RE: Why can't each Aggregate Root have its own collections of entities/ARs?

I was thinking about it. A system can have multiple domains?

To: [email protected]
From: [email protected]
Date: Thu, 29 Aug 2013 10:19:42 +0200
Subject: Re: [domaindrivendesign] Why can't each Aggregate Root have its own collections of entities/ARs?
















 



  


    
      
      
      
  
  
    
No, I would prefer to have a list of
      continents

      

      
        
          class World {
        
        
            List<Continent> continents
              = ...;
        
        
          }
        
      
      

      
      Because decoupling world and continents via id's and messages seem
      like overkill. 

      

      Now, if we were to switch from the domain of topography (the
      structure of the world), to, say, financial data associated with
      the topography (the financial domain), then I can see a
      decoupling. The Topography domain sends messages to any other
      domain, like Financial, or Demography or whatever, about changes
      in its model. 

      

      Then why not all one domain model? Because the Topography model
      may hold detailed information on the continents, countries, their
      borders, how to render, etc which has no meaning in other domains.
      In the other domains the entities just "are" a country without all
      the extra weight. By decoupling and creating adapters, each domain
      may evolve independent of the others.

      

      Tom

      

      

      

      

      

      

      On 2013-08-29 10:06, Jorge Branco wrote:

    
    
       
      
          
            
              Hi again!
              

              
              Tom,
              

              
              
                
                  Now if you take that
                      further and say that a World needs to know how
                      many Countries each Continent has to perform it's
                      responsibilities, the only data that the World AR
                      needs to hold is a map of ContinentID to Country
                      count, again maintained via domain events.
                
              
              

                
              

              
              If I understood you correctly, you're favoring a
                solution along the lines of:
              

              
              
                
                  class World {
                
                
                    Map<ContinentId,
                      Integer> continentsSize = ...;
                
                
                  }
                
              
              

              
              that from its inception on is updated via domain
                events. But if this is indeed the case, how would you
                deal with this if World needed also other things from
                continents other than size? Would you have a map for
                each one of the needed features?
              

              
              jorge
            
            

              

              On Thu, Aug 29, 2013 at 7:54 AM,
                Tom Eugelink <[email protected]> wrote:

                
                  
                     
                    
                      
                        
                           

                          Hi Dennis,

                            

                            Good question; neither I guess, I'm voicing
                            an opinion.

                            

                            In an earlier post I said that I understand
                            the reason for this disconnected model, and
                            I really do see the upside. I'm not closed
                            minded, but I also see the downside. And I'm
                            really in doubt if the gain outweighs the
                            loss.
                            
                              

                                

                                

                                

                                

                                

                                On 2013-08-29 08:41, Dennis Traub wrote:

                              
                            
                          
                          
                            
                                
                                
                                  Tom,
                                  First off: Are you asking
                                    a question or trying to prove a
                                    point? You responses to everyone
                                    trying to answer suggest the latter.
                                  That said, in my opinion
                                    Simon's answer was the first one in
                                    this thread that really broke it
                                    down to what totally makes sense.
                                    And what he proposed would radically
                                    simplify rather than complicate
                                    things. OO modelling is best done
                                    with an open mind, not a pre-defined
                                    set of "that's how we did it all the
                                    time"
                                  Best
                                  Dennis
                                  On Aug 29,
                                    2013 7:43 AM, "Tom Eugelink" <[email protected]>
                                    wrote:

                                    
                                        
                                        
                                          
                                            
                                               

                                              On 2013-08-29 00:03,
                                                Simon Fox wrote:

                                              
                                              Now if you take
                                                  that further and say
                                                  that a World needs to
                                                  know how many
                                                  Countries each
                                                  Continent has to
                                                  perform it's
                                                  responsibilities, the
                                                  only data that the
                                                  World AR needs to hold
                                                  is a map of
                                                  ContinentID to Country
                                                  count, again
                                                  maintained via domain
                                                  events.
                                              

                                              This of course is true,
                                              but it very much feels
                                              like "early
                                              optimalisation". I once
                                              knew a client (insurance
                                              company) who optimized its
                                              database for flexibility
                                              so far, that they ended up
                                              with two tables: one
                                              contained the datamodel,
                                              the other contained the
                                              values. Extremely
                                              flexible... Totally
                                              unusable, mainly because
                                              it performed awfully and
                                              newcomers had a very hard
                                              time getting date in and
                                              out of it. 

                                              

                                              Now, I'm not comparing
                                              that situation with the
                                              current discussion
                                              directly, but it
                                              illustrates that there
                                              needs to be a balance
                                              between optimization (on
                                              what ever aspect) and
                                              usability. The
                                              disconnected nature of a
                                              domain models can be a
                                              good thing, but if we go
                                              overboard, it can become a
                                              bad thing. The question is
                                              to find the sweet spot. My
                                              gut, and often in software
                                              development guts are
                                              involved, tells me that
                                              the recognition of
                                              separate DM's is wise, but
                                              bringing that down to the
                                              level of one or a few
                                              entities (like a invoice
                                              and its rows, aka
                                              aggregate roots) makes my
                                              gut go uneasy. 

                                              

                                              Basically the whole
                                              domain, inter and intra,
                                              becomes message driven.
                                              And such complete loose
                                              coupling also leads to
                                              complexer code, harder to
                                              follow business logic, and
                                              new people not easily
                                              being able to see (or
                                              trace) the big picture.
                                              Personally, that does not
                                              sit well. Like that
                                              database example; almost
                                              completely message driven
                                              software??? Hmmmmmm.

                                              

                                              Tom