Monday, 1 April 2013

Why are you not using Design by Contract?

When learning to program I distinctly remember coming across the concept of placing asserts within your code. Assert statements are primarily used for "things that cannot happen", but in my early days I was too focused on the stuff that was supposed to happen!

"Defensive programming" was also introduced. Principles such as "Never trust the user" and "80% of your code will be validation and verification" were highlighted. Despite these introductions many years ago, the concept of asserts never stuck with me. Yet I program defensively like there is no tomorrow.

The use of asserts can be extended into "Design by Contract" or DBC. In DBC the developer makes use of pre-conditions, post-conditions and invariants. Some languages such as Effiel have taken DBC as a core feature while other languages leave DBC up to libraries.

One of my favourite programming books is the Pragmatic Programmer. Having stood up to many re-reads I always found myself intrigued by the idea of DBC. Yet I never found myself following this interest through, at least in a production environment.

Our team recently came across a bug in which part of the system was using a component in a way which was deemed invalid. We had a suite of tests to accompany this feature, but these tests were unable to highlight the problem. When the object was sent across the wire, the Javascript front end was firing a null reference across, this was out of our control in the back end of the application. As the feature crossed a boundary and spoke to another system defensive programming would have been difficult. All we could do was error and inform the developer what was wrong. Even without defensive programming, the system was currently doing this anyway. We had little to gain.

Here I decided to experiment for the first time in my programming career with code contracts. A contract was applied that said the collection sent into the system must not be null or empty. If so, the second system would blow up informing the developer what was wrong. This contract was a very primitive example of a pre-condition - something that must be true in order for the rest of the following code to execute.

The benefit here came from just a few mere lines of code. Had we tried to program defensively the second systems' code base would have suffered for little gain. We would need to report the error, add error codes, introduce exception handling and so on, all for a simple defect that could be fixed immediately and potentially never occur again once the developer integrating has configured the components correctly.

One important factor to consider with DBC is the contract violations should never be caught or handled. Every single contract that is violated is a bug. To stop the violation you need to fix the code that is breaking the contract. Likewise contracts make little sense when dealing with a public API. On the edge of the system you should presume your users will make mistakes and "do the wrong thing", here you must use defensive programming.

Since this day I've liberally applied code contracts whenever we cross system boundaries or interact with the infrastructural aspects of our code, e.g. database helpers. This has increased my confidence that the system as a whole has been correctly "glued together". Another benefit is several bugs have been thwarted thanks to the contracts as unlike unit tests, contracts are always present when enabled, meaning missed boundary conditions can easily be detected.

Hand in hand with our automated test suite, code contracts make a great companion. Never alone will one suffice, but when used in conjunction they can be extremely powerful. So the question is, why aren't you using them?

3 Years at Codeweavers

Having wrote about the top ten things I discovered in my first year at Codeweavers, I figured it would be time for a follow up after the past two years. In no particular order, a collection of the biggest lessons I have experienced.

I'll expand on these topics over time in future posts.