17

I have to test a library that provides its own memory allocation routine:

void* allocation_routine(size_t size) throw();

Documentation states that this function allocates at least size bytes of memory (it's allowed to allocate more). By the way the function uses posix_memalign internally, but the implementation may change.

I'm wondering if it's possible to write a unit test for that kind of functions? How can we test whether the required amount of memory was allocated?

UPDATE:

If we can't write a unit test then what is the closest solution?

Improve this question
9
  • 3
    You will want to choose either C or C++ tags. The languages are quite different in exception handling, see How to throw an exception in C? or throw expression - cppreference.com Commented Apr 16, 2019 at 6:44
  • 1
    Possibly related: Check if a pointer points to allocated memory on the heap. Commented Apr 16, 2019 at 6:45
  • I suggest mocking the standard memory allocation functions to check if they are called (and with which size). Most implementations allow this relatively easily. For specific help, you'll need to provide which specific implementation you're working with, and (as David C. Rankin asked) specify which language you're using (C and C++ are different languages). Commented Apr 16, 2019 at 6:51
  • For example in gcc (C), you'd use memory allocation hooks. Commented Apr 16, 2019 at 7:10
  • @DavidC.Rankin This routine is from the .cpp file. But the code is very C-style. I would suppose that C-solution is more appropriate there. Commented Apr 16, 2019 at 7:36

3 Answers 3

Reset to default
16

You cannot write a unit test for this function, because you cannot allocate memory on the heap without a system call. Hence, this is an integration test, as you are unable of isolating the unit under test from the operating system.

I would create a new, small executable that calls allocation_routine for n bytes. Depending on what allocation_routine is supposed to return, you can assert that it's non-nullptr. Then, write n bytes into this region of memory. Compile and link it using the address sanitizer (available with both gcc and clang), then try to integrate it into the test runner of your application (ctest etc.).

You might also want to restrict the available heap via the POSIX setrlimit to verify the behvaior when the allocation fails.

Improve this answer
Sign up to request clarification or add additional context in comments.

7 Comments

Also I suggest to fill each allocated block with some pseudorandom sequence (for example, you can use PRNG generator based on xoroshiro hash). You can store seed value and later play back same sequence and check memory block correctness. Or you can just fill memory block with random values, compute and store it's hash, and check it later (but when you are using first method you can find which bytes damaged exactly).
And as basic test, on each step (among of thousands) you can do one of the following (select randomly): 1) allocate random sized memory block, fill it with PRNG values; or 2) check correctness of some randomly selected block and deallocate it.
Thank you! I'm new to linux. Could you please explain what did you mean by address sanitizer? I'm currently using Intel C/C++ compiler and LLVM's lit to run the tests. I'd like to avoid using external programs.
Have a look at the docs. Start with the example given, get familiar with the command line flag. gcc has a very similar sanitizer/flag, for icc I can't tell. An address sanitizer adds runtime checks that will report memory violations like out of bounds access, overflows etc.
Also, every time when selecting random size of next allocated memory block, you can increase this size a little. So blocks sizes will grow in time. I that way test will show you how custom allocator is sustaned of fragmentation.
|
0

It's not unit test, but you can use Valgrind to get different information about memory.

  • memory leaks (which are not free)
  • memory issues

It's mainly used for debugging, but it will warns you if something is not allocated well.

Improve this answer

Comments

0

Note that while you can't unit test a malloc replacement directly you can if there is any level of indirection. Otherwise as @lubgr has stated this is an integration test.

Indirection can help on both sides. If you have a malloc wrapper that counts memory for example:

extern "C" void* malloc(const size_t size) __THROW { return getAllocator().malloc(size); } struct Allocator { using MallocFunc = std::function<void*(size_t)>; unsigned long numAllocs = 0; size_t totalMemAllocated = 0; MallocFunc = __libc_malloc; void* malloc(const size_t size) { ++numAllocs; totalMemAllocated += size; return __libc_malloc(size); } }

You could instead set MallocFunc to point to your own mock version of malloc() instead of __libc_malloc to test the accounting mechanism works properly.

Likewise you can test the allocation routine itself if you provide interposed or mock versions of the system calls it uses in the unit test code. If using gcc you might consider using symbol wrapping or the LD_PRELOAD mechanism for this.

Remember that integration tests are important as well. The assumptions you make during unit testing may not hold.

Improve this answer

Comments

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.