Comment on PEP-0238
Edward Jason Riedy
ejr at cs.berkeley.edu
Thu Jul 12 13:07:50 EDT 2001
And Greg Ewing writes:
-
- It seems to me that, if you want adjustable precision at
- all, you want to adjust it in a very fine-grained way
- based on the kind of data you're using and what you're
- doing with it.
For _fixed-point_ arithmetic, this is often true. Your
monetary example is best served by an extremely flexible
arithmetic, as the laws governing the calculations are
complex. The Euro docs are a good, recent example.
- So the notion of precision being part of some sort of
- context that you establish and then go off and do your
- calculations seems fundamentally wrong-headed.
For many geometric and scientific codes using floating-point,
this is exactly what you want. You bump the precision up
once, or you multiply it after trying to get a good-enough
answer. The precision changes are always establishing a
working precision related to some desired output precision.
You change the precision rarely.
If you know of floating-point codes that mix desired
precisions frequently and intentionally, I'd really like to
hear of them.
- I don't know what the best solution to this is. I feel
- it lies somewhere in the direction of using specialised
- data types which know their own precision and also
- what precisions are needed for various combinations
- of them.
I don't have a copy of Martin Fowler's Analysis Patterns
handy, but I believe he addresses some of these issues.
There are a few articles that might relate:
http://martinfowler.com/articles.html
Jason
--
More information about the Python-list
mailing list