On Tue, May 22, 2012 at 9:41 PM, Vicente J. Botet Escriba <vicente.botet@wanadoo.fr> wrote:
Le 22/05/12 18:55, Wilfried Kirschenmann a écrit :
Dear chrono maintainer team,

Using the following example :

int main(int argc, char **argv)
{
       boost::chrono::process_real_cpu_clock::time_point start = boost::chrono::process_real_cpu_clock::now();

       for ( long i = 0; i < 1000; ++i )  std::sqrt( 123.456L ); // burn some time
       Sleep(10000); //idle some time

       boost::chrono::process_real_cpu_clock::time_point end = boost::chrono::process_real_cpu_clock::now();
       std::cout << end << std::endl;

        boost::chrono::process_real_cpu_clock::duration elapsed = (end - start);
       std::cout << "took real     : " << elapsed  << "nanoseconds\n";

       return 0;
}

I obtain negative times.

I run boost 1.49 win32 on Windows 7 x64 with Visual studio 2010.

Studying the behavior of the timer, I think that the error is one of these :
- process_real_cpu_clock::now() only stores the current time in a 32 bit integer instead of a 64 bit integer

I don't know from where are you getting this conclusion. process_real_cpu_clock is using boost::int_least64_t, as representation.

    typedef duration<boost::int_least64_t, nano> nanoseconds;    // at least 64 bits needed
    class BOOST_CHRONO_DECL process_real_cpu_clock {
    public:
        typedef nanoseconds                          duration;
        typedef duration::rep                        rep;
        typedef duration::period                     period;
        typedef chrono::time_point<process_real_cpu_clock>    time_point;
        BOOST_STATIC_CONSTEXPR bool is_steady =             true;

        static BOOST_CHRONO_INLINE time_point now() BOOST_NOEXCEPT;
#if !defined BOOST_CHRONO_DONT_PROVIDE_HYBRID_ERROR_HANDLING
        static BOOST_CHRONO_INLINE time_point now(system::error_code & ec );
#endif

    };


From the fact that it behaves as if there were a 32bit overflow. Which is in fact the case : https://svn.boost.org/trac/boost/ticket/6361
The problem is fixed in 1.50 and I no longer get negative times.
 
Wilfried K.