D GC theory

[deadalnix via Digitalmars-d]

On Tuesday, 24 February 2015 at 15:22:02 UTC, Sativa wrote:
>> Only modifications of pointers, which introduce new 
>> cross-block dependencies (so that GC knows to recheck the new 
>> dependency). Other memory access goes without slowdown.
>
> But this type of thinking is the reason why the current GC is 
> in the state it is.
>

There are limitation you have in a system language, that you 
don't in a VM language. That being said, D has a type system that 
allow for various optimization when it come to the GC, so I don't 
think it is hopeless, but we need to plug the remaining holes.