[Boost-users] file_descriptor_sink vs. pclose

I've been having difficulty using iostreams::file_descriptor_sink with popen in Linux. The problem is that popen() requires a matching pclose() rather than a close(). Now, file_descriptor_sink has a close_on_exit flag that, when set to false, supposedly prevents it from calling close() on the underlying file descriptor, but it calls close() anyway. Is this a bug in boost or am I doing it wrong? As a workaround I can call flush () on the iostream and then pclose() on the filehandle, but it seems like it would be better to actually close the iostream. I've tested the example below using boost 1.39 and using 1.43 from subversion. The result in both cases is: pclose returned error (test A): Bad file descriptor pclose returned error (test B): Bad file descriptor Hello Hello ------------------------- #include <stdexcept> #include <iostream> #include #include #include #include #include typedef boost::iostreams::streamboost::iostreams::file_descriptor_sink boost_stream; int main() { bool close_on_exit = false; FILE *pout = popen("cat", "w"); { boost_stream bs(fileno(pout), close_on_exit); bs << "Hello\n"; // destructor calls close(fileno(pout)) } if(pclose(pout)) { char *err = strerror(errno); std::cerr << "pclose returned error (test A): " << err << std::endl; } pout = popen("cat", "w"); boost_stream bs(fileno(pout), close_on_exit); bs << "Hello\n"; // bs.close() calls close(fileno(pout)) // without the bs.close(), there would be no pclose() error bs.close(); // If bs.close() is not used, bs.flush() can be used to flush data to // pipe before calling pclose(). //bs.flush(); if(pclose(pout)) { char *err = strerror(errno); std::cerr << "pclose returned error (test B): " << err << std::endl; } }