615257 - MMgc::setExact should probably be hidden

Status: VERIFIED DUPLICATE of bug 626612

(Tamarin Graveyard :: Garbage Collection (mmGC), defect)

Description

[Lars T Hansen]

In bug #612561, Tommy writes:

"One thought regarding SetHasTrace.   Do we need to allow SetHasTrace to be exposed to the mutator?  I'm thinking when safegc forces ctor invocation to move in the GC we can completely hide the trace setting.   Ie:

return MMgc::setExact(new (gc, AbcEnv::calcExtra(builtinPool))
                                AbcEnv(builtinPool, builtinCodeContext));

becomes:

return GCNewFlagsExtra<AbcEnv>(gc, 
                               kExtactTrace,
                               AbcEnv::calcExtra(builtinPool), 
                               builtinPool,
                               builtinCodeContext);

or maybe the macros could specialize GCNewExtra for AbcEnv so it'd just be:

return GCNewFlagsExtra<AbcEnv>(gc, 
                               AbcEnv::calcExtra(builtinPool), 
                               builtinPool,
                               builtinCodeContext);

...

GCNew invokes GC::Alloc and then calls your ctor using placement new.  Then it would set the trace bit.  Functionally its exactly the same, just code movement from mutator to MMgc."

The benefit of this would be to shrink the API, which seems good.  The liability is that setExact is no longer available for uncommon cases, which may not be so good - but we don't know yet.

Comment 1

[Lars T Hansen]

Subsumbed by 626612, marking as dup of that even though that's stretching the terminology a bit.

Comment 2

[Trevor Baker]

bulk verifying resolved !fixed issues