Just for the record, I've written a userland arbitrary precision arithmetic library, that provides a BigDecimal class which should be pretty similar to what you're aiming to do:
https://github.com/brick/math
This library uses the existing GMP or BCMath extensions when available, but also works without them, using a (slower, but fully functional) pure PHP implementation.
A new PHP extension that would provide native arbitrary precision numbers with an OO API would be welcome, IMO.
Maybe brick/math can help inspire your API?
Benjamin
Le 29 sept. 2018 à 01:05, Rudi Theunissen rtheunissen@php.net a écrit :
Hi everyone,
I've been working on adding arbitrary precision decimal support as an
alternative to *bcmath. *I have created an extension based on mpdecimal,
which is what Python 3's decimal module is also based on. I haven't
released or broadcast this project yet, because I wanted to discuss the API
and implementation with internals first.See: https://github.com/php-decimal/php-decimal
Any advice, commentary or objection is welcome. :)
Thank you,
Rudi Theunissen--0000000000001ef4ff0576f68080
Hi Benjamin,
Brick\Math looks awesome. I really like the static initializers and
descriptive method names. I recognize some of the patterns from OpenJDK's
BigDecimal source. :)
The major difference to me is scale vs precision, ie. number of significant
digits vs number of digits behind the decimal point. Not sure which is
better, just noticing the difference.
Your production is so good, by the way. I really appreciate it.
Rudi
On Tue, Oct 23, 2018 at 5:45 AM Benjamin Morel benjamin.morel@gmail.com
wrote:
Just for the record, I've written a userland arbitrary precision
arithmetic library, that provides a BigDecimal class which should be pretty
similar to what you're aiming to do:This library uses the existing GMP or BCMath extensions when available,
but also works without them, using a (slower, but fully functional) pure
PHP implementation.A new PHP extension that would provide native arbitrary precision numbers
with an OO API would be welcome, IMO.Maybe brick/math can help inspire your API?
Benjamin
Le 29 sept. 2018 à 01:05, Rudi Theunissen rtheunissen@php.net a écrit :
Hi everyone,
I've been working on adding arbitrary precision decimal support as an
alternative to *bcmath. *I have created an extension based on mpdecimal,
which is what Python 3's decimal module is also based on. I haven't
released or broadcast this project yet, because I wanted to discuss the API
and implementation with internals first.See: https://github.com/php-decimal/php-decimal
Any advice, commentary or objection is welcome. :)
Thank you,
Rudi Theunissen--0000000000001ef4ff0576f68080
I recognize some of the patterns from OpenJDK's BigDecimal source. :)
Indeed, Brick\Math was largely inspired by Java's implementation!
The major difference to me is scale vs precision, ie. number of significant
digits vs number of digits behind the decimal point. Not sure which is
better, just noticing the difference.
Brick\Math does not have a concept of precision or "significant digits".
It only cares about scale, and has unlimited precision.
Its aim is to always return an exact result, unless rounding is explicitly
requested.
The scale is automatically adjusted for common operations such as plus(),
minus() and multipliedBy().
For dividedBy(), you have to explicitly specify the requested scale of the
result and an optional rounding mode; if no rounding mode is provided, and
the result does not fit in this scale, you get an exception.
If you don't know the scale but do know that the division yields a number
with a finite scale, you can use the exactlyDividedBy(), which will either
return an exact result with the correct scale, or throw an exception.
This is the first difference that strikes me with your current
implementation:
0.1 / 7 == 0.01428571428571428571428571429
Because the result is an infinite repeating decimal, in my opinion, your
Decimal class should not allow such a division without explicitly
specifying a scale and a rounding mode.
In other words, I would expect an exception here.
To exactly represent the result of this division, another concept such as
Brick\Math's BigRational can be used instead.
Ben
Because the result is an infinite repeating decimal, in my opinion, your
Decimal class should not allow such a division without explicitly
specifying a scale and a rounding mode. In other words, I would expect an
exception here.
There is already an internal flag for inexact division, but is currently
ignored. If exact division is a requirement, I would rather a dedicated
type for that like Decimal\Exact or Decimal\Fraction.
Brick\Math does not have a concept of precision or "significant digits".
It only cares about scale, and has unlimited precision.
That's the main difference here. Arbitrary scale might be more intuitive
and practical than arbitrary precision - I honestly don't have an opinion
here. It would be interesting to compare some use cases and
interoperability with SQL DECIMAL, which I assume would be a common analog
for any PHP type.
On Sat, Oct 27, 2018 at 3:14 AM Benjamin Morel benjamin.morel@gmail.com
wrote:
I recognize some of the patterns from OpenJDK's BigDecimal source. :)
Indeed, Brick\Math was largely inspired by Java's implementation!
The major difference to me is scale vs precision, ie. number of
significant digits vs number of digits behind the decimal point. Not sure
which is better, just noticing the difference.Brick\Math does not have a concept of precision or "significant digits".
It only cares about scale, and has unlimited precision.Its aim is to always return an exact result, unless rounding is explicitly
requested.
The scale is automatically adjusted for common operations such as plus(),
minus() and multipliedBy().
For dividedBy(), you have to explicitly specify the requested scale of the
result and an optional rounding mode; if no rounding mode is provided, and
the result does not fit in this scale, you get an exception.
If you don't know the scale but do know that the division yields a number
with a finite scale, you can use the exactlyDividedBy(), which will either
return an exact result with the correct scale, or throw an exception.This is the first difference that strikes me with your current
implementation:
0.1 / 7 == 0.01428571428571428571428571429Because the result is an infinite repeating decimal, in my opinion, your
Decimal class should not allow such a division without explicitly
specifying a scale and a rounding mode.
In other words, I would expect an exception here.To exactly represent the result of this division, another concept such as
Brick\Math's BigRational can be used instead.Ben