18/05/2007

 The good, the bad and the ugly

The bad old days

Long ago I strongly believed in deterministic approaches for software engineering. At school I learned the classical way to build software. It’s based on the experience of how we build stuff in the concrete world. In software development literatur you find the analogy between building a house and building a piece of software. First you need an architect who will draw the plan of the house. Every aspect of what will be in the house has to be described in the plan. When the plan is ready the entrepreneur can start with building the house. The architect will inspect regularly the house making sure that everything is like he described on his plan.


For a large majority of software professionals this is the model inspiring the way they build software.  Big upfront design has to take place and the result of it is a detailed analyse. Only when the analyse is complete we can start developing. We are supposed to follow what is specified and at the end we should get the application that our customer desire. I remember the time when I learned engineering software this way. It was at the university, I already was graduated but I decided to further specialise in what was called IT management. On that time I already worked for the IT industry. So because I couldn’t be present on the daily courses I had made a deal with my professors to see them once a week to review what I had studied during the week. This provided me the opportunity to better interact with my teachers and I remember a talk with my professor for software engineering.

During that week I had reviewed the famous waterfall model of software engineering. This model is completely based on the classical approach: you start with an analysis, you design the architecture, you build the software after you test every piece and finally you integrate the whole. I found the concept appealing but I couldn’t recognize this way of working in the organization I was working for at that time. So I asked my professor if he had really seen this type of model working in the real life. This was the only time when I saw the man, for which I had and still have a huge respect, showing signs of contempt. He overreacted, basically he answered that in his organization he conducted projects based on this model but that lots of organization didn’t because they where unprofessional.

I remember that in the beginning of my career I constantly was frustrated because I never encountered projects where this recipe actually worked. Especially when working for smaller organizations the customer was not willing to play this type of game. I thought that the reason for this was that they didn’t understand what software engineering is and how complex it is. Now I realize that we are wrong. We are all actors playing the software engineering game, fooling ourselves and the customer.

Usually we begin by interviewing the customer and try to understand what his problem is and how we could help him solve that specific problem. Because we are self confident and because the customer has a business to run this type of interview remains mostly short. We go back to our offices and writes documents describing what should be build. Writing this type of document is fastidious and time and resource consuming. When, months later we present this document to our customer where we describe in details what we intend to build he’s not willing to spend days reading it. In fact I never had a customer that actually reed and understand what was described in the analyses, just because they had a business to run and couldn’t afford spending hours reading this document as big a phonebook full of incomprehensive technical terms. It’s the reason why we make powerpoint presentations summarizing what is described in the analyses. What happen usually is that the developers working in the project don’t read the analyses too. Partially for the same reason as the customer, they have a work to do. But the main reason why developers don’t read the document is simply because they know that it isn’t valid anymore. Between the time the analysis was written and that the developers start coding, things have changed.

The ugly rules of software engineering

This is the first rule I discovered: customers change their minds all the time, especially when you show them the final product! But why do customers change their minds? Simply because when they are confronted with the actual product they suddenly dispose of more information on the product as when it was only a mental representation. Changes occur also because you realize that some aspects of the application should be improved. When you are building the product piece by piece you are confronted to some aspects of the reality you hadn’t anticipated before. Also you are getting more and more info from the customer and your understanding of the problem domain constantly evolves. I remember projects we begun to truly understand what we were building when we had already build a consequent part of the application.
The point is that we already are getting late on the project because all this analyses has taken a lot of time and the time we used to make the analyses will not help us as it should be. When the project finally arrives in the development team the expert developers try to figure out what has to be build. The system is divided into the technical core concepts, the user interface, and the backend. Code skeletons are generated and assigned to the developers for implementation. It soon becomes apparent that the modelled solution cannot be implemented without problems. The result is that what is implemented isn’t what is described in the analyses and the analyses document becomes completely obsolete. At the end the project is not completed on time and after months of pressure we produce an inconsistent system that contains many errors. Subsequently the development is chaotic. Users find errors in the application and establish that many of their requirements were not implemented adequately. The developers are informed of this and they try to implement the changes as quickly as possible. In this process, the internal structure of the application degenerates and the cost of future changes increase further. In the end the maintenance of the software becomes a nightmare and the cost of maintenance rises exponentially till we decide to re-build the software from scratch.

Another well know problem we constantly encountered is that analysts tend to increase the complexity of the system in order to foresee every possible problem. Recently I participated on a project where the project started without a real customer and without clear business objectives. The analysis was full of technical requirement describing how the portal should be build but their was nearly no reference on what the portal should actually contain and for what it should be used. The result was that the analyst and developers increased the complexity of the modelling in order to foresee every possible problem. This leads to unused technologies, which simply means a waste of money. So the second rule in software engineering is: deciding what to build is more important than how.

The classical process of software engineering leads also to separation of people. Customer, analyst, developers works not with each other but separated by place and time. In general, it is bad when analysis, design and construction are separated. This separation enhance the gaps between the vision of what should be build and how. It is not unusual that the people who has initiated the project and so has dictated the vision are not available anymore when the actual implementation occurs. This leads to misunderstandings on the actual goals that should be fulfilled. This is the third rule of software engineering: Most of problems that arise in software projects can be tracked back to lack of communication.

Adopting good practices
People who are interested in software engineering know that this kind of problems has been detected long ago and people have begun to think about other methods for making software. Other development methods have come to my attention over the last few years. It has come to my attention when I was realizing that building a house is not the same as building software. The classical rational approaches are very good in dealing with complex but immutable things like constructions or mathematics. But when dealing with rapidly changing environments like software engineering these methods tend to be ineffective. When working for things that tend to change constantly you need more agility. This is why Agile methodologies have arisen and nearly are becoming the norm now. Agile methodologies strive for a slimmer, more lightweight development process. In my opinion when it comes to time and resource consuming factors like processes, the principle that should be haunted is that every process should have a clear justification for everything that it incorporates. Even agile methodologies are sometimes going against this rule but they are nevertheless a breath of fresh air. They help us at better managing change. They definitely enforce a good communication between everybody on the team and so help us to build what the customer wants.

Agile methodologies come with many variants, one of them is Xp. Xp is certainly one the most used agile development process used these days. The father of Xp is Kent Beck, he published the first book on Xp but Xp is not the invention of one man. A number of very respected software engineers like Ron Jeffries, Martin Fowler and Robert C. Martin started to use Xp and a community has grown up around it.
Advantages of agile methodologies
Agile methodologies encourage rapid feedback from the customer and simple practices that leads to a better quality and rapid discovery of defects. Some of these practices are:
- Pair programming: developers always work in pair so that each developer can check the work of the other and help to discover errors soon.
- Continous integration: In the classical approach every module is build by a developer and at the end all modules are assembled, this usually lead to lots of integration problems because the pieces don’t fit to each others. In an agile project the modules making the entire solution are constantly integrated so that integration problems are detected immediately.
- Test driven programming: programmer constantly writes test that constantly test the software

Many studies have proven that agile practices lead to the following advantages:

- Cost effective: One of the major advantages of agile methodologies over traditional methodologies is that agile methodologies are less costly. The practices that the agile methodology promotes have all a common goal: reducing the cost of change. In place of trying to foresee everything in advance we use our energy to make things easier to change.









source: http://www.ambysoft.com/essays/whyAgileWorksFeedback.html


Higher customer satisfactionAgile methodologies favour an incremental approach. The increment is determined by the customer. The customer prioritises the requirements and the developers are responsible for estimating the cost.
- Rapid time to marketIn place of spending months on documents that will never be used the stakeholders simply describes their requirements then the developer spend several hours or days implementing the feature that can be rapidly show to the stakeholders.
- Lower financial riskRapid time to market leads to a shorter payback period.


I’m convinced that nobody should ever follow blindly any process. Each team and each project is unique and therefore needs its own custom-made process. A process must be adapted to the locale and be continuously improved. It requires constant reflexion to constantly improve and adapt a local process. That’s why it’s so important to be aware of your environment and what the process used elsewhere are. Post-mortems are meetings where everybody of the team can speak about what went wrong during the project. These meetings can help us continuously improve the development process and to avoid the same mistakes twice. Finally I should stress the fact that at the end it’s the people that make the difference and not the process. This is why every good process should tend to make it self irrelevant.

16:20 Écrit par Geoffrey Vandiest dans Web | Lien permanent | Commentaires (0) | Tags : agile, waterfall, software engineering, architecture, xp |  Facebook |

13/02/2007

End To End testing

I truly believe in TDD (Test driven development).  Nevertheless I think that just TDD and refactoring don’t imply that the developers are writing first time right code.  TDD will only test the case the developer has thought of.  It’s a coding practice that is helping in improving the design but it’s not a guarantee for perfect quality.   

On the project we are currently working on, we’ve planned some end to end testing sessions. Each testing session’s goal is to test a couple of stories. We hope that these tests will be helpful to better understand what our end-users really want.   As we don’t have a dedicated QA team the responsibility of the tests is handled by our usability team.  They invited some end users for testing sessions. We already do a lot of tests but this is the first time we are organizing formal end to end tests during the development process. 

Although the first testing session will be held tomorrow I’ve the feeling that the investment we made by preparing these tests is already paying off. Usually a couple of weeks before the delivery date a lot of pressure are put on the head of the team.    The sessions are motivating the team to deliver a complete set of features at regular moments in time.  Because we organised these tests all over the development period we’ve divided the delivery of the product into small releases. The developers are therefore making regular sprints and this is improving our productivity and quality. 

Knowing that the fearless end-users are coming to evaluate our product stimulates us to deliver more quality.  It helped us discover bugs sooner because someone else that is not as close to the code check that the code is doing what he expects.

22:15 Écrit par Geoffrey Vandiest dans Web | Lien permanent | Commentaires (0) | Tags : testing, xp, tdd, usability, end to end |  Facebook |