Re: [DotGNU]Developing a IL -> jvm translator

[Gopal V]

If memory serves me right, Sylvain Pasche wrote:
> I think about using the coder architecture developed in pnet. That is,
> to write a coder objet that will output java bytecode during the
> method verification process.

Coder is a simple way to avoid the messy IL loading funcs ... 

> 1) The translator outputs opcodes in textual form (printf). In this
> case, most of the ildasm routines will need to be used to output all
> code other than methods. The output will have to be assembled
> afterward to get a .class file.
..........
> I seem to prefer the first, as I think it gives a more easy development.

I vote for using an assembly stage .... that's easier to debug. 

But in case if you are interested in using pnet's assembler for this 
work, it still has problems with .class output. The issue is more of
the meta-data handling than due to design .... I think maybe doing a
set of meta-data functions would make it possible to use the current
ilasm (for rhys's JL) for assembling to .class files.

But anyway the issue of the libraries depending on native libs (PInvoke)
remain ...

Gopal
--
The difference between insanity and genius is measured by success