GUI Architectures
martinfowler.com
martinfowler.com
Should we be encouraging "correctness" here or is it ok that every new MV* framework reinterprets these patterns?
Fast forward to today and the UI can now be decoupled from the back-end technology so it is natural to evaluate whether or not the old patterns apply to new practices. In the old model, event based development and widgets had little advantage, event's where overly complex for a service oriented architecture, where a unit of processing is done and then a response is returned. It made little sense to bring in the indirection of event based programming. Further black boxing UI elements into widgets had little advantage because they could not directly respond to events that happen on the client side, in the old way, if a button was clicked, the server had to be notified and it would require complex routing to get that request back to a discreet widget that would have to be recreated or cached and looked up with every request. MVC with procedural C's where the best pattern available to deal with this reality.
Now jumping back to the present, we are not under the same constraints, code can be delivered to the client, the client can respond to events and the client can request processing and data from the back end. Based on this reality, procedural C's don't make a lot of sense, rather events and widgets become a more discreet solution. Designed well, a UI can actually rid itself of almost all procedural glue code external to widgets, via event routing and syndication. In this new reality there is little need to have a controller that says you do this and then you do this. Rather you can build widgets that say when this happens I will do this and when that happens I will tell everyone about this.
Given the dogmatic view of MVC being a best practice people sometimes try to draw parallels and call what they are doing MVC. A better description of MVC or widgets is separations of concerns which is still valid. You should not have JavaScript in HTML and you should not generate HTML from strings in JavaScript. Rather just like in MVC a widget should have an HTML template that represents the view. Some people call widgets micro-MVC which I suppose is correct, but to my MVC implies a controller orchestration a workflow in which I think we are trying to force definitions to not look like the guy that does not know what he is doing. That being said, I think trying to square peg round hole Event/Widget UI development into a MVC definitions is causing a lot of confusion. I think MVC is a relic of web 1.0 page/post programming. What people are really trying to say when they say MVC on the client, is separation of concerns which is still a very relevant and good practice.
I think the pattern you're referring to is more like a Model2 (JSP/Rails) style pattern.
The thing is now we try to explain evented systems in terms of MVC and I believe that we are trying to force the definition when what we are really trying to say is separation of concerns. Here is a good thread from that period that goes over how the two became jumbled together: http://c2.com/cgi/wiki?WhatsaControllerAnyway it's long the tldr; is smalltalk took the traditional view that controllers are independent of UI, they don't even know what the UI will look like. Where widget based systems such as swing where utilizing the same definition, but yet widgets where tightly coupled to the view they control by being the intelligent manipulators of a discreet view, but proper separation is observed in that all view elements exist in a HTML (or other UI) files, and the controller is a code-behind file, together they represent a discreet element.
Traditionally, this has not been considered MVC until the web UI started moving back to the client and now due to MVC achieving almost religious omnipotence people are wary of something not being MVC even if they have to make the definition match the practice.
on an interesting side note, I often wonder had MVC been as popular before the web, as it was after the web, would an event model have been chosen for the browser?
Sorry to reply to my comment but it is too late to edit it, on the above quote, I realized that I jumped around and it reads like I am implying that swing uses HTML, that was not my intent, I jumped to HTML in my mind because I was thinking about a pure view technology which HTML is probably the best example, I mixed thoughts there but it was not my intent it would have been more proper to say like HTML not HTML as it reads like swing utilizes HTML as it's view technology which was not the concept I was trying to convey.
For example: There is one widget (let's call it a Table) that displays data. The breakdown is as follow:
Model => Data View => Table Controller => Something that updates view based on changes in model (undefined originally)
Even if it wasn't limited to one widget, it was limited to a one simple View of the app (the other popular example is to use the DocumentWidget and DocumentModel "real-life" sample).
The original MVC was definitely not well-thought for rich desktop application PLUS automated testing with limited dependency to the Presentation layer (or view).
I believe Java SWING library utilizes MVC model at the widget levels.
(From the article: One of the central ideas of the Smalltalk-80 system was the ModelViewController (MVC) UserInterfaceParadigm?. Within that paradigm, the term "controller" had a well-defined, limited meaning.)
MVC today does not have a specific method or implementation and that was my point MVC has come to be known as a fuzzy definition for separations of concerns, as such I believe just saying separation of concerns is a better definition of what is desired. Because at some point people look at what was modeled in Smalltalk or Struts 1.0 which adhere to that concrete definition and become confused with the more nebulous definition that MVC has morphed into.
The age of an article has no bearing on how interesting a good hacker would find it.