[Python-Dev] Decimal data type issues

[Ka-Ping Yee]

On Thu, 15 Apr 2004, Batista, Facundo wrote:
> [Ka-Ping Yee]
>
> #- Aren't you going to need an optional argument anyway to let the
> #- user specify a context?
>
> What for?

So the user can specify how much precision the resulting Decimal value
should have.

> #- I assume the purpose of from_float is to prevent people from being
> #- confused about whether Decimal(1.1) produces exactly 1.1 or the machine
> #- value of the double-precision rounded float closest to 1.1.  Is that
> #- correct?
> #-
> #- If so, then the real exceptional case is *exact* conversion, not float
> #- conversion.  Decimal(1.1, precision=4) is unambiguous.  Therefore:
>
> Here I think you're confused. *Precision* is a parameter of the context, and
> determines in how many digits you will get the coefficient after an
> operation.
>
> The issue when constructing from float, is how many *positions* after the
> decimal point will be relevant when creating from a float.

I never mentioned positions, and i don't see any reason why it is
necessary to isolate these two ways of specifying precision.  Sometimes
it will be useful to specify a number of significant digits and other
times it will be useful to specify a number of digits after the decimal
point.  Both kinds of specification are useful as a context for arithmetic
operations and useful when constructing from any type including float.

So it would be nice if contexts were able to specify precision in either
fashion.

But that is a separate issue from the one i was trying to raise.  To
bring up this precision vs. position question is to miss my point.

Here is my point, again: Decimal(1.1, significant_digits=4) is unambiguous.

Converting from float is not confusing when the precision is specified
(either in terms of significant digits or decimal places).

Therefore, it is not necessary to make all float conversions inconvenient.
It is only necessary to make *exact* float conversions inconvenient.

Focusing on the true source of the problem also lets us use a more
descriptive name for this special case.

> #- That's what I'm suggesting: that you can supply a context object,
> #- or ask for the default context, when constructing any Decimal value.
>
> Do you want to supply a context that *will affect* the creation process, or
> just supply one that the Decimal will keep to use in other operations?

I want to supply a context for constructing the Decimal value.

It would not be persistent.  You would use some other method (presumably
something you have already conceived and named) for setting the Decimal
context for the entire thread.



-- ?!ng