Deepcov

See test results for every line of Python code

Deepcov is a pytest plugin + VSCode extension which combines test status and code coverage data.

Making Sense of Test Runs

Tests can often fail due to unreliable infrastructure, such as network dependencies. When this happens, it can be difficult figuring out if we should let it distract us from what we're working on.

What if we just wanted to pay attention to tests running through components we have changed?

Pytest can tell us if tests pass, but can't tell us if we changed their behaviour, whereas coverage data can tell us if tests are running through a component, but not if they are passing.

Tutorial

We are going to debug a toy test suite using deepcov.

Before starting, ensure you have python and VSCode installed.

Run tests with deepcov attached

1. Fork and clone this repo
2. code pytest-deepcov/python-cli/tests/resources
3. Open src/lib.py and src/test_lib.py
4. Create yourself a new terminal window in VSCode
5. (optional) virtualenv .venv; . .venv/bin/activate
6. pip install pytest-deepcov
7. pytest --cov=src
8. the tests should fail, and a new directory .deepcov should be created

Visualise deepcov data

1. Install the VSCode extension. It will add a 'deepcov' button to the top right of your editor. Don't click it yet :)
2. Configure your CLI location: Open VSCode settings, then search for deepcov. Follow the instruction to connect with the python package.
3. Enable deepcov using the 'deepcov' button to view your pytest test results from above.
4. src/lib.py and src/test_lib.py should have annotations on every line with test results.
5. Disable deepcov using the same button

Troubleshooting

1. If it complains that it can't find the CLI, you will need to (a) ensure the CLI is correct, (b) close and reopen VSCode to refresh the config
2. Are you on Windows? If so, please try windows subsystem for linux.
3. Please log an issue on this repo and I will help
4. If you are feeling brave, try using the CLI to debug (see below)

Deepcov Internals

The VSCode extension fetches data for each line of code from the deepcov python package.

deepcov /Users/a/git/treebeardtech/deepcov/python-cli/tests/resources/src/lib.py

{
  "lines": {
    "3": {
      "passed": [],
      "failed": []
    },
    "1": {
      "passed": [
        "ran on startup"
      ],
      "failed": []
    },
...
    "5": {
      "passed": [
        "src.test_lib.TestLib.test_divide2[3.0]",
        "src.test_lib.test_divide",
        "src.test_lib.test_divide2[3.0]"
      ],
      "failed": [
        "src.test_lib.TestLib.test_divide2[0]",
        "src.test_lib.test_divide2[0]"
      ]
    }
  }
}

(Note: this cli must be run inside the .deepcov output directory)

You can use this approach to create plugins for other text editors.

Limitations

• Project structure: Your pytest root directory must be the same as your vscode root directory.
• coveragepy is used to capture coverage data using the Python trace hook. This means:
  1. You cannot use the debugger and deepcov at the same time. Deepcov detects when the debugger is active and disables itself.
  2. Everything takes 30% longer. It is best to only use deepcov to rerun subsets of a failing test suite.

Diving Deeper

Having test results for every line of code opens up the possibility of implementing spectrum-based fault localisation

• https://arxiv.org/abs/1607.04347 contains a description of this technique and expected results
• https://github.com/saeg/jaguar implements the described heuristics in java