james-server-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Simon Funnell <simonfunn...@googlemail.com>
Subject Alien Ideas
Date Fri, 20 Nov 2009 00:27:28 GMT
Stefano Bagnara wrote:
> I'm not sure that "next generation frameworks" means "declarative logging.
Neither do I, but by definition one generation follows the next and the 
current generation doesn't properly address the issue while the next 
does. Its not just logging however, there are other things, like 
inheriting exception handling policies, security and other cross cutting 
> IMHO declarative logging (the way you described it) cannot and will
> never replace standard logging. It could be used together, but not
> replace all of logging.
I disagree of course, but only in principle. It's a simple matter of 
adding logging(and other things) 'after' the functionality instead of 
'before' it in the inheritance chain.

Many components in James extend AbstractLogEnabled putting it 'before', 
I realise this is a stumbling block but its not a massive problem.
> It would be a PITA to mantain that style of declarative logging. 
Maybe, but I don't think so. If everything is well documented, it 
shouldn't be a problem.
> if you have more documentation, examples, code,  evangelization
> articles, to help me why declarative logging will be used in the next
> generation framework I'd be very happy to read it. I'm very interested
> in improving the logging aspect of libraries/application as I feel the
> logging one of the most problematic issues in current design. I don't
> like per class logging (Log.getLogger field), either static or per
> instance and I prefer injecting logging services, but I still don't
> see how to completely remove logging in favor of a declarative
> approach. 
How about writing components with no references to loggers or logging 
whatsoever? (Your mind will instinctively reject this idea...as time 
goes by you will accept it as the best way...a bit longer and it will 
become the norm....its already the norm in some organisations....)  It's 
really simple, you add logging (and other things) after, not before. 
This is the point of components, they are not classes, a good component 
model supports coupling/binding of functionality at any time. In my eyes 
(and with the design I am following) a component 'is' a framework, all 
of its elements are pluggable and you can change individual elements, or 
add more elements, on a per request basis.

Generational inheritance is not suitable for everything.

Look at it this way, if you view James as a community of components and 
initially this community does not have a logging component, so sorts 
mail without logging. A logging component now comes along but because we 
are using generational inheritance the new logging component literally 
has to mate with existing components to create a new generation of 
components that have the logging gene{for example, getting a logger 
property (gene?) by extending the parent AbstractLogEnabled). This is 
not want we want, instead what we want is our new logging component to 
come along and share its inherent ability with the rest of the 
components in a transparent fashion. In this alternative scenario other 
components in the community that make up the application inherit logging 
capabilities through 'nurture' rather than 'nature', i.e. components can 
become more sophisticated by inheriting from peers as well as parents.
> Can you take a class from our repository and show how to
> refactor it to use declarative logging (also including the declarative
> logging part needed to produce the same log).
> Stefano
This is probably a good idea and something I have been looking at doing 
but I would like to continue studying the code base at present while I 
get a clear idea of the way forward. The SMTPServer package looks a good 
target, it has separate classes for each command and is the kind of 
thing I am looking for. Would it be a good idea to have a wiki page with 
some clear terms and definitions like what is a container, what is a 
framework, what is a component, etc? I think it would be good at least 
summarise their responsibilities and relate them to James concepts. I 
think it mentions something along those lines on the wiki somewhere and 
I don't mind starting it off. You may already have definitions for these 
terms and if so please point me in their direction.



P.S. I am using logging as an example of cross cutting concerns, which 
is a much more general problem.

>> Regards,
>> Simon
>> Stefano Bagnara wrote:
>>> Simon,
>>> Sorry for being pragmatic, but your framework will have to deal with
>>> this comments if you want to start telling "our framework" instead of
>>> "my framework".
>>> If the framework needs so much text to be explained then it will
>>> probably be not good for James. We need something that people already
>>> know, or alternatively something that people is able to grasp without
>>> reading pages of docs. Maybe you want to try with concrete examples,
>>> diagrams, but not so much text. I admin I read all you wrote and I
>>> still don't understand how your framework works. Bad thing.
>>> You propose a new container framework using interfaces to be
>>> implemented. in 2009 this is a bad practice, currently frameworks use
>>> @annotations, AOP, agnostic (pojo) dependency injection. If we want an
>>> interface based framework then Avalon does a damn good job here,
>>> already.
>>> You also will need a good understanding of other frameworks so to
>>> publish a comparison and use common terms defining what makes your
>>> framework special. What is the most similar framework around, what
>>> does your framework do better? Does your framework try to solve
>>> container issues (class loading, component isolation,
>>> componentreloading, deployment) or application assembly (dependency
>>> injection, component configuration, service discovery/linking,
>>> configuration)?
>>> Stefano
>>> 2009/11/18 Simon Funnell <simonfunnell@googlemail.com>:
>>>> Hi,
>>>> By the way, its only 'my' framework by virtue of the fact there is no
>>>> 'our'
>>>> to refer to. I could probably refer to it as 'the' framework with you
>>>> being
>>>> able to infer what I am referring to from experience, but its actually
>>>> called 'Platformed' and I intend to make it open source. Also, after some
>>>> research and reflection, I actually think its much more suitable (it was
>>>> designed for this purpose). There is no escaping some of the differences
>>>> in
>>>> approach however. I am quite willing to invest some time implementing
>>>> some
>>>> things which you can take a look at, it really wouldn't take to long to
>>>> modify some existing code. Avalon defines interfaces such as Loggable(?)
>>>> but
>>>> there is no equivalent concept with what I am doing because all things
>>>> are
>>>> inherently loggable, components do have to be built in a way that enables
>>>> loggability (and lots of other things) though. It basically means
>>>> breaking
>>>> down some of the existing code into smaller chunks that can pieced
>>>> together.
>>>> These smaller chunks could be recomposed into an alternative
>>>> implementation
>>>> of existing classes and I could also use these smaller chunks in a
>>>> different
>>>> arrangement. As I say, I could implement an example which would allow
>>>> people
>>>> to make an evaluation about any sort of decision or what not.
>>>> I think you might actually like it, it was written a while ago and its
>>>> only
>>>> by looking at it closely again that I remembered how much work I put into
>>>> it. My last email didn't accurately reflect the whole architecture, the
>>>> micro virtual machines I have touched on are extensible through
>>>> 'operations'. Current James classes have methods such as initialize,
>>>> service, destroy etc, I would have to replace these methods with actual
>>>> classes (which implement an interface called Operation). Operations are
>>>> stateless instances that implement functionality, micro virtual machines
>>>> provide the operations with a context upon which to operate. Different
>>>> micro
>>>> virtual machines can have contexts like application/session etc depending
>>>> on
>>>> what's required. If you wanted to log the state of any given machine, at
>>>> any
>>>> stage, you just plug in a logging operation at the necessary place. This
>>>> can
>>>> be done dynamically, like for example an individual client matching some
>>>> criteria. The smaller things are, the more reusable they are. I could
>>>> very
>>>> easily create an adapter that takes existing mailet/matchers and
>>>> transforms
>>>> them into Operations for  micro virtual machines processing the spool.
>>>> I've probably said too much/too little so I wont post again until I get
>>>> some
>>>> feed back, or have something working, which ever comes first. I am still
>>>> interested in producing some documentation as well, suggestions are
>>>> welcome.
>>>> Thanks,
>>>> Simon
>>>> ---------------------------------------------------------------------
>>>> To unsubscribe, e-mail: server-dev-unsubscribe@james.apache.org
>>>> For additional commands, e-mail: server-dev-help@james.apache.org
>>> ---------------------------------------------------------------------
>>> To unsubscribe, e-mail: server-dev-unsubscribe@james.apache.org
>>> For additional commands, e-mail: server-dev-help@james.apache.org
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: server-dev-unsubscribe@james.apache.org
>> For additional commands, e-mail: server-dev-help@james.apache.org
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: server-dev-unsubscribe@james.apache.org
> For additional commands, e-mail: server-dev-help@james.apache.org

To unsubscribe, e-mail: server-dev-unsubscribe@james.apache.org
For additional commands, e-mail: server-dev-help@james.apache.org

View raw message