Why can't clang with libc++ in c++0x mode link this boost::program_options example?

Question

Compiling this example code for boost::program_options: http://svn.boost.org/svn/boost/trunk/libs/program_options/example/first.cpp

...on MacOS Lion (10.7.2), using boost-1.48.0 installed with MacPorts:

$ clang++ -v
Apple clang version 3.0 (tags/Apple/clang-211.12) (based on LLVM 3.0svn)
Target: x86_64-apple-darwin11.2.0
Thread model: posix
$ clang++ -std=c++0x --stdlib=libc++ -lc++ -I/opt/local/include -L/opt/local/lib -lboost_program_options first.cpp -o first
Undefined symbols for architecture x86_64:
  "boost::program_options::options_description::options_description(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&, unsigned int, unsigned int)", referenced from:
      _main in cc-6QQcwm.o
  "boost::program_options::operator<<(std::__1::basic_ostream<char, std::__1::char_traits<char> >&, boost::program_options::options_description const&)", referenced from:
      _main in cc-6QQcwm.o
  "boost::program_options::abstract_variables_map::operator[](std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&) const", referenced from:
      boost::program_options::variables_map::operator[](std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&) const in cc-6QQcwm.o
  "boost::program_options::detail::cmdline::set_additional_parser(boost::function1<std::__1::pair<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> >, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > >, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&>)", referenced from:
      boost::program_options::basic_command_line_parser<char>::extra_parser(boost::function1<std::__1::pair<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> >, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > >, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&>) in cc-6QQcwm.o
  "boost::program_options::detail::cmdline::cmdline(std::__1::vector<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> >, std::__1::allocator<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > > > const&)", referenced from:
      boost::program_options::basic_command_line_parser<char>::basic_command_line_parser(int, char const* const*) in cc-6QQcwm.o
  "boost::program_options::to_internal(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&)", referenced from:
      std::__1::vector<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> >, std::__1::allocator<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > > > boost::program_options::to_internal<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > >(std::__1::vector<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> >, std::__1::allocator<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > > > const&) in cc-6QQcwm.o
  "boost::program_options::invalid_option_value::invalid_option_value(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&)", referenced from:
      void boost::program_options::validate<int, char>(boost::any&, std::__1::vector<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> >, std::__1::allocator<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > > > const&, int*, long) in cc-6QQcwm.o
  "boost::program_options::validation_error::validation_error(boost::program_options::validation_error::kind_t, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&)", referenced from:
      std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const& boost::program_options::validators::get_single_string<char>(std::__1::vector<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> >, std::__1::allocator<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > > > const&, bool) in cc-6QQcwm.o
  "boost::program_options::value_semantic_codecvt_helper<char>::parse(boost::any&, std::__1::vector<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> >, std::__1::allocator<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > > > const&, bool) const", referenced from:
      vtable for boost::program_options::typed_value<int, char> in cc-6QQcwm.o
ld: symbol(s) not found for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)

The same code compiled/linked with g++4.7 installed with MacPorts:

$ g++-mp-4.7 -std=c++0x -I/opt/local/include -L/opt/local/lib -lboost_program_options -o first first.cpp

... works fine. As does using clang without libc++:

clang++ -std=c++0x -I/opt/local/include -L/opt/local/lib -lboost_program_options first.cpp -o first

What's wrong? Why does boost::program_options and libc++ not work together?

Solution

You need to rebuild boost using clang++ -stdlib=libc++.

libc++ is not binary compatible with gcc's libstdc++ (except for some low level stuff such as operator new). For example the std::string in gcc's libstdc++ is refcounted, whereas in libc++ it uses the "short string optimization". If you were to accidentally mix these two strings in the same program (and mistake them for the same data structure), you would inevitably get a run time crash.

This accident is exactly what has occurred in your case.

In order to turn this run time crash into a link time error, libc++ uses a C++11 language feature called inline namespace to change the ABI of std::string without impacting the API of std::string. That is, to you std::string looks the same. But to the linker, std::string is being mangled as if it is in namespace std::__1. Thus the linker knows that std::basic_string and std::__1::basic_string are two different data structures (the former coming from gcc's libstdc++ and the latter coming from libc++).

Comments

[Dmitry Isaev]

P.S. How to recompile boost for clang / libc++: stackoverflow.com/questions/8486077/…

[Shital Shah]

A quick question. If I was doing my_stdlib_str.c_str() then I would eventually get set fault, right? I mean in that case above linker trick would be by passed?

[Howard Hinnant]

Correct. The inline namespace trick isn't bullet proof. It has now been enough years that I suspect dual std::libs on macOS is no longer a problem.