On Wednesday 14 October 2009 23:03:10 Matthias Schabel wrote:
I thought that the operators for std:;complex can't deduce the return type correctly for Boost.Units. Anyway "The effect of instantiating the template complex for any type other than float, double or long double is unspecified." (26.2)
Right, as usual... Maybe flesh out the more flexible complex type that we provided in our example or lobby the standards committee to make std::complex less stupid? Is there any good reason to have unspecified behavior for std::complex for UDTs?
Yes. This is a huge problem with extended-/finite-precision code where the type of a + b is different from the type of a or b. In this case, std::complex<10-bit type> + std::complex<10-bit type> leads to std::complex<11-bit type> and the problem gets much more complicated for multiplication and division. I work around it currently by specializing std::complex(!) in the std namespace (against all guidelines) for my fixed-point types. Any better solution would be greatly welcome. (For the mathematicians here, the problem is that std::complex<fixed-point type> does not form a ring, much less a field.) Regards, Ravi