垃圾收集器：对象图的数据结构

Question

I read many resources on garbage collecting where they explained different algorithm. However, I didn't find any explaining the representation of the graph object.

My idea is quite simple : an oriented graph where vertex represent allocated memory block (on heap) and the edges the owner relationship. Example : Consider 2 memory blocks m1 and m2, if m1 contains reference to a block inside m2 then add an edge (m1, m2). These edges are weighted with the number of references to m2 that m1 contains (here just 1). Finally I've got a "virtual" memory vertex representing the stack, call it M0. Every Mi reachable from M0 mustn't be garbage collected.

Okay, now consider you want to add a memory block to the graph. If we keep the vertices inside a set, then the complexity of adding a memory block should be of O(log(n)). First question : Can we do better ?

Idem for deleting.

Now, I'm asked to use this algorithm with a reference counting mechanism in C++ (shared_ptr). Firstly, is the reference counter isn't redundant with the in-degree of a vertex ?

Secondly, the key idea is to use the best of the reference counter (O(1) deleting/adding) with the best of the garbage collector algorithm (cleaning reference cycles), but is the overhead of adding/deleting each node in the object graph isn't a bit non-efficient ?

What are the complexities of adding/removing in known garbage collector (java / C# / …) ?

Thanks !

Answer 1

Well... your premise is off. Known garbage collectors don't maintain much state actually, at most a couple bits per object and some structure but that's it. Instead, they build up some state at each collection pass and let it die at the end of the pass. This way, they need little (to no) instrumentation of the ownership relationships.