ANN: Decimal.py 0.0.0

[Tim Peters]

[Niki Spahiev]
> Why not long instead of tuple?

[Aahz]
> ...
> Basically, the problem with longs is that you're still using binary
> arithmetic, not decimal arithmetic.  That's fine for addition and
> subtraction, but doesn't work as well for multiplication and division.

Na, it works fine, and binary longs are *especially* more efficient for
multiplication and division on binary HW.  Scaling by powers of 10 (a very
special case of * and /) is much quicker with decimal sequences, though, and
that's required "at the end" of floating operations (to normalize and round
the result to the destination precision).

> In addition, sticking to decimal arithmetic all the way guarantees
> that you're O(N) on converting to and from the internal representation,
> whereas it's O(N^2) if the internal format is binary.

That's the main point, IMO.  string<->FixedPoint conversion time swamps
computation time for most uses of FixedPoint, because binary<->decimal
conversion is expensive, and especially so once you go beyond the range of a
native C long.

I'll add that I encouraged Aahz to implement a sequence-based prototype in
Python, so that if this pans out, the algorithms can be migrated easily to C
for speed (functions slinging tuples of small ints in Python will be easy to
recast as C functions slinging pointer-to-small-int+length pairs).