If I specify a default value for an argument of type "std::string &" in C++, could that cause a memory leak?

Question

I have a method declared like so:

/*!
\brief Removes the leading and trailing white space from a string.
\param s The string to remove the white space from.
\param white_chars Characters to be considered as whitespace.
*/
std::string Trim(const std::string &s, const std::string &white_chars = " \t\n\r");

The definition of the method is uninteresting, but here it is anyway:

std::string Trim(const std::string &s, const std::string &white_chars)
{
    size_t startidx = s.find_first_not_of(white_chars);
    if (startidx == std::string::npos) return "";
    size_t endidx = s.find_last_not_of(white_chars);
    return s.substr(startidx, endidx - startidx + 1);
}

Now in most usages of this method, I supply just the first argument. Valgrind gives me the following warning

==3338== 68 bytes in 2 blocks are possibly lost in loss record 4,639 of 7,212
==3338==    at 0x4C2B1C7: operator new(unsigned long) (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==3338==    by 0x728CA88: std::string::_Rep::_S_create(unsigned long, unsigned long, std::allocator<char> const&) (in /usr/lib/x86_64-linux-gnu/libstdc++.so.6.0.16)
==3338==    by 0x728E2B4: char* std::string::_S_construct<char*>(char*, char*, std::allocator<char> const&, std::forward_iterator_tag) (in /usr/lib/x86_64-linux-gnu/libstdc++.so.6.0.16)
==3338==    by 0x728E414: std::basic_string<char, std::char_traits<char>, std::allocator<char> >::basic_string(std::string const&, unsigned long, unsigned long) (in /usr/lib/x86_64-linux-gnu/libstdc++.so.6.0.16)
==3338==    by 0x728E441: std::string::substr(unsigned long, unsigned long) const (in /usr/lib/x86_64-linux-gnu/libstdc++.so.6.0.16)
==3338==    by 0x6965F0A: str::Trim(std::string const&, std::string const&) (appbase.cpp:90)
==3338==    by 0x5F481D7: netopt::Element::set_ID(std::string const&) (netopt_elem.cpp:85)

Note "blocks are possibly lost"? I've seen memory leaks where Valgrind tells me "blocks are definitely lost", but this is a less certain message.

So the question becomes, am I causing a memory leak by assigning a default value to an std::string &? If so, what am I doing wrong?

Answer 1

There is no technical issue, but philosophically creating a temporary std::string at each call is not so nice. Especially with libstdc++ (which you appear to use) since it causes a memory allocation each time (lack of Short String Optimization).

Since find_first_not_of has an overload taking a char const*, it would be better to provide two overloads instead:

// First, the one with the default argument, but without a temporary
std::string Trim(std::string const& s, char const* white_chars = " \t\n\r");

// Second, for convenience, one with a `std::string`
std::string Trim(std::string const& s, std::string const& white_chars);

This also means that when calling Trim(xx, "abc") you will avoid the temporary std::string being generated :)

Of course, the overkill solution is to reuse already written code: Boost String Algorithm has many algorithms for string manipulation, including trim.

Answer 2

No problem.

The temporary string is constructed at each call, and automatically destroyed again at the end of the statement. As long as you don't save a (dangling) reference to it, there is no problem.