That's some really interesting use cases :D Yes especially when accuracy is important I can imagine that the analytic solution is more efficient. Complex step diff blew my mind first time I heard about it, it seemed too good to be true :O
@cantanima you are bringing a lot of good points to the table :) Most of this is probably outside my abilities but implementing a few of them at least could be an interesting experience. Cutting-edge performence isn't really any ambition I have but something that's faster than doing it by hand is a win in my eyes. I have discovered something very exciting though! It seems BinaryBuilder has support for symengine so if I can get it to work with Nimterop there's a chance I won't have to battle against dependecy hell... And then we could have a working, fast library that people can use. And I can fiddle with my own pure Nim implementation in my spare time :-)
