This failure means that the code must be developed further for the tests to pass, so the green phase continues. Online research suggests the following code which uses a lookup table and executes much faster (Figure 9). The lookup table has 33 values, and interpolation is used to compute the value for each angle.

Figure 9: Fast sine/cosine calculation for embedded application in C. Source: AtWillys.de

The code builds and when the test cases are run on this code, they all pass.

This is clearly a step forward but the code is very difficult to understand and would therefore be difficult to maintain, contravening the objectives of functional safety and cybersecurity standards such as those referenced earlier and severely offsetting the advantage of this concise, efficient, “clever” code.

The role of coding standards

C is an imperfect language. Unlike other languages such as Java, it does not trap errors as soon as they are found and code can execute unpredictably instead. It also exhibits fundamental language-related causes of failure in the form of undefined behaviour, implementation defined behaviour and unspecified behaviour. For the unwary, such behaviours can lead to software bugs and to security issues. Other features of the language are simply difficult to understand, and hence the cause of unnecessary mistakes.

Coding standards such as MISRA C ensure that a subset of the language is applied to avoid such pitfalls. Coding standards are required by most functional safety and cybersecurity standards for exactly that reason and are just as applicable whether TDD is used, or not.

Performing a MISRA C check on the “clever” sample code from the example reinforces that position. Trying to refactor the code to make it compliant with MISRA C is just not feasible and the only solution is to redesign the function.

Using a lookup table is a good idea, but it is better to avoid the complex interpolation by reducing the angle to a value in a 90° quadrant and then look up the value from a table that contains 90 values, one for each degree. This consumes a little more memory than the previous example, but the resulting code is crystal clear, easy to understand, maintain and test – characteristics that should help make it MISRA C compliant (Figure 10).

Figure 10: Fast sine/cosine calculation that is also easily understood.

When the tests are regressed they all pass, meaning that the actual values are equal to the expected values ±0.1% and also that in every example, the tests execute within the allocated time budget.

Third iteration: Addressing the timing issue (green phase)