Skip to content

whot/unit-tests-for-static-functions

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Unit-testing static functions in C

This repo shows a method to write unit-tests for static functions in C.

C source files usually have a number of static functions. By definition, these functions are local to the file and not available elsewhere. Many of these functions are helper functions and ideal candidate for unit-tests - except that you can't test them, they are local after all.

The most likely candidate that matches these criteria is the function that parses your commandline arguments. Rarely unit-tested but complex enough to need tests. But often declared as a static void parse_args(...) or similar.

Options you have to test static functions in C code:

  • move those functions into other files and/or declare them in headers, thus making them non-static. This is the best solution in many cases but you may end up polluting your internal namespace with all the new function names you now need to come up with.
  • customize the build with e.g. #ifdef TESTING to change how these functions are accessible during test builds. I find this makes the code worse and harder to understand.
  • #define static /* * / in your test build to make all static functions unconditionally public. You'll need -Wno-implicit-function-declarations to not get swamped by compiler warnings. This may not work if two static functions use the same name.
  • ... ?

The solution illustrated here

  • can run against a single source file at a time,
  • can easily setup different tests against the same source file,
  • does not modify the source code at all,
  • allows for easy mocking of functions

It relies on linker to ignore missing symbols (these are unused anyway) but otherwise uses standard C features.

The solution outlined below is available in this repo, run meson builddir && ninja -C builddir && ninja -C builddir test to reproduce.

Example

Let's say our source file example.c contains the following code:

#include <stdio.h>
#include <stdbool.h>

static bool is_acceptable_id(unsigned int id) {

    if (database_id_exists(id))
        return 0;

    return id > 1000 && id < 10000;
}

void some_function(int argument) {
    function_somewhere_else(argument);
}

int main(int argc, char **argv) {
    int id = atoi(argv[1]);
    int is_acceptable = is_acceptable_id(id);

    printf("ID is acceptable: %d\n", is_acceptable);

    return 0;
}

some_function() can be unit-tested, but is_acceptable_id() is static. But we really want to unit-test is_acceptable_id() that one, it's simple and important. Writing tests for it means having to declare it properly, possibly move it to a different file, etc. If you have worked on larger C projects, you can probably think of a number of candidates analogous to this example.

Solution

The solution has three components: using #include to get access to the static methods, moving main out of the way where necessary and a set of linker flags to work around the resulting difficulties.

This approach does not restrict you in the type of test suite you can use. In the examples below, I'll simply use assert().

Step 0: write the test

Let's write the test cases first, our test-example.c file looks like this:

#include <assert.h>

#include <test-suite-header.h>
#include <other-stuff-you-need.h>

static void test_id(void) {
    assert(!is_acceptable_id(10));
    assert(!is_acceptable_id(100));
    assert(!is_acceptable_id(1000));

    assert(is_acceptable_id(5000));
    assert(is_acceptable_id(3000));
}

int main(int argc, char **argv)
{
    test_id();
    return 0;
}

Fairly straighforward. As mentioned above you are not restricted in the type of test suite to use.

Step 1: include the real source file

Including the source file gives us access to all the static functions within the file:

#include "example.c"    // <---- include the file

#include <assert.h>

#include <test-suite-header.h>
#include <other-stuff-you-need.h>

static void test_id(void) {
    assert(!is_acceptable_id(10));
    assert(!is_acceptable_id(100));
    assert(!is_acceptable_id(1000));

    assert(is_acceptable_id(5000));
    assert(is_acceptable_id(3000));
}

int main(int argc, char **argv)
{
    test_id();
    return 0;
}

Compile with gcc -o test test-example.c but of course this won't work because we have two main() methods here. On to step 2 which is optional if you are testing a source file without a main().

Step 2: move main out of the way

Skip this step if the source file doesn't have a main().

Let's rename main to some other name with a #define. This one is best supplied with a compiler argument (-Dmain=_main_disabled) so it applies to all source files you provide. But we want our test's main() to work, so we just #undef main before defining that one.

#include "example.c"

#include <assert.h>

#include <test-suite-header.h>
#include <other-stuff-you-need.h>

static void test_id(void) {
    assert(!is_acceptable_id(10));
    assert(!is_acceptable_id(100));
    assert(!is_acceptable_id(1000));

    assert(is_acceptable_id(5000));
    assert(is_acceptable_id(3000));
}

#undef main                     // <--- stop renaming 'main'
int main(int argc, char **argv)
{
    test_id();
    return 0;
}
        

Compile this with gcc -Dmain=_main_disabled -o test test-example.c and we're almost there, except we get a bunch of missing symbols.

Step 3: mock the missing pieces

We can't compile this file yet, because we're missing a bunch of functions like database_id_exists(). Plus, this function is one that we need to mock anyway because we don't have a database during testing.

So we mock that one:

#include "example.c"

#include <test-suite-header.h>
#include <other-stuff-you-need.h>

bool database_id_exists(unsigned int id)        // <--- mock a missing function
{
    return false;
}

static void test_id(void) {
    assert(!is_acceptable_id(10));
    assert(!is_acceptable_id(100));
    assert(!is_acceptable_id(1000));

    assert(is_acceptable_id(5000));
    assert(is_acceptable_id(3000));
}

#undef main
int main(int argc, char **argv)
{
    test_id();

    return 0;
}

Simple enough. Still won't work because we are missing all the functions defined in other source files, function_somewhere_else() in our example.c.

Step 4: hook up the compiler and linker

We still can't compile, because we don't have function_somewhere_else(). But that one is never called by our test, so we just tell the linker to ignore the error with --unresolved-symbols=ignore-all.

So our compiler command is now gcc -Dmain=_main_disabled -Wl,--unresolved-symbols=ignore-all -o test test-example.c. And that's basically it, we now have test-example.c being a unit-test for the static functions within example.c.

However, for (largely cosmetic) reasons you need a few more flags which are illustrated in this meson snippet:

e = executable('test-example',
               'example.c', 'test-example.c',  # the source files
               dependencies: [dep_ext_library],
               c_args: ['-Dmain=_main_disabled'
                        '-Wno-missing-prototypes'],
               link_args: ['-Wl,--unresolved-symbols=ignore-all',
                           '-Wl,-zmuldefs',
                           '-no-pie'],
               install: false),
)
test('test-example', e)

Full explanation of everything here:

  • executable is meson lingo for "build an executable" and the first argument is the name of the resulting binary
  • example.c and test-example.c are the input source files. You don't technically need example.c but it helps meson re-trigger the test built when that file changes.
  • dependencies is the list of external/internal libraries we need to link to. You must include the dependencies whose header files you require, and any libraries you want to call into as part of your test.
  • c_args are compiler flags. Both are only required if you are testing a source file with a main() method.
    • -Dmain=_main_disabled renames the main() to _main_disabled()
    • -Wno-missing-prototypes stops compiler warnings about missing prototypes (which you'd get for _main_disabled()).
  • link_args are the linker flags:
    • --unresolved-symbols=ignore-all: tell the linker to ignore any unresolved symbols. This means even if example.c uses functions from some other source file, we can successfully link our test executable.
    • -zmuldefs: allow symbols to be defined more than once. example.c defines multiply() but it gets redefined when it is included in test-example.c. This triggers linker errors that we need to ignore. This is cosmetic only, you can drop example.c from the sources list instead.
    • -no-pie: required wherever pie is enabled by default (e.g. Arch). We'll likely end up with undefined symbols, position-independent code gets relocated on execution which will go boom if we have undefined symbols. So let's not do that for this test.

Alternatively, if you don't want to deal with unresolved symbols, just build with all the source files that the original binary needs. Mocking gets harder but it may be sufficient for your use-cases. You'll only need -zmuldefs then.

example = executable('example', <list of source files>)
e = executable('test-example',
               'test-example.c', <list of source files>,
               dependencies: [dep_ext_library],
               c_args: ['-Dmain=_main_disabled'
                        '-Wno-missing-prototypes'],
               link_args: ['-Wl,-zmuldefs'],
               install: false),
)
test('test-example', e)

Note how test-example.c is the first file. This way the mocked function is defined first and the linker uses it instead of the real function defined in the other file.

Note: I'm not sure if this is guaranteed linker behaviour and safe to rely on

Caveats

It's a hack, so it's not as polished as it needs to be. Specifically:

  • Linker errors are incredibly frustrating to debug. Since we turn off all the warnings, you won't get much feedback why things break. Best solution here is to try to add more dependencies until it works and removing the linker instructions to search in the warnings for something that isn't related to your code file.
  • It's a re-build of the code so you're not testing the exact same bits as you ship. How much that is an issue depends on you and your use case.

About

An example how to unit-test static functions in C

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published