Rethinking our test strategy

[vringar]

To understand where we want to go we need to accurately describe the current situation of our tests. What we have:

• A lot of integration tests that enable different instruments and then assert that the database contains the elements that we expect to be there
• A web server that gets automatically started when the tests begin running
• The extension being built automatically every time the tests get run

What are the shortcomings of the current approach:

• The rebuild of the WebExtension on a fast computer takes at least 40s meaning that rerunning a fixed tests takes 40s + actual runtime of the test => This makes TDD impossible since the delay is too big
• We mostly/only test the happy path. Failures in our error handling are only discovered when tests are accidentally broken and the failure triggers a chain reaction
• Since we spin up the complete platform running a single test more than 10s

My suggestions to fix this:

• Test the different instruments by only creating an MPLogger and an in-memory StorageProvider, and using Selenium directly to control the browser, leaving the entire TaskManager/BrowserManager machinery out of this, similar to what was proposed in #330
• Test the TaskManager by monkey patching out the normal BrowserManager and replacing it with a TestBrowserManager, that different exceptions at different points in time to simulate the browser failing in different ways
• Test the StorageProviders and StorageController by just creating them stand-alone and then feeding them different messages through the socket. This allows us to test a bunch of edge cases without having to mess around with the platform
• Remove all autouse fixtures and instead explicitly inject them where they are needed.
• Maybe don't rebuild the extension as part of the test suite (I see the use, but the drawback is also significant)