Pull vs. Push

By: Jon S. Stevens
$Date$

This document is to explain a different type of philosophy for working with Context based MVC tools like WebMacro (WM) and Velocity (V) than I think that the Turbine community is used to. The requirement for this philosophy has come up recently during Scarab development and I would like to share and document it with you all. Many thanks to Brian B for being patient with me and trying to explain exactly what he wants to see. So, let me start off by giving a little bit of history and then moving on from there...

The current encouragement that Turbine gives to developers is to do a mapping between one Screen (Java) and one Template (WM or V). The way it works is that you build up a Context object that is essentially a Hashtable that contains all of the data that is required to render a particular template. Within the template, you refer to that data in order to format it for display. I will refer to this as the "Push MVC Model." This IMHO is a perfectly acceptable, easy to understand and implement approach.

We have solved the problem of being able to display the data on a template without requiring the engineer to make modifications to the Java code. In other words, you can modify the look and feel of the overall website without the requirement of having a Java engineer present to make the changes. This is a very good step forward.

However, it has a shortcoming in that it makes it more difficult to allow the template designer (ie: a non programmer) the ability to move information on a template from one template to another template because it would require the logic in the Java code to be modified as well. For example, say you have a set of "wizard" type screens and you want to change the order of execution of those screens or even the order of the fields on those screens. In order to do so, you can't simply change the Next/Back links you need to also change the Java code.

Another example of this is if you have a form on a page and you want to be able to split that form across several steps across several pages. In the current model, you would have to potentially write/modify several Java classes. In reality, this should be something that is easily modifiable by just a single template engineer who doesn't necessarily even know Java.

There are several considerations that one must take into consideration regarding the design of a system to provide this level of abstraction. For example, when someone submits the form and there is an error. Proper UI would suggest that you should re-display the page with the previous form data filled in as well as an error message that details the problems. You also need to consider the ability to display the same form that may be pre-populated with information from the database instead of as an error from the user.

So, beginning with Scarab, we are going to try another model which I will describe as the "Pull MVC Model". What this entails is the ability to create an object that is able to "pull" the required information out at execution time within the template. Thus, it becomes the job of the template designer to understand the API of objects available to him/her to take advantage of.

Instead of the developer telling the designer what Context names to use for each and every screen, there is instead a set of a few objects available for the template designer to pick and choose from. These objects will provide methods to access the underlying information either from the database or from the previously submitted form information.

Gone are the days where one would have a Java class that would be responsible for building the context up for the template. Instead, there would be a single base class that places the few objects into the context for every request and there is a documented API that the template designer can refer to in order to access the information that he/she needs to get at. Of course this information is only retrieved when requested and can also be managed in a cache for reuse.

By moving to a "Pull" model, it becomes possible to more easily achieve complete independence from the Java engineers in order to not only change the look and feel (UI) of the site but also the information architecture (IA) layout and flow of the site.

While this puts more requirements on the Java engineer to make sure that the template designer has a well defined API, it will save the Java engineer many hours further down the road when the client requests IA changes because now the responsibility is on the template designer to make the changes. Therefore the increased initial developer time is justified in order to realize the long term goals of the project.

I hope that this makes sense to everyone. I'm sure that some of you are probably saying "well, duh!." However, this is really not the model that has been encouraged so far within Turbine nor within other products such as XMLC (which actually operates *only* on the Push model), so I believe that it is somewhat uncharted territory for some people. The only products that I can think of right now that encourage this is JSP taglibs and Tea, however, I still feel as though they have missed the boat on this in one way or another.

Comments are appreciated. Please post them to the Turbine mailing list.