Valgrind reports memory leak when assigning a value to a string

Question

Valgrind reports a memory leak when assigning a value to a string.

I used the following simple code to test an memory leak reported by Valgrind.

/******************************************
* FILE: t3.c
* Compiled using : g++ -g t3.c -o t3
*
* $ g++ -v
* Reading specs from /usr/lib/gcc/i686-pc-linux-gnu/3.4.6/specs
* Configured with: ./configure --prefix=/usr --infodir=/share/info --mandir=/share/man
*      --enable-languages=c,c++ --with-system-zlib --program-suffix=-3.4 --enable-threads=posix
* Thread model: posix
* gcc version 3.4.6
 ******************************************/


#include <iostream>
#include <string>

using namespace std;

/**************************************************************
 **************************************************************/
int main(int argc, char *argv[])
{
   string test = "XXXXXXXXX";
   cout << "this is a test " << test << endl;
   exit(0);
}

I compile using this command:

$ g++ -g t3.c -o t3

And when I run Valgrind it reports a memory leak when I try to assign a value to a string. I'm using this simple test to investigate some memory leak in the real program, and it seems that using string can cause some sort of problem.

By 0x8048A6F: main (t3.c:23) is the line : string test = "XXXXXXXXX"; Can someone give some hint on such strange behaviour?

[enzo@P0101222 C]$   valgrind --leak-check=full  ./t3
==3910== Memcheck, a memory error detector.
==3910== Copyright (C) 2002-2007, and GNU GPL'd, by Julian Seward et al.
==3910== Using LibVEX rev 1732, a library for dynamic binary translation.
==3910== Copyright (C) 2004-2007, and GNU GPL'd, by OpenWorks LLP.
==3910== Using valgrind-3.2.3, a dynamic binary instrumentation framework.
==3910== Copyright (C) 2000-2007, and GNU GPL'd, by Julian Seward et al.
==3910== For more details, rerun with: -v
==3910==
this is a test XXXXXXXXX
==3910==
==3910== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 25 from 1)
==3910== malloc/free: in use at exit: 102 bytes in 3 blocks.
==3910== malloc/free: 4 allocs, 1 frees, 126 bytes allocated.
==3910== For counts of detected errors, rerun with: -v
==3910== searching for pointers to 3 not-freed blocks.
==3910== checked 194,136 bytes.
==3910==
==3910== 16 bytes in 1 blocks are definitely lost in loss record 1 of 3
==3910==    at 0x4017846: malloc (m_replacemalloc/vg_replace_malloc.c:149)
==3910==    by 0x4018E05: realloc (m_replacemalloc/vg_replace_malloc.c:306)
==3910==    by 0x41B441A: argz_append (in /lib/libc-2.2.5.so)
==3910==    by 0x41593B9: __newlocale (in /lib/libc-2.2.5.so)
==3910==    by 0x40E010B: std::locale::facet::_S_create_c_locale(__locale_struct*&, char const*, __locale_struct*) (c++locale.cc:99)
==3910==    by 0x407EF6F: std::locale::facet::_S_initialize_once() (../../.././libstdc++-v3/src/locale.cc:172)
==3910==    by 0x407EFB4: std::locale::facet::_S_get_c_locale() (../../.././libstdc++-v3/src/locale.cc:185)
==3910==    by 0x407A422: std::ctype<char>::ctype(unsigned short const*, bool, unsigned) (/usr3/BUILD/gcc/gcc-3.4.6/i686-pc-linux-gnu/libstdc++-v3/include/i686-pc-linux-gnu/bits/ctype_noninline.h:104)
==3910==    by 0x40801D5: std::locale::_Impl::_Impl(unsigned) (/usr3/BUILD/gcc/gcc-3.4.6/libstdc++-v3/libsupc++/new:92)
==3910==    by 0x4080EED: std::locale::_S_initialize_once() (/usr3/BUILD/gcc/gcc-3.4.6/libstdc++-v3/libsupc++/new:92)
==3910==    by 0x4080F84: std::locale::_S_initialize() (../../.././libstdc++-v3/src/locale_init.cc:155)
==3910==    by 0x4080FE7: std::locale::locale() (../../.././libstdc++-v3/src/locale_init.cc:102)
==3910==
==3910==
==3910== 22 bytes in 1 blocks are possibly lost in loss record 2 of 3
==3910==    at 0x4017C38: operator new(unsigned) (m_replacemalloc/vg_replace_malloc.c:163)
==3910==    by 0x40BF2C4: std::string::_Rep::_S_create(unsigned, unsigned, std::allocator<char> const&) (/usr3/BUILD/gcc/gcc-3.4.6/i686-pc-linux-gnu/libstdc++-v3/include/ext/new_allocator.h:81)
==3910==    by 0x40C1CE4: char* std::string::_S_construct<char const*>(char const*, char const*, std::allocator<char> const&, std::forward_iterator_tag) (/usr3/BUILD/gcc/gcc-3.4.6/i686-pc-linux-gnu/libstdc++-v3/include/bits/basic_string.tcc:150)
==3910==    by 0x40C1E15: std::string::string(char const*, std::allocator<char> const&) (/usr3/BUILD/gcc/gcc-3.4.6/i686-pc-linux-gnu/libstdc++-v3/include/bits/basic_string.h:1386)
==3910==    **by 0x8048A6F: main (t3.c:23)**
==3910==
==3910== LEAK SUMMARY:
==3910==    definitely lost: 16 bytes in 1 blocks.
==3910==      **possibly lost: 22 bytes in 1 blocks.**
==3910==    still reachable: 64 bytes in 1 blocks.
==3910==         suppressed: 0 bytes in 0 blocks.
==3910== Reachable blocks (those to which a pointer was found) are not shown.
==3910== To see them, rerun with: --leak-check=full --show-reachable=yes
[enzo@P0101222 C]$

Solution 1

Because you call exit(0), so the string destructor is never invoked. Just use return 0.

To elaborate, the constructor of std::string allocates heap memory to store the string, relying on the destructor to deallocate that memory. If you declare a string object on the stack, the destructor will automatically be invoked when the string object goes out of scope, thus freeing the memory. But exit is really a C mechanism; it immediately exits the program without performing stack-unwinding meaning that C++ destructors for local stack objects will not be called.

Solution 2

If you allocate five strings, do you get five times the memory leak, or is it still the same amount? If it's the same amount, then you probably don't have a leak at all. Some libraries allocate memory for internal bookkeeping/efficiency/et cetera that doesn't get released until after valgrind stops looking. These get picked up as memory leaks because your program caused the allocation but never caused a deallocation. If it's five times the amount, then your implementation of string may be at fault. I agree with Charles Salvia though... try again with return 0; instead of exit(0); and see if that changes anything.

Solution 3

In one of my computer science classes we were told that Valgrind outputs information about strings that we shouldn't worry about. Here's the suppression file that they gave us for strings: https://sites.google.com/site/complingfiles/files/string.supp

Solution 4

Despite having no exit(0) at the end of program I had similar problem with false positives with std::string . I was statically linking with libstdc++. Switching linking option to shared and compiling with GLIBCXX_FORCE_NEW suppressed the warnings.

View more solutions

Comments

[Nate Glenn]

Yeah, the school switched that class to java. I'll see if I can dig it up anywhere.

[Nate Glenn]

@Riot: found the file and posted a new link.

[BobTuckerman]

It appears that compiling with the GLIBCXX_FORCE_NEW flag doesn't actually do anything. According to the libstdc++ docs for mt_allocator, it's an environment variable. " If the GLIBCXX_FORCE_NEW environment variable is set, it sets the bool _S_force_new to true and then returns.". So, simply do something like export GLIBCXX_FORCE_NEW=1; and then run valgrind. This resolved a lot of issues I had with std::string giving false positives.