On Tue, Sep 13, 2011 at 12:06 AM, Nick Coghlan <ncoghlan@gmail.com> wrote: ..
2. The 'math' module is not the place for this, a new, dedicated module is more appropriate. This is mainly due to the fact that the math module is focused primarily on binary floating point, while these algorithms should be neutral with regard to the specific numeric type involved.
While I agree that a new dedicated module is an appropriate place for this functionality, I am concerned that generic type-neutral algorithms may be sub-optimal for float-only calculations. I think at least at the start, the new module should follow the same design as the math module - namely provide numerically stable efficient algorithms for double precision floating point data. Other types should be supported by coercing to float. This means that support for long integers, decimal type, fractions, complex numbers will necessarily be limited. For example, you will not be able to compute average of 100-digit integers accurately even when the exact answer is an integer. Once the semantics of operations on general iterables yielding numbers are in place, the first optimization should probably be to special case operations on arrays of doubles. Proper support for decimal type, fractions, etc would probably need a separate project.