rfc:support_object_type_in_bcmath

PHP RFC: Support object type in BCMath

Introduction

BCMath currently only supports procedural functionality and does not support object types. Object-based programming is mainstream these days, so it's a bit outdated from that perspective. Also, the GMP extension, which is likely to have similar needs to BCMath, already supports object types.

Of course, it is also possible to treat BCMath like an object if you devise some ideas in userland. However, things like operator overloading cannot be implemented in userland.

Proposal

This RFC proposes support for object types in BCMath. Please take a look at the usage example first.

use BCMath\Number;
 
$num = new Number('1');
$num2 = new Number('2');
$result = $num + $num2;
 
$result->value; // '3'
var_dump($num > $num2); // false

Operator overloading allows you to treat objects as if they were computing or comparing primitive values. Also, it supports not only calculations using operators, but also calculations using methods.

The following example is equivalent to the above example.

use BCMath\Number;
 
$num = new Number('1');
$num2 = new Number('2');
$result = $num->add($num2);
 
$result->value; // '3'
var_dump($num->gt($num2)); // false

One important thing to note first: existing global settings related to scale are not used by Number. Therefore, no matter what is set in the global settings, it has no effect on the behavior of Number.

stub

As explained in more detail in subsequent chapters, most methods have optional arguments for scale and rounding mode. However, there are some conditions for using powmod, and since it never use scale and rounding modes, I do not include them in the arguments.

namespace BCMath {
  final readonly class Number implements \Stringable {
    public string $value;
    public int $scale;
 
    public function __construct(string|int $num) {}
 
    public function add(Number|string|int $num, ?int $scale = null, int $roundingMode = PHP_ROUND_HALF_UP): Number {}
 
    public function sub(Number|string|int $num, ?int $scale = null, int $roundingMode = PHP_ROUND_HALF_UP): Number {}
 
    public function mul(Number|string|int $num, ?int $scale = null, int $roundingMode = PHP_ROUND_HALF_UP): Number {}
 
    public function div(Number|string|int $num, ?int $scale = null, int $roundingMode = PHP_ROUND_HALF_UP): Number {}
 
    public function mod(Number|string|int $num, ?int $scale = null, int $roundingMode = PHP_ROUND_HALF_UP): Number {}
 
    public function powmod(Number|string|int $exponent, Number|string|int $modulus): Number {}
 
    public function pow(Number|string|int $exponent, int $minScale, ?int $scale = null, int $roundingMode = PHP_ROUND_HALF_UP): Number {}
 
    public function sqrt(?int $scale = null, int $roundingMode = PHP_ROUND_HALF_UP): Number {}
 
    public function floor(): Number {}
 
    public function ceil(): Number {}
 
    public function round(int $precision = 0, int $mode = PHP_ROUND_HALF_UP): Number {}
 
    public function comp(Number|string|int $num, ?int $scale = null): int {}
 
    public function eq(Number|string|int $num, ?int $scale = null): bool {}
 
    public function gt(Number|string|int $num, ?int $scale = null): bool {}
 
    public function gte(Number|string|int $num, ?int $scale = null): bool {}
 
    public function lt(Number|string|int $num, ?int $scale = null): bool {}
 
    public function lte(Number|string|int $num, ?int $scale = null): bool {}
 
    public function format(?int $scale = null, int $roundingMode = PHP_ROUND_HALF_UP, string $decimalSeparator = '.', string $thousandsSeparator = ''): string {}
 
    public function __toString(): string {}
  }
}

namespace

Use namespace “BCMath” for this class according to the following RFC: https://wiki.php.net/rfc/namespaces_in_bundled_extensions

In the RFC example, the symbols use common names, so I follow suit and use the symbol “Number”.

Therefore, the fully qualified name of the class would be:

BCMath\Number;

It is immutable

There is a concept called “value object” that is often seen in domain-driven design, etc., and Number is exactly what should be treated as a value object. That is, the object must be immutable.

No matter what we do, the original object remains unchanged and always returns a new object.

final readonly class

BCMath\Number is a final readonly class. Since it is an immutable class, the property must be readonly.

If this class is made inheritable, and if there is a property defined by the user, it is difficult to know which operand property to pass, the left or right operand, to the object resulting from the calculation using operator overloading. Therefore, use the final modifier.

Since this is final, there is no difference between making the property read-only and making the class itself read-only. For ease of understanding, this RFC uses readonly classes.

final readonly class Number{}

Stringable

In addition to referencing the value property to obtain the value of Number, you can also obtain the value by casting to the string type.

use BCMath\Number;
 
$num = new Number('1');
var_dump((string) $num); // '1'

Constructor

The constructor signature is:

public function __construct(string|int $num) {}

Since there is no need to consider errors for integers in the range that can be represented by ints, it is reasonable to accept ints in the constructor.

The scale of the value is always implicitly calculated from the given $num.

See code example.

use BCMath\Number;
 
new Number('2'); // value is '2', scale is 0
new Number('0.12345'); // value is '0.12345', scale is 5
new Number('2.0000'); // value is '2.0000', scale is 4

Poperties

This class has properties “value” and “scale”.

namespace BCMath;
 
final readonly class Number
{
    public string $value;
    public int $scale;
}

Methods

In addition to calculations using operators, Number also supports calculations using methods. Basically, it corresponds to the bcXXX functions. See stub.

The bcXXX functions accept numbers to calculate as string, while Number accepts Number instance, string, and int.

Major difference from bcXXX functions

There are two major differences from bcXXX.

  • If do not specify a scale, bcXXX uses the global settings. BCMath\Number does not use global values and automatically calculates the scale.
  • If the calculation result does not fit within the scale, bcXXX always truncates it. BCMath\Number rounds the value according to the specified rounding mode (That is, calculate one extra digit and then round the value).

For example, the existing behavior of bcadd is:

bcadd('1.23', '2.111'); // Global settings are used. If set to 0, '3' is returned.
bcadd('1.23', '2.111', 1); // '3.3' is returned.

With Number it works like this:

use BCMath\Number;
 
$num = new Number('1.23');
$num2 = new Number('2.111');
// If scale is omitted, the larger scale of $num and $num2 is used.
// In this example, the scale of $num2 is larger, so the calculation is done with scale = 3.
$result = $num->add($num2); // value is '3.341', scale is 3.
 
$num = new Number('1.23');
$num2 = new Number('2.111');
$result = $num->add($num2, 10); // value is '3.3410000000', scale is 10.
 
$num = new Number('1.23');
$num2 = new Number('2.111');
$result = $num->add($num2, 1, PHP_ROUND_AWAY_FROM_ZERO); // value is '3.4', scale is 1.

If $num2 is not Number, the following behavior:

use BCMath\Number;
 
// $num2 is int
$num->add(4); // Equivalent to "$num->add(new Number('4'));"
 
// $num2 is string
$num->add('2.3355'); // Equivalent to "$num->add(new Number('2.3355'));"

When calculating, the value is always implicitly converted to the bc_num structure that Number has internally.

If omit the scale (i.e. specify it as null), the scale will be automatically determined by calculation. How the scale is determined is explained with an example later in the RFC.

Comparison method

These are equivalent to the following operators:

method operator
comp <=>
eq ==
gt >
gte >=
lt <
lte <=

However, if specify $scale as a method argument, the comparison will be performed using up to the specified scale, like the existing bccomp(). When comparing using operator overloads, it is purely comparing values.

format

This behaves similar to number_format(). However, the arguments are slightly different. See code example:

use BCMath\Number;
 
$num = new Number('123456.789');
 
$num->format(); // '123456.789'
$num->format(1); // '123456.8'
$num->format(1, PHP_ROUND_TOWARD_ZERO); // '123456.7'
$num->format(1, PHP_ROUND_TOWARD_ZERO '.', ','); // '123,456.7'
 
$num->format(2, PHP_ROUND_HALF_UP, ',', ' '); // '123 456,79' french notation

Operator overload

See the table below for supported operators.

type support
comparison yes
add yes
sub yes
mul yes
div yes
mod yes
pow yes
bit shift no
bit wise no

Calculations with the operator behave as if the corresponding method's optional arguments were not specified. That is, scale is always specified as null and roundingMode is always specified as PHP_ROUND_HALF_UP.

Therefore, calculations such as the following are allowed:

use BCMath\Number;
 
$num = new Number('1.23');
$result = $num + 2;
$result->value; // '3.23'
$result->scale; // 2
 
$num = new Number('1.23');
$result = $num + '1.23456';
$result->value; // '2.46456'
$result->scale; // 5

Increment and decrement

Incrementing and decrementing Number behaves similarly to GMP objects.

use BCMath\Number;
 
$num = new Number('1.23');
$numA = $num;
 
$num++; // Here, $num will change to a new object.
 
$num->value; // '2.23'
$numA->value; // '1.23'

Exception

There are several times when you should throw an exception, such as division by 0. This is the same as the existing bcXXX functions. Therefore, I will not prepare a new exception class specifically for Number.

Detailed examples

I present some examples to clarify the criteria for automatically determining the scale. The following is the behavior when scale is omitted.

For div, pow, and sqrt, the scale of the calculation result may be infinite. Therefore, these three calculations have the concept of “maximum expansion scale” of the scale. This is the number of digits to extend relative to the original scale of the left operand. This is the value used only if no scale is specified and cannot be changed from userland.

This RFC uses a “maximum expansion scale” of 10.

For calculations that use “maximum expansion scale”, the scale of the result is at least the scale of the left operand and at most the scale of the left operand + “maximum expansion scale”.

add

use BCMath\Number;
 
$num = new Number('1.23');
$num2 = new Number('2.000000');
$result = $num + $num2; // value is '3.230000', The larger scale of the two values is applied. (2 < 6, so 6 is used)

sub

use BCMath\Number;
 
$num = new Number('1.23');
$num2 = new Number('2.000000');
$result = $num - $num2; // value is '-0.770000', The larger scale of the two values is applied. (2 < 6, so 6 is used)

mul

use BCMath\Number;
 
$num = new Number('1.23');
$num2 = new Number('2.456');
$result = $num * $num2; // value is '3.02088', The resulting scale is the sum of the scales of the two values. (2 + 3 = 5)
 
$num = new Number('1.25');
$num2 = new Number('4.00');
$result = $num * $num2; // value is '5.0000', The resulting scale is the sum of the scales of the two values. (2 + 2 = 4)

div

use BCMath\Number;
 
// maximum expansion scale is 10
 
$num = new Number('1.23');
$num2 = new Number('3.333');
$result = $num / $num2; // value is '0.369036903690', The max scale is the sum of the dividend scale and maximum expansion scale. (2 + 10 = 12)
 
$num = new Number('1.25');
$num2 = new Number('5');
$result = $num / $num2; // value is '0.25', The result fits within the maximum scale, so an implicit scale of 2 is set.
 
$num = new Number('1.25000');
$num2 = new Number('5');
$result = $num / $num2; // value is '0.25000', The result fits within the maximum scale, so an implicit scale of 5 is set.
 
$num = new Number('1.25000');
$num2 = new Number('5.00');
$result = $num / $num2; // value is '0.25000', The result fits within the maximum scale, so an implicit scale of 5 is set.

mod

use BCMath\Number;
 
$num = new Number('6.234');
$num2 = new Number('1.23');
$result = $num % $num2; // value is '0.084', the larger scale of the two values is applied. (3 > 2, so 3 is used)

powmod

use BCMath\Number;
 
$num = new Number('4');
$exponent = new Number('5');
$modulus = new Number('3');
$result = $num->powmod($exponent, $modulus); // value is '1', The scale is always 0 because the result is always an integer.

pow

use BCMath\Number;
 
$num = new Number('1.23');
$exponent = new Number('3');
$result = $num ** $exponent; // value is '1.860867', The value of the left operand scale multiplied by exponent becomes the resulting scale. (2 * 3 = 6)
 
$num = new Number('1.23');
$exponent = new Number('0');
$result = $num ** $exponent; // Scale is always 0 because the 0th power is always 1.
 
$num = new Number('1.23');
$exponent = new Number('-3');
$result = $num ** $exponent; // value is '0.537383918356', The maximum scale is the sum of the left operand's scale and maximum expansion scale. (2 + 10 = 12)

sqrt

use BCMath\Number;
 
$num = new Number('1.23');
$result = $num->sqrt(); // value is '1.109053650641', The max scale is the sum of the $num scale and maximum expansion scale. (2 + 10 = 12)
 
$num = new Number('16.00');
$result = $num->sqrt(); // value is '4.00', The result fits within the maximum scale, so an implicit scale of 2 is set.

floor

use BCMath\Number;
 
$num = new Number('1.23');
$result = $num->floor(); // value is '1', The scale is always 0 because the result is always an integer.

ceil

use BCMath\Number;
 
$num = new Number('1.23');
$result = $num->ceil(); // value is '2', The scale is always 0 because the result is always an integer.

round

use BCMath\Number;
 
$num = new Number('1.23');
$result = $num->round(1); // value is '1.2', Implicitly sets the scale from the rounded value.

comparison methods

use BCMath\Number;
 
$num = new Number('1.23');
$num2 = new Number('1,23456');
 
$num->comp($num2); // -1, Same as '1.23' <=> '1.23456'
$num->comp($num2, 2); // 0, Same as '1.23' <=> '1.23'
 
$num->eq($num2); // false, Same as '1.23' == '1.23456'
$num->eq($num2, 2); // true, Same as '1.23' == '1.23'
 
$num->gt($num2); // false, Same as '1.23' > '1.23456'
$num->gt($num2, 2); // false, Same as '1.23' > '1.23'
 
$num->gte($num2); // false, Same as '1.23' >= '1.23456'
$num->gte($num2, 2); // true, Same as '1.23' >= '1.23'
 
$num->lt($num2); // true, Same as '1.23' < '1.23456'
$num->lt($num2, 2); // false, Same as '1.23' < '1.23'
 
$num->lte($num2); // true, Same as '1.23' <= '1.23456'
$num->lte($num2, 2); // true, Same as '1.23' <= '1.23'

About the initial value of rounding mode

BCMath\Number has two types of rounding: explicit rounding using round() and implicit rounding if the result does not fit within the scale during calculation. In the existing bcXXX functions, the value is always rounded down, which corresponds to the rounding mode PHP_ROUND_TOWARD_ZERO.

A dilemma arises here.

If the initial value of the rounding mode for calculations to match the behavior of the existing function (i.e., set PHP_ROUND_TOWARD_ZERO as the initial value), it will be inconsistent with the round() method.

On the other hand, focusing on consistency with the round() method will lead to inconsistency with existing functions.

This RFC takes a secondary vote on the mode of implicit rounding during calculations.

Note that PHP_ROUND_TOWARD_ZERO always truncates the value, so don't have to calculate an extra digit for rounding, and is better at that than PHP_ROUND_HALF_UP.

Backward Incompatible Changes

The class BCMath\Number is no longer available in userland.

Proposed PHP Version(s)

Next minor version (currently 8.4)

RFC Impact

To SAPIs

None.

To Existing Extensions

Only BCMath is affected.

To Opcache

None;

New Constants

None.

php.ini Defaults

None.

Open Issues

None.

Unaffected PHP Functionality

There is no effect on anything other than BCMath.

Future Scope

None.

Proposed Voting Choices

There is a yes/no choice whether to accept this RFC and requires a 2/3 majority vote to be accepted.

It will also hold a secondary vote on whether the initial value for implicit rounding should be PHP_ROUND_HALF_UP or PHP_ROUND_TOWARD_ZERO. The option that receives more than 50% of the votes will be selected. In other words, the candidate with the most votes will be selected.

If the numbers are exactly the same, use PHP_ROUND_HALF_UP.

Voting started on 2024-05-01 and will end on 2024-05-16 00:00 GMT.

Support object type in BCMath
Real name Yes No
alcaeus (alcaeus)  
alec (alec)  
crell (crell)  
devnexen (devnexen)  
ericmann (ericmann)  
galvao (galvao)  
gasolwu (gasolwu)  
girgias (girgias)  
hirokawa (hirokawa)  
jimw (jimw)  
kguest (kguest)  
kocsismate (kocsismate)  
mauricio (mauricio)  
mbeccati (mbeccati)  
nielsdos (nielsdos)  
petk (petk)  
ramsey (ramsey)  
saki (saki)  
sergey (sergey)  
timwolla (timwolla)  
weierophinney (weierophinney)  
Final result: 18 3
This poll has been closed.

Initial value for implicit rounding mode
Real name PHP_ROUND_HALF_UP PHP_ROUND_TOWARD_ZERO
alcaeus (alcaeus)  
derick (derick)  
devnexen (devnexen)  
ericmann (ericmann)  
galvao (galvao)  
gasolwu (gasolwu)  
girgias (girgias)  
hirokawa (hirokawa)  
jimw (jimw)  
kguest (kguest)  
mauricio (mauricio)  
mbeccati (mbeccati)  
ramsey (ramsey)  
saki (saki)  
sergey (sergey)  
theodorejb (theodorejb)  
weierophinney (weierophinney)  
Final result: 16 1
This poll has been closed.

Patches and Tests

Prototype: https://github.com/php/php-src/pull/13741

Not all features have been implemented yet.

Implementation

References

https://externals.io/message/122651 (Mailing list thread before creating RFC)

https://externals.io/message/122735 (Mailing list thread RFC Discussion)

Rejected Features

None.

rfc/support_object_type_in_bcmath.txt · Last modified: 2024/09/04 02:14 by saki