RE: [picocontainer-dev] Bug in DefaultPico?

> -----Original Message-----
> From: Paul Hammant [mailto:[email protected]]
> Sent: Wednesday, December 19, 2007 4:25 AM
> To: dev-qxt/[email protected]
> Subject: Re: [picocontainer-dev] Bug in DefaultPico?
> 
> Mike,
> 
> >
> >> You're quite right - its lame as is.
> >> What do you think it should be changed to ?  One using plain refs
> >> rather than
> > integers ?
> >
> > I don't see why we shouldn't use plain refs.  As long as the refs
> > are cleared
> > after dispose() -- I mean, perhaps something else will show itself
> > as a better
> > option later, but for now, I'd take care of one step at a time.
> 
> Oh yes I remember.  It was to do with garbage collection.  If they are
> in a list/set as refs then they are likely to be prevented from being
> eligible for GC.  I can't remember why that would ever arise ... I
> mean when you'd want to dispose of CAs but not the container that
> contained them.. :-(

Yeah, I was thinking about that -- Let me play with it for a day or two and I'll
try to come up with a properly encapsulated alternative.

> > I'm also thinking of moving the start, stop, and dispose operations
> > to a
> > Visitor.  (The visitor doesn't have to be transient, so it would be
> > possible for
> > it to keep track of what CA's and what Child containers were
> > started.)  It
> > should help move several of the additional data structures sitting
> > in Pico to
> > other more reusable objects.
> 
> Meh.
> 
> With Pico2 I deleted Aslak's visitor abstraction and lifecycle impl of
> it.  I did not like the way that from root it would traverse the
> entire tree.
> 
> With Pico, you can chain containers together.  Each in a parent,
> child, grandchild situation might have a different set of behaviors
> incl lifecycle.  By each Pico instance delegating to the next's
> start() methods, there a decent mechanism to subvert the traversal.  I
> think that superior - even though its slightly more code.

True -- And even then I never thought about traversing child containers.  But
since there's so much data being used to handle lifecycle events, I still think
that the behavior needs to be encapsulated.  I'll work up a proposal and post
the diff to jira.

									-Mike




---------------------------------------------------------------------
To unsubscribe from this list please visit:

    http://xircles.codehaus.org/manage_email