How can I detect if a program is running from within valgrind?

Question

Is there a way to identify at run-time of an executable is being run from within valgrind? I have a set of C++ unit tests, and one of them expects std::vector::reserve to throw std::bad_alloc. When I run this under valgrind, it bails out completely, preventing me from testing for both memory leaks (using valgrind) and behavior (expecting the exception to be thrown).

Here's a minimal example that reproduces it:

#include <vector>
int main()
{
    size_t uint_max = static_cast<size_t>(-1);
    std::vector<char> v;
    v.reserve(uint_max);
}

Running valgrind, I get this output:

Warning: silly arg (-1) to __builtin_new()
new/new[] failed and should throw an exception, but Valgrind
   cannot throw exceptions and so is aborting instead.  Sorry.
   at 0x40192BC: VALGRIND_PRINTF_BACKTRACE (valgrind.h:319)
   by 0x401C823: operator new(unsigned) (vg_replace_malloc.c:164)
   by 0x80487BF: std::vector<char, std::allocator<char> >::reserve(unsigned) new_allocator.h:92)
   by 0x804874D: main (vg.cxx:6)

I'd like to modify my unit test to simply skip the offending code when it's being run from within valgrind. Is this possible?

Answer 1

I looked at the valgrind doucmentation and didn't find an easy answer. But here are a couple of things you can try:

• Write your own wrapper around the offending new operation and raise the exception before valgrind gets its private new function going.

• Try as poster above suggested except that instead of a command-line option (which requires plumbing) use an environment variable:

  MYAPP_UNIT_TESTS_DISABLED="NEW_MINUS_ONE,FLY_TO_MOON,DEREF_NULL" valgrind myapp
  

Then you can easily write a function

bool unit_test_enabled(const char *testname);

to protect your unit test based on the value returned by getenv(3).

Answer 2

You should look at this page from the Valgrind manual, it contains a RUNNING_ON_VALGRIND macro (included from valgrind.h) which does what you want.