Published on 12 June 2013.
Today’s thought is about how you can refactor code with higher confidence.
Imagine how you would change your attitude towards refactoring code if you knew that the refactored version worked exactly like the original for a large (automatically generated) set of inputs.
It is a common technique to cover a piece of code with tests before you change it. That way you can change it and have a higher confidence that it still works afterwards. But what tests should you write?
One thing you can do is to call the function you want to refactor and see what the output is. Then record that output as the expected result in a test. If you do this for a few inputs, you get some coverage. The tests might not serve as great documentation, but they will serve as a safety net when refactoring.
You end up with a set of tests like this where xn are the different inputs and yn are the different outputs:
f(x1) = y1 f(x2) = y2 f(x3) = y3
Then when you refactor the function f, the tests should still pass.
What if you could automate this process? What if you could somehow generate the different inputs automatically, would it then be possible to test the refactored function automatically? Yes, it would. The following assertion should hold where f’ is the refactored version of f:
assert f(x) = f'(x)
This says that for any input to the function f, it should produce the exact same output as the refactored version f’.
This assertion depends on the inputs being somehow automatically generated and on the two functions existing at the same time.
Tools can probably be developed to help write tests of this kind so that automatically generating inputs becomes easies and so that duplicating the functions does not become that tedious.
Site proudly generated by Hakyll.