id	summary	reporter	owner	description	type	status	milestone	component	version	severity	resolution	keywords	cc
10639	lexical_cast<double>(string) wrong in C++11	richard.kreckel@…	Antony Polukhin	"It's a well-known fact that printing a double precision floating-point number to 17 decimal digits is enough to recover the original double precision number (c.f. Theorem 15 in David Goldberg's ""What Every Computer Scientist Should Know About Floating-Point Arithmetic"").

This useful property has been broken somewhere between Boost 1.40 and Boost 1.55 if the program is compiled with -std=c++11 or -std=c++14 (GCC or CLang compiler switches).

Here is a test program which should print the same number three times:
{{{
#include <iostream>
#include <string>
#include <boost/lexical_cast.hpp>
#include <boost/math/special_functions/next.hpp>
#include <boost/format.hpp>

int main()
{
    double x = 1.0316536643319239e-255;
    std::printf(""x == %.17g\n"", x);

    std::string s = (boost::format(""%.17g"") % x).str();
    std::printf(""s == %s\n"", s.c_str());

    double y = boost::lexical_cast<double>(s);  // ERROR
    std::printf(""y == %.17g\n"", y);

    if (x != y) {
        std::cout << ""x and y are "" 
                  << boost::math::float_distance(x, y) 
                  << "" ULP apart"" << std::endl;
        return 1;
    }

    return 0;
}
}}}

However, with Boost 1.55 or 1.56 it prints:
{{{
x == 1.0316536643319239e-255
s == 1.0316536643319239e-255
y == 1.031653664331924e-255
x and y are 1 ULP apart
}}}

I'm testing this on an AMD64 machine running Linux with both G++ 4.9.1 and CLang 3.5.0. In both cases, the compiler switch -std=c++11 or -std=c++14 has to be set in order to trigger the problem."	Bugs	closed	Boost 1.58.0	lexical_cast	Boost 1.56.0	Problem	fixed		pbristow@…