Acutest Readme

What Is Acutest

Acutest is C/C++ unit testing facility aiming to be as simple as possible, not to stand in the developer's way and to minimize any external dependencies.

To achieve that, the complete implementation resides in a single C header file, and its core depends only on few standard C library functions.

Overview

Main features:

Unit tests in C or C++ are supported.
No need to install/setup/configure any testing framework. Acutest is just a single header file, acutest.h.
The header provides the program entry point (function main()).
Minimal dependencies: Core features only depend on few standard C headers, optional features may use more if available on the particular system.
Trivial interface for writing unit tests: Few preprocessor macros described further below.
Rudimentary support for Test Anything Protocol (use --tap option).
Rudimentary support for xUnit-compatible XML output (use --xml-output=FILE).

C++ specific features:

Acutest catches any C++ exception thrown from any unit test function. When that happens, the given test is considered to fail.
If the exception is derived from std::exception, what() is written out in the error message.

Unix/Posix specific features:

By default, every unit test is executed as a child process.
By default, if the output is directed to a terminal, the output is colorized.
If the system offers Posix timer (clock_gettime()), user can measure test execution times with --time=real (same as --time) and --time=cpu.

Linux specific features:

If a debugger is detected, the default execution of tests as child processes is suppressed in order to make the debugging easier.

Windows specific features:

By default, every unit test is executed as a child process.
If a debugger is detected, the default execution of tests as child processes is suppressed in order to make the debugging easier.
By default, if the output is directed to a terminal, the output is colorized.
Acutest installs a SEH filter to print out uncaught SEH exceptions.
User can measure test execution times with --time.

Any C/C++ module implementing one or more unit tests and including acutest.h, can be built as a standalone program. We call the resulted binary as a "test suite" for purposes of this document. The suite is then executed to run the tests, as specified with its command line options.

We say any unit test succeeds if and only if:

all condition checks (as described below) called throughout its execution pass;
the test does not throw any exception (C++ only); and
(on Windows or Unix) the unit test subprocess is not interrupted/terminated (e.g. by a signal on Unix or SEH on Windows).

Writing Unit Tests

Basic Use

To use Acutest, simply include the header file acutest.h on the beginning of the C/C++ source file implementing one or more unit tests. Note the header provides implementation of the main() function.

#include "acutest.h"

Every test is supposed to be implemented as a function with the following prototype:

void test_example(void);

The tests can use some preprocessor macros to validate the test conditions. They can be used multiple times, and if any of those conditions fails, the particular test is considered to fail.

TEST_CHECK is the most commonly used testing macro which simply tests a boolean condition and fails if the condition evaluates to false (or zero).

For example:

void test_example(void)
{
    void* mem;
    int a, b;

    mem = malloc(10);
    TEST_CHECK(mem != NULL);

    mem = realloc(mem, 20);
    TEST_CHECK(mem != NULL);
}

TEST_ASSERT is similar to TEST_CHECK but, if it fails, it aborts execution of the current unit test instantly, either by calling abort() if the test is executed as a child process, or via longjmp() if it is not.

(Therefore you should use it only if you understand the costs connected with such brutal abortion of the test, like e.g. unflushed file descriptors, memory leaks, C++ objects destructed without calling their destructors etc., depending on what your unit test does.)

Note that the tests should be completely independent on each other. Whenever the test suite is invoked, the user may run any number of tests in the suite, in any order. Furthermore by default, on platforms where supported, each unit test is executed as a standalone (sub)process.

Finally, the test suite source file has to list the unit tests, using the macro TEST_LIST. The list specifies name of each test (it has to be unique) and pointer to a function implementing the test. I recommend names which are easy to use on command line: especially avoid space and other special characters which might require escaping in shell; also avoid dash (-) as a first character of the name, as it could then be interpreted as a command line option and not as a test name.

TEST_LIST = {
   { "example", test_example },
   ...
   { NULL, NULL }
};

Note the test list has to be ended with zeroed record.

For a basic test suites this is more or less all you need to know. However Acutest provides some more macros which can be useful in some specific situations. We cover them in the following sub-sections.

Testing C++ Exceptions

For C++, there is an additional macro TEST_EXCEPTION for verifying the given code (typically just a function or a method call) throw the expected type of exception. The check fails if the function does not throw any exception or if it throws anything incompatible.

For example:

void test_example(void)
{
    TEST_EXCEPTION(CallSomeFunction(), std::exception);
}

Richer Failure Diagnosis

If a condition check fails, it is often useful to provide some additional information about the situation so the problem is easier to debug. Acutest provides the macros TEST_MSG and TEST_DUMP for this purpose.

The former one outputs any printf-like message, the other one outputs a hexadecimal dump of a provided memory block.

So for example:

void test_example(void)
{
    char produced[100];
    char expected[] = "Hello World!";

    SomeSprintfLikeFunction(produced, "Hello %s!", "world");
    TEST_CHECK(strcmp(produced, expected) == 0);
    TEST_MSG("Expected: %s", expected);
    TEST_MSG("Produced: %s", produced);

    /* Or, if the function could provide some binary stuff, we might rather
     * use TEST_DUMP instead in order to output a hexadecimal dump of the data.
     */
    TEST_DUMP("Expected:", expected, strlen(expected));
    TEST_DUMP("Produced:", produced, strlen(produced));
}

Note that both macros output anything only when the most recently checking macro has failed. In other words, these two are equivalent:

if(!TEST_CHECK(some_condition != 0))
    TEST_MSG("some message");

TEST_CHECK(some_condition != 0);
TEST_MSG("some message");

(Note that TEST_MSG requires the compiler with variadic macros support.)

Loops over Test Vectors

Sometimes, it is useful to construct the testing function as a loop over some data providing a collection of test vectors and their respective expected outputs. For example imagine our unit test is supposed to verify some kind of a hashing function and we've got a collection of test vectors for it in its specification.

In such cases, it is very useful to get some name in the output log so that if any check fails, it is easy to identify the guilty test vector. However, it may be impractical to add such name to every checking macro.

To solve this, Acutest provides the macro TEST_CASE. The macro specifies a string serving as the test vector name. When used, Acutest makes sure that the provided name precedes any message from subsequent condition checks in its output log (until TEST_CASE is used again or the whole unit test ends).

For example lets assume we are testing SomeFunction() which is supposed, for a given byte array of some size, return another array of bytes in a newly malloc-ed buffer. Then we can do something like this:

struct TestVector {
    const char* name;
    const uint8_t* input;
    size_t input_size;
    const uint8_t* expected_output;
    size_t expected_output_size;
};

struct TestVector test_vectors[] = {
    /* some data */
};

void test_example(void)
{
    int i;
    const uint8_t* output;
    size_t output_size;

    for(i = 0; i < sizeof(test_vectors) / sizeof(test_vectors[0]); i++) {
        struct TestVector* vec = &test_vectors[i];

        /* Output the name of the tested test vector. */
        TEST_CASE(vec.name);

        /* Now, we can check the function produces what it should for the
         * current test vector. If any of the following checking macros
         * produces any output (either because the check fails, or because
         * high `--verbose` level is used), Acutest also outputs the  currently
         * tested vector's name. */
        output = SomeFunction(vec->input, vec->input_size, &output_size);
        if(TEST_CHECK(output != NULL)) {
            TEST_CHECK(output_size == vec->expected_output_size);
            TEST_CHECK(memcmp(output, vec->expected_output, output_size) == 0);
            free(output);
        }
    }
}

To explicitly reset the test vector name, typically after such a loop if the unit test makes some subsequent checks not related to any particular test vector anymore, use NULL as the macro's parameter:

void test_example(void)
{
    ...

    /* Reset the test vector name. */
    TEST_CASE(NULL);

    ...
}

Custom Log Messages

Many of the macros mentioned in the earlier sections have a counterpart which allows to output a custom messages instead of some default ones.

All of these have the same name as the aforementioned macros, just with the underscore suffix added. With the suffix, they then expect printf-like string format and corresponding additional arguments.

So, for example, instead of the simple checking macros

TEST_CHECK(a == b);
TEST_ASSERT(x < y);
TEST_EXCEPTION(SomeFunction(), std::exception);

we can use their respective counterparts with a custom messages:

TEST_CHECK_(a == b, "%d is equal to %d", a, b);
TEST_ASSERT_(x < y, "%d is lower than %d", x, y);
TEST_EXCEPTION_(SomeFunction(), std::exception, "SomeFunction() throws std::exception");

You should use some neutral wording for them because, with the command line option --verbose, the messages are logged out even if the respective check passes successfully.

(If you need to output some diagnostic information just in the case the check fails, use the macro TEST_MSG. That's exactly its purpose.)

Similarly, instead of

TEST_CASE("name");

we can use richer

TEST_CASE_("iteration #%d", 42);

However note all of these can only be used if your compiler supports variadic preprocessor macros. Variadic macros became a standard part of the C language with C99.

Building the Test Suite

When we are done with implementing the tests, we can simply compile it as a simple C/C++ program. For example, assuming cc is your C compiler:

$ cc test_example.c -o test_example

Running Unit Tests

When the test suite is compiled, the resulted testing binary can be used to run the tests.

Exit code of the test suite is 0 if all the executed unit tests pass, 1 if any of them fails, or any other number if an internal error occurs.

By default (without any command line options), it runs all implemented unit tests. It can also run only subset of the unit tests as specified on the command line:

$ ./test_example                # Runs all tests in the suite
$ ./test_example test1 test2    # Runs only tests specified
$ ./test_example --skip test3   # Runs all tests but those specified

Note that a single command line argument can select a whole group of test units because Acutest implements several levels of unit test selection (the 1st one matching at least one test unit is used):

Exact match: When the argument matches exactly the whole name of a unit test then just the given test is selected.
Word match: When the argument does not match any complete test name, but it does match whole word in one or more test names, then all such tests are selected. (Note that space , tabulator \t, dash - and underscore _ are recognized as word delimiters in the test names.)
Substring match: If even the word match failed to select any test, then all tests with a name which contains the argument as its substring are selected.

By adopting an appropriate test naming strategy, this allows user to run (or to skip if --skip is used) easily whole family of related tests with a single command line argument.

For example consider test suite test_example which implements tests foo-1, foo-2, foomatic, bar-1 and bar-10:

$ ./test_example bar-1   # Runs only the test 'bar-1' (exact match)
$ ./test_example foo     # Runs 'foo-1' and 'foo-2' (word match)
$ ./test_example oo      # Runs 'foo-1', 'foo-2' and 'foomatic' (substring match)
$ ./test_example 1       # Runs 'foo-1' and 'bar-1' (word match)

You may use --list or -l to just list all unit tests implemented by the given test suite:

$ ./test_example --list

To see description for all the supported command line options, run the binary with the option --help:

$ ./test_example --help

FAQ

Q: Wasn't this project known as "CUTest"?

A: Yes. It has been renamed as the original name was found to be too much overloaded.

Q: Do I need to distribute file README.md and/or LICENSE.md?

A: No. The header acutest.h includes URL to our repo, copyright note and the MIT license terms inside of it. As long as you leave those intact, we are completely fine if you only add the header into your project. After all, the simple use and all-in-one-header nature of it is our primary aim.

License

Acutest is covered with MIT license, see the file LICENSE.md or beginning of acutest.h for its full text.

More Information

The project resides on github:

http://github.com/mity/acutest

You can find the latest version of Acutest there, contribute with enhancements or report bugs.

arr2036 / acutest