Deprecations for 3.0

[Nick Wellnhofer]

On 17/09/10 23:14, Andrew Whitworth wrote:
> Those macros, SET_PMC_REF and SET_STRING_REF work well for me. We need
> to think carefully about how we want to add these things, however,
> considering that our GC systems are (in theory anyway) selectable. If
> every single one of these becomes an indirect function call, we could
> end up eating a lot of performance.

If we want to have truly pluggable GC systems, this is indeed a problem. 
 From a performance point of view, the write barrier must be at least 
partially inlined. If we ever want to support GC backends with different 
write barriers, I see no other way than making that selection 
compile-time dependent, and maybe change the macros to function calls 
for external code.

Nick