Re: eliminate gcjh?

[sf <sf at b-i-t dot de>]

Per Bothner wrote:
> Christian Stuellenberg wrote:
>
> > If I remind right, [coverant returns]
> > is "only" at language level, since they
> > did not want to change the VM (to stay compatible e.g.).
...
> > 1.5 does
> > only translation of some syntactic sugar to other valid java code on
> > the fly I think...
>
>
> Well, I'm sure it it doesn't translate to "other java code".  And
> it most definitely generates .class files you could not create with
> earlier versions.
>
> sf wrote:
>  > Covariance has always existed at bytecode level. See also
>  > http://gcc.gnu.org/bugzilla/show_bug.cgi?id=9861 .
>
> I'd rather say "support for covariance in the sense of overloading based
> on return type" as always existed.  Covariance implies that the method
> with the changed return types overrides the method in the base class;
> in the bytecode they are completely unrelated.

If you put it this way you are right. Overloading, based on return type 
as well as parameter type, is part of a language specification at a 
higher level. The Java language supports one but not the other.

So until 1.5, covariant methods just never occured at the bytecode level 
when compiling java programs.

Regards
	Stephan