A testing system for C, contained in 1 header file.
-
Small, Portable, Lightweight
greatest doesn't depend on anything beyond ANSI C89, and the test scaffolding should build without warnings when compiled with
-Wall -Wextra -pedantic. It is under 1,000 LOC (SLOCCount), and does no dynamic allocation. -
Permissive License
greatest is released under the ISC License. You can use it freely, even for commercial purposes.
-
Easy To Set Up
To use, just
#include "greatest.h"in your project. There is very little boilerplate. Most features are optional. -
Un-Opinionated
When a command-line test runner is useful, greatest can provide one, but it can also run as part of other programs. It doesn't depend on a particular build system or other tooling, and should accommodate a variety of testing approaches. It actively avoids imposing architectural choices on code under test. While greatest was designed with C in mind, it attempts to be usable from C++.
-
Modular
Tests can be run individually, or grouped into suites. Suites can share common setup, and can be in distinct compilation units.
-
Low Friction
Specific tests or suites can be run by name, for focused and rapid iteration during development. greatest adds very little startup latency.
There are some compile-time options, and slightly nicer syntax for parametric testing (running tests with arguments) if compiled with a C99 or later language standard.
I wrote a blog post with more information. While it's several years old, it's still accurate about the main functionality.
theft, a related project, adds property-based testing.
#include "greatest.h"
/* A test runs various assertions, then calls PASS(), FAIL(), or SKIP(). */
TEST x_should_equal_1(void) {
int x = 1;
/* Compare, with an automatic "1 != x" failure message */
ASSERT_EQ(1, x);
/* Compare, with a custom failure message */
ASSERT_EQm("Yikes, x doesn't equal 1", 1, x);
/* Compare, and if they differ, print both values,
* formatted like `printf("Expected: %d\nGot: %d\n", 1, x);` */
ASSERT_EQ_FMT(1, x, "%d");
PASS();
}
/* Suites can group multiple tests with common setup. */
SUITE(the_suite) {
RUN_TEST(x_should_equal_1);
}
/* Add definitions that need to be in the test runner's main file. */
GREATEST_MAIN_DEFS();
int main(int argc, char **argv) {
GREATEST_MAIN_BEGIN(); /* command-line options, initialization. */
/* Individual tests can be run directly in main, outside of suites. */
/* RUN_TEST(x_should_equal_1); */
/* Tests can also be gathered into test suites. */
RUN_SUITE(the_suite);
GREATEST_MAIN_END(); /* display results */
}Output:
$ make simple && ./simple
cc -g -Wall -Werror -pedantic simple.c -o simple
* Suite the_suite:
.
1 test - 1 passed, 0 failed, 0 skipped (5 ticks, 0.000 sec)
Total: 1 test (47 ticks, 0.000 sec), 3 assertions
Pass: 1, fail: 0, skip: 0.Test cases should call assertions and then end with PASS(), SKIP(),
FAIL(), or one of their message variants (e.g. SKIPm("TODO");).
If there are any test failures, the test runner will return 1,
otherwise it will return 0. (Skips do not cause the test runner to
report failure.)
Tests and suites are just functions, so normal C scoping rules apply. For example, a test or suite named "main" will have a name collision.
(For more examples, look at example.c and example_suite.c.)
greatest runs all tests by default, but can be configured to only run suites or tests whose names contain a filter string, and/or exclude tests whose name contains a filter string. When test name filtering and exclusion are used together, exclusion takes precedence.
void greatest_set_suite_filter(const char *name);
void greatest_set_test_filter(const char *name);
void greatest_set_test_exclude(const char *name);
These correspond to the following command line test runner options:
`-s SUITE`: Only run suites whose names contain the string "SUITE"
`-t TEST`: Only run tests whose names contain the string "TEST"
`-x EXCLUDE`: Exclude tests whose names contain the string "EXCLUDE"
For example, to run any tests with "tree" in the name, in suites with "pars" in the name (such as "parser"), but exclude any tests whose names also contain "slow":
./test_project -s pars -t tree -x slow
The string matching includes optional test name suffixes.
Assertions fail the current test unless some condition holds. All
assertions have a "message" variant (with an m suffix), which takes a
custom failure message string as their first argument. For example, the
assertion ASSERT_EQ(apple, orange); could instead be used like
ASSERT_EQm("these should match", apple, orange). Non-message
assertions create a default message.
Assert that COND evaluates to a true (non-zero) value.
Assert that COND evaluates to a false (zero) value.
Assert that EXPECTED == ACTUAL. To print the values if they
differ, use ASSERT_EQ_FMT. To compare with custom equality test
and print functions, use ASSERT_EQUAL_T instead.
Assert that EXPECTED == ACTUAL. If they are not equal, print their
values using FORMAT as the printf format string.
For example: ASSERT_EQ_FMT(123, result, "%d"); will call printf
like printf("Expected: %d\nGot: %d\n", 123, result); if its
EXPECTED and ACTUAL arguments don't match.
Note: EXPECTED and ACTUAL will be evaluated more than once on
failure, so they should not be a function call with side effects.
(Since their type is not known by the macro, they cannot be
captured in a local variable. typeof is a GCC extension.)
Assert that ACTUAL is within EXPECTED +/- TOLERANCE, once the values
have been converted to a configurable floating point type
(GREATEST_FLOAT).
greatest does not depent on floating point math.
It is only used within ASSERT_IN_RANGE's macro expansion.
Assert that the strings are equal
(i.e., strcmp(EXPECTED, ACTUAL) == 0).
Assert that the first SIZE bytes of the strings are equal
(i.e., strncmp(EXPECTED, ACTUAL, SIZE) == 0).
Assert that the first SIZE bytes of memory pointed to by EXPECTED and ACTUAL are equal. If their memory differs, print a hexdump and highlight the lines and individual bytes which do not match.
Assert that the enum value EXPECTED is equal to ACTUAL. If not, convert
each enum value to a string using ENUM_STR_FUN before printing them.
ENUM_STR_FUN should have a type like:
const char *FUN(int x);
Assert that EXPECTED and ACTUAL are equal, using the greatest_equal_cb
function pointed to by TYPE_INFO->equal to compare them. The
assertion's UDATA argument can be used to pass in arbitrary user data
(or NULL). If the values are not equal and the TYPE_INFO->print
function is defined, it will be used to print an "Expected: X, Got: Y"
message.
Assert that COND evaluates to a true value. If not, then use
longjmp(3) to immediately return from the test case and any
intermediate function calls.
If built with GREATEST_USE_LONGJMP #defined to 0, then all
setjmp/longjmp-related functionality will be compiled out. This also
reduces memory usage by sizeof jmp_buf, which may be several hundred
bytes, depending on the platform.
Groups of suites or tests can be run in random order by using
GREATEST_SHUFFLE_SUITES and GREATEST_SHUFFLE_TESTS, respectively.
This can help find and eliminate accidental coupling between tests.
The shuffling depends on the seed and the test/suite count, so a consistent seed will only lead to reproducible ordering until the group's count changes.
Shuffling suites:
SHUFFLE_SUITES(seed, {
RUN_SUITE(suite1);
RUN_SUITE(suite2);
RUN_SUITE(suite3);
RUN_SUITE(suite4);
RUN_SUITE(suite5);
});
Shuffling tests:
SHUFFLE_TESTS(seed, {
RUN_TEST(test_a);
RUN_TEST1(test_b, 12345);
RUN_TEST(test_c);
RUN_TESTp(test_d, "some_argument");
RUN_TEST(test_e);
});
Note: Any other code inside the block will be executed several times.
The shuffling macro expands to a loop with (count + 1) iterations -- the
first pass counts, and the following passes only execute the next chosen
suite/test. In particular, avoid running tests directly inside of a
SHUFFLE_SUITES block without a suite, because the test will be run
on each iteration.
greatest_set_test_suffix can be used to set an optional name suffix
for the next test:
for (i = 0; i < row_count; i++) {
const struct table_row *row = &table[row_count];
greatest_set_test_suffix(row->name);
RUN_TEST1(test_with_arg, row);
}
This will cause the test name to print with a _ separator and the
suffix in all pass/fail/skip messages (i.e., test_with_arg_KEY). This
is especially useful when running a test several times with different
arguments, in shuffled order. The name suffix is included when using
name-based filtering.
The test name and optional suffix are copied into an internal buffer.
Its size can be configured by #defineing the constant
GREATEST_TESTNAME_BUF_SIZE. (If not #defined, it defaults to 128
bytes.) If the buffer is not large enough for the name and suffix, it
will truncate after size - 1 bytes, to ensure that it is properly
\0-terminated.
The name suffix pointer is cleared after each RUN_TEST* call, so a
suffix can be constructed in a stack allocated buffer without later
dereferencing a pointer that has gone out of scope.
Because of how PASS(), ASSERT(), FAIL(), etc. are implemented
(returning a test result enum value), calls to functions that use them
directly from test functions must be wrapped in CHECK_CALL:
TEST example_using_subfunctions(void) {
CHECK_CALL(less_than_three(5));
PASS();
}
This is only necessary if the called function can cause test failures.
Test runners build with the following command line options:
Usage: (test_runner) [--help] [-hlfv] [-s SUITE] [-t TEST]
-h, --help print this Help
-l List suites and tests, then exit (dry run)
-f Stop runner after first failure
-a Abort on first failure (implies -f)
-v Verbose output
-s SUITE only run suite w/ name containing SUITE substring
-t TEST only run test w/ name containing TEST substring
-t EXCLUDE exclude tests containing string EXCLUDE substring
Any arguments after -- will be ignored.
If you want to run multiple test suites in parallel, look at parade.
These command line options are processed by GREATEST_MAIN_BEGIN();.
Rather than producing a command line test runner (which checks the
command line options, and exits with a pass/fail return code after
running tests), greatest can be used more like a library. Instead of
using GREATEST_MAIN_BEGIN(), use GREATEST_INIT() to (re-)initialize
greatest, then use either GREATEST_PRINT_REPORT() to print the report
to GREATEST_STDOUT, or use greatest_get_report(&report) to get the
pass, fail, skip, and assertion counters.
The command line flags above have corresponding functions:
greatest_stop_at_first_fail()greatest_abort_on_fail()greatest_list_only()greatest_set_suite_filter(const char *filter)greatest_set_test_filter(const char *filter)greatest_set_test_exclude(const char *filter)greatest_set_verbosity(unsigned int verbosity)
Most of the macros have prefixed and unprefixed forms. For example,
SUITE is the same as GREATEST_SUITE.
Check the source for the list -- search for GREATEST_USE_ABBREVS.
These aliases can be disabled by #define-ing GREATEST_USE_ABBREVS to 0.
If you want color output (PASS in green, FAIL in red, etc.), you can
pipe the output through the included greenest script in contrib/:
$ ./example -v | greenest(Note that greenest depends on a Unix-like environment.)
greatest itself doesn't have built-in coloring to stay small and portable.
There is an awk script provided, contrib/entapment, that converts the
verbose output from the default CLI test runner to TAP version 13
format:
./example -v | contrib/entapment