In the worst cases, where 'application' involves hardware, repeatable testing could even be impossible. TDD will be hard because of the intrusion of the application. You will have do an awful lot of work to produce any form of automated test. It is even worse in the embedded scenario: you could be forced to have real equipment, or an adequate simulation, just to run the code. You might even have to resort to primary evidence: good old
ACCU RIP ZIP CODE
one where the output code is mixed with the state transition code, typically you would have to run the application, stimulate it somehow, and look for secondary evidence that the state machine is working. To test a 'traditional' state machine, i.e. And this in turn makes the testing complicated. Whatever the context, the state machine will be exposed to the application, and will therefore have dependencies on that application code. an embedded system), it might involve writing to hardware, turning on motors, lighting lights etc. While this code might be trivial, its impact is often not so. implement the 'action' part of the transition. Having got that far it is very easy to simply add a line or two of code to finish the job, i.e. The effort has already been put into decoding the state and handling the new input event. It makes a lot of sense, and it is the path of least resistance. Whichever approach is taken, most people will naturally include code, in the state machine, that is concerned with implementing the outputs of the states. If you take this approach, surely the result will be perfect. Now you are in the familiar territory of 'proper' OO and patterns. Pattern-based design with a class per state and maybe a state factory. You might decide to go for a fully fledged GoF But you know you can do better than that. It meets the requirements and it will probably work. You might go so far as to have a function per state, with anotherįor each input event. Matthew Jones seperates their concerns to make testing easier.Īnyone can code up a state machine, but can you make such a machine fully testable? Can you prove that it is? Can you do this repeatably? In other words, can you present an inquisitor with a test suite that proves your state machine fully implements a given state transition diagram?ĭepending on your background, you might choose to write your state machine using the C State machines are a common design pattern.
0 kommentar(er)
