Hi Alan,
On 02/23/2016 04:33 PM, Alan Bateman wrote:
On 22/02/2016 21:41, Mandy Chung wrote:
:
I added new method to Cleaner:
public boolean helpClean() { }
I’m in two minds of making this a public method. An explicit way to
enqueue pending references as well as invoke Cleanable::clean. If it
goes in, the method name needs to be renamed.
I'm also not sure that it's a good idea to have this be part of the
Cleaner API. If feels like it's exposing too much of the internals. Is
it really needed?
Without it, the DirectBufferAllocTest fails.
As John Cage (the character from Ally McBeal) would say: "I'm puzzled!".
My previous iterations used similar method on the internal CleanerImpl
class and there were suggestions to expose the functionality as public API.
I don't think this method exposes too much of internals. In the sense
where it would prevent any possible future rearrangement of the
internals or evolving of the Cleaner API. It doesn't talk about any
queue(s) or ReferenceHandler thread. It keeps the Cleaner implementation
details private. It just enables giving CPU time to assist with cleanup
and if used as hinted in @implNote, provides back-pressure to allocating
threads. I think this is a nice way of solving the problem of balancing
allocation with deallocation. The same problem concurrent GC algorithms
have, like G1 or CMS. If they can not keep-up with allocation threads,
they pause them and do the cleanup while the world is stopped.
Regards, Peter
