Message modeling and XSD generation

As an integration consultant I work almost daily with XML messages. In my opinion in order to work efficiently with XML you need to have XML schemas. XML schemas makes it possible to validate your messages (including those hard to find typo’s in mappings), they can be used to generate documentation, they define your service contracts and can be used to generate a skeleton of your code. if and when validation should be enabled is a different discussion. Perhaps in the future I will write another article about it.

In order to benefit from XML schemas they need to be clear, precise, flexible and interoperable with the different technologies you are going to use on your project. Amongst us colleagues we regularly have lively discussions on how to achieve this. We all have the same ideas on the general guidelines but sometime we disagree about some details. Mostly it boils down to the choice of technology we are used of working with. But I am relatively sure I can work with the schemas created by my peers.

One major downside of XML schemas is that it is very technical and functional analyst don’t always understand it very well and why should they? They want to model the messages in their favorite modelling tool. In the perfect world you can generate the XSD’s from the model. This way you can enforce the policy you have defined to which the XSD’s should conform.

So what is wrong with this? Nothing! I even encourage you to do it. If it is done correctly and you keep in mind that in the end a developer has to import the XSD’s in his tool and work with them.

On a recent project I had to import an XSD from a third party in order to interact with them. In their documentation they were very proud of their UML model and how clever they were modelling there data. With the generated XSD they were less concerned. From what the XSD should be: simple, flexible, easy to understand … nothing was achieved. I spend 2 days trying to import them in my tool (IBM Integration Toolkit). In the end I gave up as I could no longer justify the time spent to my client.  I wrote my own (very simple) XSD’s that conform to the messages we need to send and receive and used those within our tools.

For those thinking: then don’t use IBM Integration toolkit. I have quite some experience with IBM tooling and in my career I never before had so much problems importing XSD’s. I find the XML support of IBM tools excellent. We tried to import the XSD’s in different tools and they all failed.

So to conclude I want to give you some advice:

• Pay attention to your XML Schemas
• Define guidelines/rules to which you XML schemas should adhere within your organization
• For a public interface make sure the XML schema does not use too advanced schema features (UN/CEFACT Naming and Design Rules may help you there)
• Model your data and generate the XML Schemas from the (UML) model but let your developers validate the generated XSD’s

XML Schemas should be an aid not a burden ! Keep it Simple !

Author: Jef Jansen