, so a cast using (as <class.x> y) is of no help here.

Er, Kawa appears to be correct.  If you transcribe this code to Java, it 
would not compile.  There is not getModulus or getPublicExponent in the 
PublicKey interface.

> E.g. Java.security is full of interfaces:
> http://java.sun.com/j2se/1.4.2/docs/api/java/security/interfaces/RSAPublicKey.html

But your code doesn't use <java.security.interfaces.RSAPublicKey>, it 
uses <java.security.PublicKey>.  If we change the declaration of pubkey 
to <java.security.interfaces.RSAPublicKey>  two of the warnings go away.

The other warnings can be shut up using casts to <byte[]>, though of 
course it is preferable if Kawa knows the types of bigint-n-bytes and 
gen-pkcs1:

> ;;TODO KeyMod may have 129 bytes, need 128
> (define (bigint-n-bytes n big :: <java.math.BigInteger>)
>   (invoke big 'toByteArray))

It's best to explicitly specify the return type as <byte[]>, though it 
doesn't seem to make a difference in this case.  I go back and forth on 
whether Kawa should attempt to infer return types of functions and just 
use <object> if it isn't declared.  The problem is that until Kawa has 
solid and consistent typechecking it's hard to specify under which 
circumstances Kawa can infer a return type, and it is important that 
this be consistent because it is wired into the Java method type.

If you declare bigint-n-bytes before rnd-pkcs1 and compile then as a 
file, then Kawa knows the return type of bigint-n-bytes.  (You can also
force them to be compiled together with a 'begin' form.)

I expect to do an overhaul of Kawa's type-checking and inference at some 
point in order to implement correct static type-checking for Qexo.  No 
idea when that will happen.

> It further seems that type declarations for variables are ignored by subsequent code in interactive mode.

It works when compiling as a module.  Compiling with with -C produces no 
warnings:

(define random-gen :: <java.security.SecureRandom>
  (invoke-static <java.security.SecureRandom> 'getInstance "SHA1PRNG"))
(define (foo key-gen :: <java.security.KeyPairGenerator>)
  (invoke key-gen 'initialize (as <int> 1024) random-gen))

I don't know why the cast of 1024 to <int> is needed.  Probably Kawa 
could be smarter here.
--
	--Per Bothner
per@bothner.com   http://per.bothner.com/