lexical_cast crashes on certain conditions for floating point types

[alday.boost.trac@…]

In the presence of BOOST_LCAST_NO_COMPILE_TIME_PRECISION, there is no specialization of lcast_src_length<Source> for floating types, so lexical_converter_impl will allocate only 2 bytes and then try to do a sprintf, making the program crash.

While the specialization for integral types checks for the presence of BOOST_NO_LIMITS_COMPILE_TIME_CONSTANTS, and ifdefs an arbitrary value or one calculated via std::numeric_limits, the specialization for floats gets wiped out completely in the presence of BOOST_LCAST_NO_COMPILE_TIME_PRECISION.

Could you please do for floats what you do for integrals? I attach a patch that assigns an (arbitrary) max length of 156 to the representation of floats.

Basically, instead of

#ifndef BOOST_LCAST_NO_COMPILE_TIME_PRECISION
template<class Source>
        struct lcast_src_length<Source, BOOST_DEDUCED_TYPENAME boost::enable_if<boost::is_float<Source> >::type>
        {
...
            BOOST_STATIC_CONSTANT(std::size_t, value =5 + lcast_precision<Source>::value + 6);
        };
#endif // #ifndef BOOST_LCAST_NO_COMPILE_TIME_PRECISION

I'd like to move the ifndef inside, as in

 template<class Source>
        struct lcast_src_length<Source, BOOST_DEDUCED_TYPENAME boost::enable_if<boost::is_float<Source> >::type>
        {
#ifndef BOOST_LCAST_NO_COMPILE_TIME_PRECISION
...
            BOOST_STATIC_CONSTANT(std::size_t, value =5 + lcast_precision<Source>::value + 6);
#else
            BOOST_STATIC_CONSTANT(std::size_t, value = 156);
#ifndef BOOST_LCAST_NO_COMPILE_TIME_PRECISION

The easiest way to reproduce it is:

#include <string>
#define BOOST_LCAST_NO_COMPILE_TIME_PRECISION
#include <boost/lexical_cast.hpp>
int main() {
        boost::lexical_cast<std::string>(2.12345);
}

[Antony Polukhin]

Fixed in ​develop branch.

Will merge to the master branch as soon as the tests will cycle through.

Great thanks for providing patch!