Spa2010: TDD at the System Scale

Spa2010 is a conference for of people from all over the software community, run (and named after) the BCS special interest group for Software Professional Advancement. The conference is known for its rich mix of topics and heavy bias towards hands-on sessions, and annually takes on 2 forms-  the main 3 day event in London and MiniSpa, a free one day “Best Of Spa” compilation where the highest rated sessions are run again. This year I was lucky enough to attend the full conference thanks to some training budget I found, so I headed off to meet some hear some very interesting people run sessions on everything from testing to behaviours in civil architecture which could be applied in software engineering.

The first talk I attended was by Nat Pryce and Steve Freeman, co-authors of “Growing Object-Oriented Software, Guided by Tests”, who ran an interesting session on how to write and test (or should that be “test and write”- test first!) large scale systems. The session started with a great overview of the important lessons learnt and went on to some practical exercises based on the attendees hunting down testing issues. Here is a bit of what I learnt from it all:

Testing from the boundaries inwards, not centre-out

The biggest point made for me personally was that it is too easy to start by writing tests and implementations for our domain classes and settle for extensive unit test coverage without giving much thought to testing at larger scales.

Using only TDD at the unit scale means that we end up defining the “idealised” version of each unit- our domain classes define what we wish our domains looked like, our processing methods reflect how these perfect domains would be manipulated and so on until we hit the outer boundary of our program… where we end up with a rift between the ideal model we’ve built and the reality we need to integrate with.

The solution is simply to not start with our low level unit tests first, but to broadly scope and write tests at the limits of our system. This ensures we have some appreciation of the boundaries of our software and means we start working with the API’s we need to integrate to at the start. Any restrictions and needs placed on us by these API’s are considered from the beginning of development instead of as the last step when the pressure to deliver is on.

Use Anti-Corruption Layers or Simplicators

Another important point made was that we can lessen the impact of imperfect API’s by abstracting them behind a “simplicator” or “anti-corruption” layer- basically a small, separate project which consumes the ugly/unwieldy API’s we need to integrate with and exposes a more ideal API for our system to consume.

This also has the effect of largely de-coupling the implementation of our system from the specific system we are building against. An advantage I saw in this is that bad design can be isolated and removed when projects are improved or replaced; the issues in project A need not affect project B when it integrates with A; this also possibly avoids the issues being forced into the design for the replacement for A when it needs to ensure compatibility with B.

Testing Asynchronous Systems

The session went on to some excellent examples of tests which flickered (passed or failed somewhat randomly) and explored issues of false-positives and false-negatives caused in an asynchronous system. The lessons learnt from this, in summary were:

If some state changes in your asynchronous system, make damn sure that your tests are synchronised in some way:polling the state just doesn’t cut it. Sometimes the test will miss the change by testing the state before it has happened. Adding a wait or thread sleep might work, but it makes your tests take longer to run than they should. Polling after the state has changed, then been changed back again looks like a failure despite the correct behaviour being played out.

The session was crammed with useful tips and examples, I strongly recommend going to it if you get a chance at another event; it certainly makes a lot of testing and design issues known which I am certain I’d have wondered into over the next couple of years. Many thanks to Nat and Steve for all the time and effort that went into it!


2 thoughts on “Spa2010: TDD at the System Scale

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s