How thorough are your tests? Do they test all parts of the code?

Code Coverage is a measure of how much of your source code is tested or “exercised” by unit tests. It measures how many functions, methods, statements, and branches are tested.

There are “coverage” tools for almost any programming language that will gather statistics on what lines of program were executed, which branches were taken, and which functions called.

Python Code Coverage

The coverage package performs code coverage analysis for Python. It analyzes how much of your code is executed when a program is run.

Coverage works with unit tests on any Python app, including Django.

Install Coverage

1. Install coverage package. The instructions are different for each OS, so read the instructions. In many cases you just type

    pip install coverage 
   

2. Check the installed version of coverage and read the list of command line uses:

    coverage --version
    coverage help
   

   The commands you’ll probably use most are:

    coverage run some_program.py    # run a program with code coverage reporting
    coverage report                 # view coverage report as text
    coverage html                   # create an html report
    coverage erase                  # delete data and report from previous run
   

3. Use coverage run instead of python3 to run a program and gather data on what statements in the program are executed.
   Do this for unit tests to see how much of the target code (code being tested) is actually executed using your tests. Enter:

   coverage run -m unittest auction_test.py
   

   or use auto-discovery to run all files named something_test.py:

   coverage run -m unittest discover -p "*_test.py"
   

4. View the coverage report as plain text:

    coverage report
   

5. Create HTML pages containing detailed information with links:

   coverage html
   

   The output is in subdirectory htmlcov.
   Open the file htmlcov/index.html in a web browser. Click on any file name to see what statements were “covered” (or not) during the run.

Branch Coverage

By default, coverage performs only statement coverage.
To request that it also verify that all branches in the code are executed add the --branch option:

   coverage run --branch  -m unittest auction_test.py
   # the report (and html) will contain a new column for branch coverage
   coverage report

How Many Unit Tests Are Enough?

Consider this example of computing the discount for a Sale:

If the sale total is at least 1,000 Baht, give customer a 50 Baht discount.
If the customer pays using cash, then give a 2% discount.

def total(sale):
    """Compute total price of a sale, including discounts."""
    total_price = sale.get_total()
    if sale.payment.type == PaymentType.CASH:
        total_price = 0.98*total_price
    if total_price >= 1000:
        total_price -= 50
    return total_price

How many test cases do we need to thoroughly test this code?

Cyclomatic Complexity is a measure of the complexity of a method, class, or package. For a method, cyclomatic complexity is the number of different paths of execution through the method.

A simple formula for computing it is:

   complexity = #branches - #decision_points + 2

The complexity score is usually a bound for the number of tests needed to cover every path. Wikipedia has more details and examples.

Configure Coverage Using .coveragerc

You can configure what files are analyzed by coverage using a .coveragerc file in your project directory, as described in the Coverages Docs.

One useful option is to request “branch coverage”, to count how completely “if - else …” branches are executed.

For code that uses a library or framework, you want to exclude the library or framework from coverage analysis, since it’s not useful and distorts the results.

For Django projects you want to exclude migrations, settings.py, manage.py, static files, and anything else you don’t write unit tests for.

In the Django Polls tutorial, I used:

[run]
# measure branch coverage
branch = True
# don't measure python standard library (this should be the default)
cover_pylib = False
# omit uninteresting stuff
omit =
    __init__.py
    /usr/*           # Linux location of Python libraries
    mysite/*         # application configuration files
    #TODO            omit migrations
    #TODO            omit unit test files and directories
    #TODO            omit __init__.py files
# explicitly include the main app
include =
    polls/*

# in the report, exclude some methods we don't need to test
[report]
exclude_lines =
    def __str__    # example
    def __repr__   # example

100% Code Coverage Does Not Mean Tests are Perfect

You can have 100% code coverage and still miss defects. Here are a couple of cases and examples.

• Both code and unit tests neglect some Use Cases.
  • Example: in Django polls tutorial, a user can manually submit a vote for a choice that doesn’t exist. Our code and tests don’t check for invalid choices.
• A method has multiple conditional branches, and not all combinations are tested.
  • See the example below.

This is an e-commerse app. Sales with value over 1,000 Baht get a 50 Baht discount, and Platinum customers always get a 3% discount. What happens if a Platinum customer has a purchase of over 1,000 Baht?

import pos.model.Sale;
import pos.model.Customer;

class Register {
    private static final PLATINUM_DISCOUNT = 0.03;

    public double computeTotalWithDiscount(Sale sale) {
        double total = sale.getTotal();
        Customer customwe = sale.getCustomer();
        if (customer.isPlatinum()) {
            total -= PLATINUM_DISCOUNT*total;
        }
        else if (total > 1000.0) {
            // everyone gets this discount
            total -= 50.0;
        }
        return total;
    }
}

1. We might have unit tests that test for “Platinum” customer discounts and the 50 Baht discount, but not both. We might miss the cases where both discounts apply.
2. We might miss an unclear requirement: what if a platinum customer buys 1,020 Baht?
   • Should he get both discounts?
   • Which discount should be applied first? (the total discount will be different!)

CodeCov for Automated Coverage Analysis

You can perform automatic code coverage with a CI server (like Travis-CI). For this, you can use an online service like CodeCov.

Codecov.io is an online code coverage service that integrates with CI pipelines, including Travis-CI and Circle-CI.

For Python projects, Codecov.io uses the Python “coverage” package to perform the actual coverage analysis. Codecov.io manages generation and display of reports, and keeps a history of coverage results.

To get started, go to codecov.io and register using your Github ID. Give Codecov access to the project(s) you want it to pull and analyze. It is similar to the way you give Travis permission to pull and build a project. Then modify your .travis.yml file to perform coverage analysis and send the data to codecov.io.

You can add or remove projects later by going into your Setting page on Github, look for the Codecov app, and add/remove projects. This is the same procedure as adding Travis-CI to a Github project.

Invoke Codecov from the Command Line

You can send your coverage reports directly to codecov.io from your own computer. To do this you need a few things.

1. Install the codecov package.

    pip install codecov
   

2. Run “coverage” to create coverage data:

   coverage run your_python_tests.py
   

3. Get your CODECOV_TOKEN from codecov.io and add it as an environment variable:

   export CODECOV_TOKEN="02468aef00-1234..."
   

4. Send the coverage data to codecov.io:

   codecov
   

   You can combine steps 3 and 4 into a simple bash script.

Then go to your project page on codecov.io and see the results.

Once you know how Codecov works, you can easily add it to your Travis-CI configuration file and let Travis do the work for you.

Integrate Codecov in a Travis-CI Build

To add Codecov to your CI build, add it to your .travis.yml file.

You need to change and add a few things in your travis config file. There is more than one way to do this, so please read about on the web. This is just one example.

install:
  - pip install codecov

script:
  - (run your unit tests with "coverage" to generate code coverage data)

after_success:
  - codecov

We installed codecov as a package in the “install” phase rather than adding it to requirements.txt. That’s becuase codecov is only needed for CI, not for running our app in production.

Example Python Project has example and explanation of configuring your project for codecov.

Code Coverage for Java Projects

The most popular Java code coverage tools (still being maintained) are:

• Clover by Atlassian
• JaCoCo - the web page mentions Eclipse but JaCoCo works with any Java project, not just Eclipse. JaCoCo computes complexity and reports how much “complexity” was tested or missed.
• IntelliJ IDEA has a built-in coverage tool for Java

---

Resources

• Python Coverage Documentation
• Introduction to Code Coverage on DZone has example using Javascript and links to some popular code coverage tools.
• Linting Python in VS Code: How to use Pylint, Flake8, etc. in VS Code

For Java:

• JaCoCo a code coverage tool for Java.
• OpenClover code coverage and source analysis for Java based on Clover.
• Comparison of Code Coverage Tools on Atlassian (owner of Clover code coverage tool) compares features of some Java tools, including the free open-source tools JaCoco, JCov, and PIT.