How are death tests supposed to work?

[krzikalla]

Actually I know how to use them. But I wonder, if their interaction with NDEBUG is fine.
Simple example:

void foo(const std::string& s, int index)
{
  assert(index < s.size());
  std::cout << s[index];
}

// Should I take this one?
EXPECT_DEATH(foo("abc", 5));

// Or this one?
EXPECT_DEBUG_DEATH(foo("abc", 5));

The problem with the above code is, that you don't know, if the program will crash. A triggered assertion means, that the program is in an inconsistent state and that anything (crash as well as nothing) can happen afterwards. So neither EXPECT_DEATH nor EXPECT_DEBUG_DEATHwill work reliable, when NDEBUG is defined.
Actually I am looking for a macro, which executes the statement only, if NDEBUG is not present. But I haven't found something in google test. Then what is the idea behind EXPECT_DEBUG_DEATH?