Re: Solution

[Simon Funnell]

Stefano Bagnara wrote:
Please, take real code from James and show us the real refactoring you propose.
There's no way we'll keep reading so big abstract discussions.
  
I don't think this is the right approach, forget about my framework and 
declarative logging, its missing the point.
Instead leave things how they are create some high level documentation 
that describes the roles and responsibilities of James components.
This documentation will highlight what needs to be done and is a perfect 
basis for designing a set of interfaces.
Once a good set of interfaces in designed, then is the time to 
migrate/implement things.
In this way, the current effort can continue as is and re-factoring when 
it comes is much easier.

Nobody has to do anything at this stage (well very little), let me learn 
more about James while I create some documentation.

If I misunderstand something, those who know James intimately can 
correct me. This will be useful for everyone, including new users.
I don't think that what I see here will reduce the time to write code.
Extracting logging into special classes is a PITA most time. This will
require big refactoring in order to allow an external class to log
what you want to log. 
That's the whole point, pluggability. Yes it takes longer to write 
initially, but you don't (or very rarely) need to write it again, it 
does save you time in the long run. You write your logging code once and 
reuse it.
So you have to write code thinking about what is
needed for logging, and then later you have to apply logging. 
Yes, but its worth the trouble, because its more flexible.
IMHO
this is a no go. But I'll be happy to change my mind when I see a damn
clear code example.
  
Its not about logging alone, its all cross cutting concerns. I will help 
document the James architecture and develop interfaces but I am not hacking.
Checkout James trunk, take a module, refactor it the way you think it
is "cool". Open a JIRA and attach your diff. Everyone here has more
time for diffs/jira than for discussions.
  
Please don't ask me to code first and design later, I learnt not do that 
a long time ago. (Wish I learnt it when I first heard the Tortoise and 
Hare story though).
You should better use your discussion time to create a website for
your ideas and make your message documented in that website, so that
people wanting to set the new standards will have something
*organized* to read. If you want to set new standards you will need
some clear code example that will convince anyone that it is the right
direction.
  
This is probably true and I am sorry for confusing matters, but again 
its not relevant. James components are not as reusable as they could be.
This is not a problem as such, things still work. And by the way, its 
not my message. I am only stating facts.

As a side note, I know many project using AspectJ, but none of them is
using it for logging. I think there something to learn there.
  
AspectJ has its uses, but its not right for all cases.


At this stage I will repeat what I have said before, I think it is worth 
working me on some documentation.
In this way I can add value without stalling the current effort.
James is more important than my framework at present, that can come later.

Regards,

Simon

Stefano

2009/11/21 Simon Funnell <simonfunn...@googlemail.com>:
  
Hi,

Sorry, I will keep it under 'Solution' for now.

Norman Maurer wrote:
    
Im still not agree with you about that. Imho logging should be part of the
"main code". And why is this any better then injecting the Logger like every
other object. In trunk we even use JSR250 for that so every Franework which
knows howto handle this Standards should just work...
      
If you want to have declarative logging and other wonderful such features,
you can't put it in explicitly. That's not my opinion, its a fact, a known
self evident fact, but its not all about logging. Components can still work
with standards, that's not a problem. There is one thing about standards to
note, not every one follows standards, because there are those that set the
standards (if you are ahead of the game then there is no one to follow, in
this case logic clearly states that you 'are' the standard). Why not have
James set the standards while still adhering to others, its not impossible,
in fact its easy, it just requires engineering discipline. For example,
Platformed is not just a bunch of code, it encourages good practice. It
defines six roles that would be useful to James and Tomcat (both of which I
use heavily). They are split into three groups of high, medium and low. The
two low level roles are 'engineer' and 'developer'. These roles define clear
responsibilities and while close in nature they do different things.
Engineers do lower level things and developers higher level. This
distinction is not made (or is not apparent to me) in James and it would
certainly benefit from making these distinctions (an improved
architecture/API). In Platformed for example, developers don't have to worry
about concurrency issues because the engineers solved the problem for them
'once'. The engineers effectively create a structured development
environment that allows developers to focus on real world problems while
they take care of machine level problems. It's a proper separation of
concerns.
    
Are you not bored of explicitly writing logging code in every component?

        
Nope im not bored.
      
Ok, if you write less code it will take less time and you will earn more
money or if you have time left over, you can make it a better program. If
other people are doing it and you are not, they will probably get the
contract because they can get it done quicker. That said, money isn't
everything, its also fun making software. You have heard of the 'write once'
principle right? If you are writing logging code over and over again, you
probably need to understand something important. We designed computers with
software languages and they are very good at 'automating' things, like
automatically adding logging code to your software. You are manually doing a
task that can be automated!!! Is this not very obvious, its madness!? Sorry
to be sarcastic, I am a really nice guy but I can be a firebrand at times :)
it does bug me when software developers do not utilise software.

I have been looking at the HeloCmdHandler and it is problematic because it
mixes principles(its active and passive). Passive components do not want
logging code within them, but its ok to put logging code in active ones. The
easiest way to do this is to use a good tool that solves the problem for
you, or create two classes where you usually create one (that reflect the
active and passive principles). The first class will have methods that
implement stages of processing flow (passive), then another class (or any
number of classes) can call the implemented methods in any order they like
(active). For example:

//Passive
class ThreePageBook extends AbstractBook{

public ThreePageBook (){}

  public void pageOne(){
      ...
  }     public void pageTwo(){
      ...
  }
  public void pageThree(){
      ...
  }
}

abstract class AbstractReader{
    private final Book book;

  public AbstractReader( Book theBook ){
     book = theBook;
  }

  public abstract void read():
}

//Logging implementation - active
class LoggingReader extends AbstractReader{

  private final Logger logger;

  @Inject
  public LoggingReader( Book theBook , Logger theLogger ){
      super( theBook );             logger = theLogger;
  }
  @Override
  public void read(){
      logger.log( "About to read page " + book,getNextPageNumber() );
      book.pageOne();
      logger.log( "About to read page " + book,getNextPageNumber());
      book.pageTwo();
      logger.log( "Done reading page " + book,getLastPageNumber() );
      book.pageThree();
      logger.log( "Done reading page " + book,getLastPageNumber());
  }
}

//Non-logging implementation - active

class NonLinearReader extends AbstractReader{

  @Inject
  public NonLinearReader( Book theBook ){
      super( theBook );
  }
  @Override
  public void read(){
      book.pageTwo();
      book.pageThree();
      book.pageOne();   }
}

Guice is very good, but its not an architecture. Please let me help, I am
skilled, work very hard and I am available for the next six weeks so I can
dedicate myself to James. It should be obvious from the above that I am
offering practical, straight in front of your face, solutions, its easy to
add logging and the like after the functionality. Logging code doesn't need
to be removed, just put in the right place. It's a proper separation of
concerns, good component design.

Although I code in languages from the bottom of the stack (nearly the
bottom, I have a novel understanding of machine code) to the top I am
usually employed as an architect so tend to produce interfaces (metaphysical
constructs) rather then implementations (physical constructs). Some places
automate the implementation process by generating and testing all the
various component arrangement combinations looking for successful
implementations, most implementations fail and are discarded and the ones
that are successful are usually a smallish number of specialist
implementations that between them solve all generalised cases (in principle
like one of those artificially intelligent chess programs looking ahead
through all possible combinations). They are basically programs that write
programs, very cool. But I digress.

I think the way forward is documentation for the reasons given in the
previous email, I can add value without stalling the effort. I've known
cases where good documentation makes a success of bad software, powerful
stuff. Obviously you want both good documentation and good software. :) You
don't need to use AOP frameworks, its really about component design and
structuring the code in a more modular fashion.

But there it ends. IMO a framework cannot provide an easy replacement
for the "debug" and "info" level where you want to log certain states
or events that may be of interest to the user.


This is half true, no framework cannot solve the problem alone, the
components have to solve it as well. They are two sides of the same coin.

The root of the problem with James is that it implements its own internal
framework through inheritance, with things like AbstractJamesService.

This is adding features before that should be added after. No framework can
provide an alternative, because its built in.

For example, POP3Server doesn't need to know anything about the
AbstractJamesService, but it extends it.

This is the wrong way round, it needs to be 'inverted'.

The issue is architecture/API design.

Regards,

Simon

P.S. I am sorry its long again.




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


    

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


  


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