Thursday, June 01, 2006

Agile Development Ramblings: Part III
The Development

So far I've said a lot about the process, but nothing about the actual development. No prizes for guessing that this was different to my previous experiences too.

Development Environment Setup

We had to do this ourselves, so no time was wasted. From day one we had a server and by day two we were building both it and our development desktop systems with the software we required. We had few NFR's in this department apart from the requirement to use Websphere Portal and JSR168 Portlets. We also had to use Rational Application Developer 6.0 as our development IDE, and make sure our portlets were accessible and internationalisable. I do not think it would have been a problem had we had more NFR's, as long as the required media were made available to us. The fact we did not have to rely on anyone else; or an internal process was a boon.

We were left to our own devices. With access to the internet and an IBM Partner Web account we got to work. The systems we were given had nothing installed on them aside from Windows. We quickly decided to use Maven as the build and project management environment, MySQL and Hibernate for persistence, Subversion for configuration management, and CruiseControl for continuous integration. Each of the tools we used was picked after discussion between myself and the other developer. Installation tasks were split down the middle, with Notepad “what I did” notes being created and shared as we went.

This time didn't go entirely smoothly. RAD took ages to deploy and upgrade and we gave up getting the Portal Unit Test Environments (UTE's) to work on our desktops. (This turned out to be due to a known bug.) This however introduced me to the highly pragmatic approach to development any agile approach necessitates. You've got a problem; you have a think about what the relative costs and benefits of something are; if it's worth it, and doesn't endanger the delivery of features, go for it; if not, don't. If this is a real problem, you can flag it to the rest of the team and note it in the “Obstacles” section in the Sprint Plan. After such analysis, none of our problems were significant enough to reach the plan. In the case of the UTE's we decided that due to the small size of the team, we could use the shared server running a single instance of Portal and a single database for deployment and testing. It would require us to communicate constantly to make sure we didn't tread on each others toes, but this actually helped in 99% of situations. We never ended up developing in a vacuum.

All this meant that by the end of the first week we had an environment we were happy with, which we understood and would support our chosen approach.

The Coding

Again we took the agile approach. The aim was always to be developing and delivering as much as possible. At the outset, we set few constraints on ourselves. What we did have covered coding style, SVN commit protocol, Unit Testing coverage and approach and division of labour. Later on, as required, we bolted on Checkstyle and PMD, JWebUnit and JSTL. A unit testing protocol was forced on us as the default Maven target to produce our model jar file (“jar:deploy”) required that all unit tests passed. This proved invaluable and frequently caught problems in their infancy.

Architecture wise, we kept it light. Very light for someone like me who is a trained JEE architect. For the first two weeks, “MVC” was the only pattern that got mentioned. (We didn't use any MVC framework.) In the second iteration we introduced a Session Facade and in the fourth I got to use an Observer. We were lucky in that most of what we had to do was very simple. It was a no-brainer to split the code into two projects – one for the Model and another for the View / Controller. Again Maven made this simple. This Agile approach to architecture was a real revelation to me, and took some getting used to.

Build and Project management was light too. Maven was a boon in this respect. We only had to write thirteen lines of bespoke script in one project (to copy our Hibernate hbm.xml files from one place to another) and twenty in another (to call XMLAccess to deploy our portlet war to the Server). Everything else was already there for us, provided by the Maven plugins.

Maven also supported the project management, providing an automatically generated project website (via the “site” target) which allowed us to publish our how-tos to the wider team as we produced them. At the end of the development cycle, this proved an invaluable central repository for everything we produced over the proceeding thirteen weeks.

It is worth noting some of the practices I became familiar with during the development phase. The most fundamental was the aim to “write it once”, better still use a plugin or generate it. A good example follows:

Though we used MySQL as our data store, we never had to write any SQL. We didn't even have to write the hibernate mapping files. Instead, we generated them with XDoclet tags. To get from the tags to the schema, we called pre-existing Maven targets (“xdoclet:hibernatedoclet" and “hibernate:schema-export”) to generate our schema. This not only meant we got things going more quickly, and had fewer bugs (less code), but also we had automated what would otherwise have been very commonly executed task. Most importantly, these repetitive steps were taken out of our hands so there was less chance of losing valuable development time debugging what may turn out to be user error.

As with the setup phase I also had to focus 100% on delivery. I constantly asked myself “what is the value of what I am doing? Is it meeting the delivery goals? Is there something of more value I could be doing? Is there a better way to do what I am doing? Is it over redundant complicated / over engineered?”

All in all, we completed our six Sprints having delivered all the “Must Have” features and a large chunk of the “Should Have”s and “Could Have”s. We then moved into the final week for User Acceptance Testing and handover to the development partner. It was then that in my opinion our use of the Agile process showed some weaknesses.


This however went very well. Every single AU completed all the tests, signing off every single point of functionality. As the process was agile, we were able to fix every minor bug which the testing highlighted before the final UAT, and also implement some minor “Could Have”s (such as clear buttons on search forms, default selections on drop downs, alphabetization of lists, etc.)

This was possible because we, as developers, had by this time formed a solid relationship with all our users. By seeing them really using the system in anger (we told them there were prizes for breaking it) we gained a further depth of understanding of the real usability of our system. This is something I personally have rarely experienced and it was enlightening. The project team agreed that if we had known the power of this earlier, we would have got the users to have one to one hands on sessions more frequently, earlier in the process. The benefits of this to the final product are hard to pin down quantitatively, but self evident in the many testimonials we received on the completed UAT forms.

No comments: