Re: Patch: RFC: class-related structures and conservative scanning

[Bryce McKinlay <mckinlay at redhat dot com>]

Tom Tromey wrote:
> I'm not checking this in yet, I thought I'd get some comments first.
>
> This patch changes the marking of Class objects so that only the
> direct fields of Class are marked.  Then it updates all the
> Class-related structures to be allocated with _Jv_AllocRawObj, so that
> they will be conservatively scanned.
>
> This fixes a number of memory leaks and is also useful infrastructure
> for eventually enabling incremental collection.
>   

This looks good to me. However, I think some of the fields - for 
efficiency and better precision - should be allocated with 
_Jv_AllocBytes and not _Jv_AllocRawObj:

"ioffsets", as it does not contain pointers.
"itable" and "ancestors", because they only ever contain pointers that 
are already guaranteed to be reachable via the other fields of the same 
class ("ifaces" and "superclass", respectively).

Also, For the interpreter static data, what we should really do is split 
this into separate pointer and non-pointer data blocks - thus getting 
precise scanning for free.

Bryce