Performance testing has shown that the finalize methods in the channel classes
impose a significant GC penalty upon server-type programs that must deal with
thousands of new open channels per second.  These finalize methods should be
removed.

--  xxxxx@xxxxx  2002/11/12

N/A

Yes.  --  xxxxx@xxxxx  2002/11/12

does that mean, that a socket is never closed when it isn't
closed within the user-code?

I for one think this is retarded... This is likely to lead to leaky 
channel issues. I *know* it is up to the developer to weed 
these out but they are hard to locate and it makes a lot of 
sense for the garbage collector to actually do its job and 
clean up the resource.

In so far as performance is concerned, work on optimizing the 
garbage collector so that *all* classes making use of 
finalizers will run faster, instead of hacking Channel objects.

Please at least make sure that the javadoc in all relevent places indicates the need for the channels to be explicitly closed. Many people assume or rely on finalizers cleaning up native resources eventually (even though finalizers aren't guaranteed to run), especially when there is not specific documentation warning that they won't be.

I would hope that the affected classes will register their new 
instances with a PhantomReference subclass and a 
ReferenceQueue. So that when the instance is reclaimed by 
the garbage collector, any associated resources can be closed 
when the PhantomReference is posted to the 
ReferenceQueue. Otherwise, resource leaks may occur.

A PhantomReference with a ReferenceQueue is a better way 
to track instance reclamation than finalize() methods.