[Python-Dev] Re: Decimal data type issues

[Jewett, Jim J]

Paul Moore:
> Decimal(1.1, significant_digits=34) is unambiguous, 
> but confusing.

What would the preferred solution be?

Requiring the user to explicitly type zeros so as to 
have 34 digits?

Implicitly doing a Decimal(str(1.1)) and then adding
32 zeros to the internal representation?  (Note: in
at least my environment, str(1.1) = '1.1'; it is only
repr that adds the extra digits.)

Living with the fact that it will have some confusing
non-zero digits?

Facundo Batista:
> ... what is "exact conversion" to you:

>  1.1 -> "1.1"
>  1.1 -> "1.1000000000000001"
>  1.1 -> "1.100000000000000088817841970012523233890533447265625"

machine precision of floats, whatever that happens to be.  
That implementation-dependent part is one reason to use a
longer name, like exact_float.

But if I say 

  Decimal(1.1, precision=4)
  Decimal(1.1, positions=3)

Those are unambiguous.  If the context is 43, then later 
calculations will effectively fill in the extra digits;
I have just said to fill them in with zeros, rather than 
whatever the float approximations were.


Ka-Ping Yee:
>> I never mentioned positions, and i don't see any reason 
>> why it is necessary to isolate these two ways of specifying 
>> precision. 

Facundo Batista:
> Because they are two totally different concepts. And they're 
> not two ways of specifying precision.

Sure they are; the reason to specify positions is that the
underlying data wasn't really floating point -- it was an
integer which by convention is written in a larger unit.
Example with money:

	Assume meat is 0.987USD/pound.  There are three
	price digits to multiply by the weight, but the
	final cost is rounded to an integer number of
	pennies.

	10 pounds cost 9.87, but
	1 pound costs 0.99

I don't want to say that my prices have three digits of 
precision, or I'll keep fractions of a penny.  I don't
want to say that they have only two, or I'll drop the
pennies on expensive items.  I want to say that the
precision varies depending on the actual price.

> What whould mean here "significant_digits=4"?

It means precision.  In many US classrooms, 
"precision" applies to the original measurement.  
Once you start doing calculations, it gets called 
"significant digits", as a reminder not to let your 
answer be more precise than your inputs.

Ka-Ping Yee:
>> Converting from float is not confusing when the 
>> precision is specified (either in terms of significant 
>> digits or decimal places).

> Don't confuse the terminology. Precision means only one
> thing, and it's specified in Cowlishaw work.

But for any *specific* value, specifying either the total 
number of valid digits (signficant figures, precision) 
or the number of fractional digits (position) is enough
to determine both.

> "number" can be string, int, etc., but NOT float.
> The issue with rounding at a defined position has 
> nothing to do with context.

I assume that a new Decimal would normally be created
with as much precision as the context would need for 
calculations.  By passing a context/precision/position,
the user is saying "yeah, but this measurement wasn't
that precise in the first place.  Use zeros for the
rest, no matter what this number claims."

-jJ