The Amara project uses the PyTest testing framework.
To run all unittests:
$ dev test
To stop after the first failure:
$ dev test -x
To run tests from a single module
$ dev test module.name
To run tests matching a particular pattern (matches against filenames, class name, method/function name, pytest marks, etc).
$ dev test -k [pattern]
Drop into the python debugger on errors (Very useful for debugging hard to understand failures)
$ dev test --pdb
We also have gui tests that use selenium to automate testing with a real
browser. These get run with the
dev guitest command. It accepts all the
same arguments as
dev test. GUI tests are located in the
GUITests have a special setup:
- Before the tests run, we startup a server to test against.
- We run the
setup_guitestsmanagement commands to clear out the database for the test server, and auto-populate it with some data.
- The tests themselves never access the DB.
- When writing GUI tests:
- If you need to populate the DB with some data, do it in the
setup_guitestscommand. If you are running tests from
amara-enterpriseor another submodule, you can hook into the signal defined in that command to populate the DB.
- Note which TestLodge case the test is for. Try to keep the TestLodge case up-to-date with the automated test.
- If you need to populate the DB with some data, do it in the
There’s a couple tools to help debug failing tests:
- To see the webserver output run
- To see browser screenshots for failing tests, use the
--take-screenshotsflag, then check in the
guitests/screenshotsdirectory after running the tests.
- To start up the test webserver run
dev up --guitests. This is useful if you want to manually step through a failing test.
- You can also use
dev manage --guitests`to run management commands in the GUI test environment.
dev jstest to run them.
Writing unit tests¶
Unittests test the behavior of a small unit of code – a class, small module, etc. We write unit tests with the following goals:
- To define correct behavior.
- To assist initial implemenation. It’s quicker to run unit tests to verify that you’re code works than to manually test with a browser.
- To increase confidence when refactoring. Since the tests should quickly find things that broke.
When writing unit tests, try to follow these guidelines:
- Write tests first before code. See the first two points above for why this is useful.
- Have each test check 1 piece of behavior only. Don’t add assertions that are checked by other tests. On the other hand, don’t feel like you need to follow rigid rules like 1 assertion per test.
- Mock out dependencies. This makes it so you’re only testing the code-unit at hand, not other pieces of code.
- Mock out external services. Don’t require the unittests to make network requests to things like the google API. This slows down the tests and creates extra ways for them to fail.
- Document your tests. Make it clear what each behavior test is checking. There’s nothing worse than finding a failing unit test, but having no idea what the test was trying to do.
- Target a small unit of code. Usually tests target a single class or a small module in isolation. This is good because if the test fails it’s relatively clear which piece of code caused it. If you test multiple units of code together and a test fails, then you need to check any of them.
- Put tests in the ``tests/`` directory under the app you’re testing
Writing GUI tests¶
GUI tests check high-level behavior using Selenium to automate a browser. We write GUI tests with the following goals:
- Testing regressions. Once we know a potential way for Amara to break, we want to continuously test that it won’t happen. GUI tests are great for this.
- Testing integrations with amara code. Unittests check that each individual code piece is working correctly, GUI tests check that interactions between them. For example we might create a unit test for the Video model and each of the VideoType subclasses, then we would create a GUI test for adding a video, which would check if those components were working together correctly.
- Testing interactions with extenal services. Unittests should mock out access to external APIs. GUI tests don’t mock them out which allows us catch when our code breaks because an external API change.
- Testing integrations with dependencies. When we upgrade a dependency like django, we want to makes sure the pages still function.
- Testing workflows. Unittests target low-level actions. GUI tests target user-level actions that consist of many low-level actions together. Actions like logging in, adding videos, creating subtitles, consist of multiple low-level actions strung together.
When writing GUI tests, try to follow these guidelines:
- Assume the low-level logic is correct. Unittests are a much better way to test this, so don’t try to test the business logic from GUI tests. For example, we have lots of tests for the subtitle action system to make sure that works. This means it would be good to create a GUI test that tries to click the buttons corresponding to the actions from the editor (“Update”, “Save draft”, “Endorse”, etc.). However, it would be a waste of time to try to write GUI tests to re-check the low-level logic by clicking all possible buttons from all possible states.
- Target a single user action. Something like logging in, adding a video, submitting a form, etc.
- Put the tests in the toplevel ```guitests/`` directory. Since GUI tests test several components together, we put them in a toplevel directory rather than inside an individual app.
What is do testers do, given that we’re trying to write all these automated tests? Lots of things:
- Exploratory testing. Automated testing can only check for known bugs, testers are good at finding new bugs by interacting with Amara in unexpected ways.
- Testing user experience. Automated testing can only check if a process works or not, testers can check if a process is intuitive/pleasant/simple for a user.
- Defining tests. Testers write up the regression tests in English, to provide a basis for writing the automated GUI tests. This is not solely the testers resposibility though, developers can and should also write up tests.
- Verifying new functionality. When we create new functionality, it’s the tester’s resposibility to decide when it feels good enough to merge.
- Regression testing :( . Unfortunately, we have enough GUI tests to cover all our regression testing, so testers need to do it manually. This should be fixed as soon as possible.