Programming in C

Unit Testing

Gerald Senarclens de Grancy

Purpose of Unit Tests

What is a Unit Test

  • Allows automated testing of independent code units
  • No replacement for other test categories
  • Write them as early as possible - if possible before the code it tests
  • Tests can serve as useful requirements specification
  • Create a test for each bug you encounter (refacturing!)

Which problems does unit testing solve?

Code without tests (aka. legacy code) is extremely difficult to work with

Unit tests...

  • are a first line of defense against bugs
  • improve code quality by dictating testable code
  • allow releasing higher quality early and often
  • give more confidence to programmers
  • are a prerequisit to refactoring
  • can be used as a contract with regard to desired behavior
  • prevent regressions (bugs in features priorly working properly)

Which problems does unit testing cause?

Essentially none, but writing unit tests ...

  • seems to cost time
  • demands discipline
  • is far from trivial
  • does not substitute other kinds of testing

What is a Test(case)

  • Answers a single question about the code it is testing
  • Run completely by itself (automation)
  • Results must not require human interpretation
  • Should be independent of other tests and their order of execution
  • Can serve as documentation of how (not) to use your code

Flaky Tests

  • Have different outcomes without changes to the code
  • Erode confidence in testing processes
  • Fix them (mute / skip them until they are fixed) or delete them

Software Quality

User's view (visible to the user)

  • Functionality
  • Reliability
  • Usability
  • Performance/ efficiency

Producer's view

  • Portability
  • Maintainability
  • Testability (vs. legacy code)
  • Transparency (is the code readable/ understandable?)

We Need a Unit Testing Framework

For writing, compiling and running unit tests, a unit testing framework is required.

There are many concurring unit testing frameworks for C.

At the time of this writing, Criterion offers the best features.

Example: Refactor a Function

#include <stdint.h>
// Return the sum of all values from 1 to n.
uint64_t add_to(uint32_t n) {
uint64_t sum = 0;
for (uint32_t i = 1; i <= n; ++i) {
sum += i;
}
return sum;
}
Download before_refactoring.c

After experiencing performance problems, we want to use a faster algorithm

#include <stdint.h>
// Return the sum of all values from 1 to n.
uint64_t add_to(uint32_t n) {
return (uint64_t) (n / 2.0) * (n + 1);
}
Download after_refactoring.c

How can we apply the change without risking to destroy our application?

  1. Define good test cases
  2. Implement test cases using your project's unit testing library
  3. Run the tests before the change to ensure they pass
  4. Apply the change, then run the tests again

1. Define Good Test Cases

Using pencil and paper / brainstorming: define relevant test cases

  • All code paths should be used
  • Add one or more common cases
  • Include relevant edge cases

25 is a possible common case. Add 26 as an even input. The edge cases should include 0 and 4294967295.
Expected results for these are 325, 351, 0 and 9223372034707292160.

Let's also add 500000000 (expected result 125000000250000000) as 4294967295 does not terminate with the current implementation.

2. Implement defined test cases

#include <stdint.h>
#include <criterion/criterion.h>
#include <criterion/new/assert.h>
uint64_t add_to(uint32_t n); // usually in header file

Test(sum_to, fast_test) {
cr_assert(eq(u64, add_to(0), 0), "Sum of 0 is 0.");
cr_assert(eq(u64, add_to(25), 325), "Sum of 1..25 is 325.");
cr_assert(eq(u64, add_to(26), 351), "Sum of 1..26 is 351.");
cr_assert(eq(u64, add_to(500000000), 125000000250000000),
"This function must work for large unsigned integers.");
}
// does not terminate for the old code version due to an overflow of the
// loop counter (the loop would terminate only at UINT_MAX + 1)
Test(sum_to, slow_test) {
cr_assert(eq(u64, add_to(UINT32_MAX), 9223372034707292160),
"The function must work for the largest unsigned integer.");
}
Download test_sum_to.c

3. Run the Unittests Before the Change

clang before_refactoring.c test_sum_to.c -l criterion -o run_tests
./run_tests # does not terminate; interrupt via [Ctrl]+[C]

What to do now?

./run_tests --help
./run_tests -l # list all available tests
./run_tests --filter sum_to/fast_test

4. Apply the Change, then Run the Tests Again

For the sake of the presentation, instead of applying the change, we use another source file.

clang after_refactoring.c test_sum_to.c -l criterion -o run_tests
./run_tests

Now the tests are fast, but they are failing. Fix the code, then compile and run the tests again.

Criterion

The comprehensive documentation can be found on readthedocs.

The installation is trivial, if it's available in your package manager.

Otherwise, follow the setup guide for building from source.

A slightly more comprehensive example also demonstrates text fixtures.

Test-driven development (TDD)

TDD is a methodology that emphasizes writing tests before the actual code

Helps ensure that code is tested thoroughly and avoids regressions

Repeat the cycle for each new piece of functionality

  1. Write a failing test that describes the desired functionality
  2. Write the minimum amount of code necessary to make the test pass
  3. Refactor the code to improve its structure and maintainability
  4. Make sure that all tests pass

Whenever you are tempted to type something into a print statement or a debugger expression, write it as a test instead.
– Martin Fowler

Questions
and feedback...