[conversion] Using Lexical_cast to convert string to float
In writing some unit tests to convert strings to floats I've come across a puzzling situation. I was trying to test that a std::bad_cast was thrown if a number was too large to fit into a float. The error wasn't thrown but the float was given 1.#INF000. The work around I used was to cast to a double and then do a numeric_cast on the result. This provided an exception for values larger than could fit in a float. Lexical_cast on a double though works as I would have expected. If a value is too large to fit into a double then a std::bad_cast is called. float test = boost::lexical_cast<float>("3.4e050"); //test will have the value of 1.#INF000 float test = boost::numeric_cast<float>( boost::lexical_cast<double>("3.4e50"); //std::bad_cast is thrown double test = boost::lexical_cast<double>("3.4e350"); //std::bad_cast is thrown Is the lexical_cast for float behaving correctly? Ryan P.S. This code is using boost 1.38.0
AMDG Ryan McConnehey wrote:
In writing some unit tests to convert strings to floats I've come across a puzzling situation. I was trying to test that a std::bad_cast was thrown if a number was too large to fit into a float. The error wasn't thrown but the float was given 1.#INF000. The work around I used was to cast to a double and then do a numeric_cast on the result. This provided an exception for values larger than could fit in a float. Lexical_cast on a double though works as I would have expected. If a value is too large to fit into a double then a std::bad_cast is called.
float test = boost::lexical_cast<float>("3.4e050"); //test will have the value of 1.#INF000 float test = boost::numeric_cast<float>( boost::lexical_cast<double>("3.4e50"); //std::bad_cast is thrown double test = boost::lexical_cast<double>("3.4e350"); //std::bad_cast is thrown
Is the lexical_cast for float behaving correctly?
The problem is in the standard library. (Tested with msvc 9.0 express--gcc 4.3.0 behaves sensibly). #include <iostream> #include <sstream> int main() { { std::istringstream ss("3.4e050"); float val; ss >> val; std::cout << std::boolalpha << "ss: " << static_cast<bool>(ss) << ", val: " << val << std::endl; } { std::istringstream ss("3.4e350"); double val; ss >> val; std::cout << std::boolalpha << "ss: " << static_cast<bool>(ss) << ", val: " << val << std::endl; } } ss: true, val: 1.#INF ss: false, val: -9.25596e+061 In Christ, Steven Watanabe
The problem is in the standard library. (Tested with msvc 9.0 express--gcc 4.3.0 behaves sensibly). So, Microsoft's implementation is off? It wouldn't be that surprising
Steven Watanabe wrote: though. :-)
ss: true, val: 1.#INF ss: false, val: -9.25596e+061 It looks like the "3.4e350" gets converted to a valid value (for doubles) and the only way to determine if it was out of range is to do a cast on the stream itself. Is this what lexical_cast does internally?
Ryan
AMDG Ryan McConnehey wrote:
The problem is in the standard library. (Tested with msvc 9.0 express--gcc 4.3.0 behaves sensibly). So, Microsoft's implementation is off? It wouldn't be that surprising
Steven Watanabe wrote: though. :-)
ss: true, val: 1.#INF ss: false, val: -9.25596e+061 It looks like the "3.4e350" gets converted to a valid value (for doubles) and the only way to determine if it was out of range is to do a cast on the stream itself.
There are no guarantees about what val holds.
Is this what lexical_cast does internally?
Yes. In Christ, Steven Watanabe
participants (2)
-
Ryan McConnehey -
Steven Watanabe