Re: [Python-ideas] make decimal the default non-integer instead of float?

[Gregory P. Smith]

-cc: python-dev

+cc: python-ideas

On Sat, Sep 29, 2012 at 11:39 AM, Chris Angelico <ros...@gmail.com> wrote:

On Sun, Sep 30, 2012 at 4:26 AM, Brett Cannon <br...@python.org> wrote:
> Does this mean we want to re-open the discussion about decimal constants?
> Last time this came up I think we decided that we wanted to wait for
> cdecimal (which is obviously here) and work out how to handle contexts, the
> syntax, etc.

Just to throw a crazy idea out: How bad a change would it be to make
decimal actually the default?

(Caveat: I've not worked with decimal/cdecimal to any real extent and
don't know its limitations etc.)

Painful for existing code, unittests and extension modules.  Definitely python-ideas territory (thread moved there with an appropriate subject).

I'm not surprised at all that a decimal type can be "fast" in an interpreted language due to the already dominant interpreter overhead.

I wish all spreadsheets had used decimals from day one rather than binary floating point (blame Lotus?). Think of the trouble that would have saved the world.

-gps

[Calvin Spealman]

I like the idea a lot, but I recognize it will get a lot pushback. I
think learning Integer -> Decimal -> Float is a lot more natural than
learning Integer -> Float -> Decimal. The Float type represents a
specific hardware accelleration with data-loss tradeoffs, and the use
should be explicit. I think that as someone learns, the limitations of
Decimals will make a lot more sense than those of Floats.

+1

> _______________________________________________
> Python-ideas mailing list
> Python...@python.org
> http://mail.python.org/mailman/listinfo/python-ideas
>



--
Read my blog! I depend on your acceptance of my opinion! I am interesting!
http://techblog.ironfroggy.com/
Follow me if you're into that sort of thing: http://www.twitter.com/ironfroggy
_______________________________________________
Python-ideas mailing list
Python...@python.org
http://mail.python.org/mailman/listinfo/python-ideas

[Guido van Rossum]

On Sat, Sep 29, 2012 at 1:34 PM, Calvin Spealman <ironf...@gmail.com> wrote:
> I like the idea a lot, but I recognize it will get a lot pushback. I
> think learning Integer -> Decimal -> Float is a lot more natural than
> learning Integer -> Float -> Decimal. The Float type represents a
> specific hardware accelleration with data-loss tradeoffs, and the use
> should be explicit. I think that as someone learns, the limitations of
> Decimals will make a lot more sense than those of Floats.

Hm. Remember decimals have data loss too: they can't represent 1/3 any
more accurately than floats can, and like floats they are limited to a
certain number of digits after which they begin dropping precision
even if the result *can* be represented exactly. It's just that they
can represent 1/5 exactly, which happens to be culturally important to
humans, and that the number of digits at which loss of precision
happens is configurable. (And the API to configure it may actually
make it more complex to learn.)

--
--Guido van Rossum (python.org/~guido)

[Carl Johnson]

It's true that both Decimal and float lose precision, but all school children (in America at least) are taught how to reason about the loss of precision in decimals. Generally speaking, only computer science majors ever learn to think about the loss of precision in floats. So, I think that for Python 4000, a Decimal default might be sensible, assuming performance was acceptable.

[Stefan Behnel]

Guido van Rossum, 29.09.2012 23:06:

> On Sat, Sep 29, 2012 at 1:34 PM, Calvin Spealman wrote:
>> I like the idea a lot, but I recognize it will get a lot pushback. I
>> think learning Integer -> Decimal -> Float is a lot more natural than
>> learning Integer -> Float -> Decimal. The Float type represents a
>> specific hardware accelleration with data-loss tradeoffs, and the use
>> should be explicit. I think that as someone learns, the limitations of
>> Decimals will make a lot more sense than those of Floats.
>
> Hm. Remember decimals have data loss too: they can't represent 1/3 any
> more accurately than floats can

That would be "fractions" territory. Given that all three have their own
area where they shine and a large enough area where they really don't, I
can't see a strong enough argument for making any of the three "the default".

Also note that current float is a very C friendly thing, whereas the other
two are far from it.

Stefan


anecdotal PS: I recently got caught in a discussion about the impressive
ugliness of decimals in Java, the unsuitability of float in a financial
context and the lack of a better alternative in the programming languages
that most people use in that area. I got a couple of people surprised by
the fact that Python has fractions right in its standard library.

[Serhiy Storchaka]

Instructive story about fractions:
http://python-history.blogspot.com/2009/03/problem-with-integer-division.html
.

Let's not fall into the same trap.

[Stefan Behnel]

Serhiy Storchaka, 30.09.2012 18:35:

> Instructive story about fractions:
> http://python-history.blogspot.com/2009/03/problem-with-integer-division.html

Sorry - I don't get it. Instructive in what way?

Stefan