On 9/14/05, Joel Eidsath <jeidsath@gmail.com> wrote:
What is important is what happens when you try to implement sqrt for boost::rational numbers. Or sin. Or log. Or a host of other functions. For example, what sort of answer would you get for sin(60 degrees) * sqrt(3) / 3? (The exact answer is 1/2.)
(i am worried that this is just troll bait, but assuming it is a honest question) my guess would be "something closer to 1/2 than calculating with floating point arithmetic using the same number of bits (or maybe even the same number of CPU cycles)" in case of boost::rational, i doubt it will ever have sin() -- or, more precisely: the current boost::rational tries to do two things at the same time. being a fixed precision type (where the name is analogous to float, referring to the implementation), and that may have sin(), and everything else; and being an unlimited precision type (where the name refers to the mathematical concept, and accidentally the current implementation method), which is unlikely to ever have a sin() implemented br, andras