Deprecations for 3.0

[Nick Wellnhofer]

On 17/09/10 19:34, Peter Lobsinger wrote:
> I think we need both a deprecation notice and stub implementations of
> these macros by 2.9. Being able to assign simply by using 'x = y' is
> pretty much taken for granted, and we'll be braking that assumption.

Yes, we should start with a stub implementation as soon as possible. 
Even if we don't get the interface right in the first try, we could at 
least begin to mark the relevant places in the code.

Generally, we will start with code like this:

PMC    *self;
PMC    *pmc_ref,    *pmc_new;
STRING *string_ref, *string_new;

pmc_ref    = pmc_new;
string_ref = string_new;

pmc_ref is a PMC pointer that belongs to PMC "self". pmc_new is the new 
value that is assigned to that pointer. For instance, in 
parrot_hash_put, pmc_ref can be the value pointer of a hash bucket, and 
pmc_new the new value that is assigned.

The same for strings. Then we have to replace the assignment with a 
macro like:

SET_PMC_REF(interp, self, pmc_ref, pmc_new);
SET_STRING_REF(interp, self, string_ref, string_new);

The naming is just a suggestion.

Implementing this with stubs should also allow us to find out if we need 
further adjustments. (Thinking more about the hash example, it turns out 
that we also have to pass the "self" PMC to parrot_hash_put. Or let the 
caller of parrot_hash_put do the final assignment.)

For the curious, a final implementation of SET_PMC_REF with a card 
marking scheme might look like this:

#define SET_PMC_REF(interp, self, pmc_ref, pmc_new) \
     do { \
         pmc_ref = pmc_new; \
         if (PObj_young_generation_TEST(pmc_new)) \
             interp->card_table[(size_t)self >> SHIFT] = 1; \
     } while(0)

Nick