Adrian had a terrible issue with floating point errors in .NET – he discovered afterwards that C++ managed code uses single precision arithmetic everywhere (!), but in the end he managed to force double precision.