It seems a simple enough question, but I kept pondering it and delving deeper until I realized I needed to write this monograph.
What unit tests should we write? How do we know what to test?
Ideally, unit tests should cover every path through the code. It should be your chance to see every path through your code works as expected and as needed. If you are practicing Test Driven Development then it's implied everything gets a test.
In the real world, you might not be allowed to test everything - for instance, if the testing suite ends up taking a week to run, then the world will have changed by the time it finishes and the test results will be obsolete.
Unit testing at it's basic is testing an object, a method - the smallest unit of your code that it can test independently.
It should test the inputs "goes into" and corresponding outputs "goes out of". It should test every variant of input that will exercise every code path and even try to get to exit it's code paths entirely - throw an exception, SIGINT, stack overflow, whatever. It's no time to be nice to your code. Test boundaries - test what happens when you are outside the boundaries - inputs outside what you expect, exceptions from called objects, unexpected types - strings instead of numbers, floats instead of ints.
It should be automated and fast so that whenever that "thing" (your small unit of software) changes, you can run the test and make sure it still works as expected - as it did prior to the change.
A good and complete suite of unit tests should:
·
Free
you to do exploratory testing at a higher level, manually or however
·
Give
you confidence to change, refactor fearlessly - if all the same gazinta's
produce the same gazouta's for a unit, you're good!
Test code should be treated as well as all the rest of your application code - it's not hack code, junk code, throwaway code because if it is, it doesn't belong in the source tree.
Designing for testability:
- Adhering to separation of concerns help make code
testable - focused functions that do only one thing are easier to test
- When accessing other objects, try to segregate or wrap
the calls so that mocking them becomes easier
- Segregate validation code so that it can be easily
tested
- Think of patterns that foster testability like the
Facade pattern
In our Javascript ecosystem, here are some further specific hints:
- Look at anonymous functions - very often these can be
turned to named functions and makes them testable units. If you have a 10+
line anonymous function, it's probably doing something non-trivial - so it
could probably use a test.
- As you are writing functions, look for ways to make
them composable (See https://codewords.recurse.com/issues/four/lazy-composable-and-modular-javascript for
more details on concept and techniques).
- Know what your functions return - or in other words, be
explicit about what your functions return. Don't surprise the caller.
- Try to create functions without side effects -
otherwise try to make the side effects singular in purpose and scope for
each function. Test the side effects.
- Test Failure - make sure a .catch on a .then gets
tested.
- Perhaps the use of flogs, such as '__DEV__' will help you segregate test code
you want to disappear in production.
References:
xUnit Test Patterns - Website of book by same name. Published by Addison-Wesley - well worth purchasing. Accessed Jan 19, 2018 - an excellent resource on xUnit test patterns - goes over mocks, test doubles, lots of helpful patterns.
Robert Martin of First Class Tests - Accessed Jan 19, 2018 - This article in particular makes a very good case on why you should treat your test code as a 'First Class Citizen' - an equal to the application code it supports. It's a mind shift - an important one.
Collected Wisdom of Martin Fowler - Accessed Jan 19, 2018 - lot's viewpoints on how to get value out of testing - not a one time read, more of an on-going resource.
Comments