Re: Patch: Remove JvAllocObject and add fields to Class for interpreter/JITdata

[Bryce McKinlay <mckinlay at redhat dot com>]

Tom Tromey wrote:

> > > > > > "Bryce" == Bryce McKinlay <mckinlay@redhat.com> writes:
> > > > > >            
>
> Bryce> Please comment on the following patch, particularly the names and
> Bryce> nature of the new fields in Class.h
>
> I think this general idea is great and long overdue.
>
> Bryce> 	* java/lang/Class.h (Class): Add new fields `aux_id' and 
> Bryce> 	`aux_info'.
>
> aux_id isn't used (except in the GC) or set.  What if we just remove
> it for the time being?  I think we'll need something like this, but I
> think we should hold off on it until we do some actual design on the
> pluggable execution engine idea.
>  

OK. I was planning on changing the implementation of 
_Jv_IsInterpretedClass to use aux_id, but thinking about it more I'm 
starting to think the whole idea is not so great. If the interpreter's 
resolver, etc, were more cleanely organised then it should probably not 
even need to map from java.lang.Class to _Jv_InterpretedClass. As it 
stands this mapping is quite ugly/unsafe due to the "void*" casting.

But, we do need this patch because the interpreter is currently broken 
without it - so I'm going to remove aux_id and check it in.

Regards

Bryce.