The public interface of the Mono GC is fairly limited, and
its the only one that embedders should be using:
mono_gc_reference_queue_new (mono_reference_queue_callback callback)
callback used when processing collected entries.
Create a new reference queue used to process collected objects.
A reference queue let you add a pair of (managed object, user data)
using the mono_gc_reference_queue_add method.
Once the managed object is collected callback
will be called
in the finalizer thread with 'user data' as argument.
The callback is called from the finalizer thread without any locks held.
When a AppDomain is unloaded, all callbacks for objects belonging to it
will be invoked.
the new queue.
The bridge is a mechanism for SGen to let clients override
the death of some unreachable objects. We use it in monodroid
to do garbage collection across the Mono and Java heaps.
The client can designate some objects as "bridged", which
means that they participate in the bridge processing step once
SGen considers them unreachable, i.e., dead. Bridged objects
must be registered for finalization.
When SGen is done marking, it puts together a list of all
dead bridged objects and then does a strongly connected
component analysis over their object graph. That graph will
usually contain non-bridged objects, too.
The output of the SCC analysis is passed to the
`cross_references()` callback. It is expected to set the
`is_alive` flag on those strongly connected components that it
wishes to be kept alive. Only bridged objects will be
reported to the callback, i.e., non-bridged objects are
removed from the callback graph.
In monodroid each bridged object has a corresponding Java
mirror object. In the bridge callback it reifies the Mono
object graph in the Java heap so that the full, combined
object graph is now instantiated on the Java side. Then it
triggers a Java GC, waits for it to finish, and checks which
of the Java mirror objects are still alive. For those it sets
the `is_alive` flag and returns from the callback.
The SCC analysis is done while the world is stopped, but
the callback is made with the world running again. Weak links
to bridged objects and other objects reachable from them are
kept until the callback returns, at which point all links to
bridged objects that don't have `is_alive` set are nulled.
Note that weak links to non-bridged objects reachable from
bridged objects are not nulled. This might be considered a