numeric_cast throws on exactly representable conversion

[Simeon Maxein <smaxein@…>]

If a double or float representing the smallest possible value of int64_t is cast to int64_t using numeric_cast, an exception is thrown saying that a negative overflow has occured.

I would expect the cast to work, since the value is exactly representable in the target type. Since the absolute value is a power of two, there should not be any error in the floating point representation.

This is also reproducable in trunk, according to VeXocide. I found the problem in Boost 1.47.0 on Windows, using MinGW 4.5.2.

[Simeon Maxein <smaxein@…>]

Small test program

[wellnhofer@…]

This works for me with Boost 1.58 on Ubuntu, but ldrexp(1, 63) doesn't throw. See ticket #12252.