Verifying test independence
Tests should not depend on each other to pass. In general, each test should perform its' operation in complete isolation. This is clear from standard testing framework such as JUnit, unittest, and pytest. Notice the concept of cleaning up is introduced very early in the documentation.
Each test is encouraged to handle setting up its' environment as well as cleaning up. This is good practice and can prevent sudden test failures when introducing a new test.
Have you ever run into a situation where adding a new test suddenly causes another, seemingly unrelated test to fail?
This is a difficult issue to debug. It's most likely a side-effect of the test execution order. One test somewhere in your test suite is not setting up or cleaning up properly.
So, writing independent tests is very important, but how can you be sure you've achieved this goal?
One trick I use is to randomize the order my tests are run every now and then. I use pytest as my testing framework, and there's a nice plugin to randomize test execution. I install this plugin and use the --random option to mix things up a bit.
Mix your tests up every once in a while and see if something fails. There's a good chance any failures you see (assuming your tests pass beforehand) are due to subtle dependency issues.
There's one downside. Running with the --random option once doesn't guarantee you are free of test dependencies. Remember, the order is random. The random order can hide the dependencies just as your normal test order can.
There's not an easy way to tell how many times you need to run with
to verify this. Sure, you could do the math, track all the possible
permutations and verify you try them all. You could even write code to do
However, periodically running tests in a random order provides a higher degree of confidence in your test independence, and it's easy with pytest and the pytest-random plugin.
Published: 10-29-2014 08:46:12