commons-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Phil Steitz <>
Subject Re: [pool] 2.0 factory interfaces WAS: Re: svn commit: r1506685 - in /commons/proper/pool/trunk/src: main/java/org/apache/commons/pool2/ main/java/org/apache/commons/pool2/impl/ test/java/org/apache/commons/pool2/ test/java/org/apache/commons/pool2/impl/ test/java/org/apache/commons/pool2/perfo...
Date Sat, 03 Aug 2013 15:15:52 GMT
On 7/30/13 9:16 AM, Phil Steitz wrote:
> I have started working on this.  Should have something to commit at
> least for GOP in the next day or two.

I am stuck on SoftReferenceObectPool.  I have been able to handle
everything else collapsing down to just Pooled(Keyed)ObjectFactory /
Base(Keyed)ObjectPool.  To stick with (and get factory / monitoring
benefits of) PooledObjectFactory / PooledObjects, I need to create a
version of DefaultPooledObject that wraps SoftReferences and these
have to be mutable - i.e., when an instance is returned a new soft
reference is created and stored in the pool, but we want to maintain
PooledObject state.  To do this, I need to be able to recognize the
returning object.  I could do this by searching the pool, but that
will cost something.  On the other hand, the lack of search now
means returnObject is not idempotent (like it is in the rest of the
2.0 pools).  Multiple returns will put multiple soft references to
the same object in the pool.

So, I would appreciate some feedback on the following options:

1) Add a search for the returning object and create a
PooledSoftReference subclass of DefaultPooledObject within
SoftReferencePool to be stored in the pool.

2) Keep PoolableObjectFactory just for this pool and leave it alone,
adding a warning to returnObject (not a new issue, just different
from other 2.0 pools).

> Phil
> On 7/29/13 10:56 AM, Phil Steitz wrote:
>> On 7/24/13 1:06 PM, Mark Thomas wrote:
>>> On 24/07/2013 21:01, wrote:
>>>> Author: markt
>>>> Date: Wed Jul 24 20:01:34 2013
>>>> New Revision: 1506685
>>>> URL:
>>>> Log:
>>>> Create two new factory interfaces that work with PooledObject instances rather
than Object instances and switch Gop and GKOP to use them.
>>> One area I'd particularly like some comment on is PooledObject &
>>> PooledObjectImpl.
>>> I considered just having a single PooledObject implementation class in
>>> o.a.c.pool2 but decided that as implementation it belonged in
>>> o.a.c.pool2.impl. That lead to needing PoolImplUtils.
>>> I'm not completely happy with the current arrangement but neither have a
>>> found a better one. Thoughts?
>> I wonder if we really want / need to retain the original "dumb" (not
>> in the sense of bad design, but no tracking) pooling infrastructure
>> from 1.x.  Thinking about making it easy for users to grokk the
>> setup and get a GOP or GKOP working, I wonder if it might be better
>> to drop the base classes and just start with simple, refactored pool
>> and factory interfaces that create and manage PooledObjects
>> directly.  Users will still only absolutely *have* to implement
>> makeObject in their factories and the default code will take care of
>> everything else.  So you just end up with PoolableObjectFactories
>> sourcing and managing PooledObjects.  GOP, GKOP still return
>> unwrapped objects via borrow and there is an
>> AbstractPoolableObjectFactory with makeObject abstract and the rest
>> provided.  I have not played with this yet (hopefully will have some
>> time in the next couple of days), but I wonder if it might not be
>> better / simpler.  Also, adding methods to GOP, GKOP that return
>> PooledObject instances (maybe stripped down) might be useful to
>> clients.  Sorry if above is naive / old ground.  I just want to make
>> sure what we end up with is a simple as possible.
>> Phil
>>> Mark
>>> ---------------------------------------------------------------------
>>> To unsubscribe, e-mail:
>>> For additional commands, e-mail:

To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message