rfc:operator_functions

PHP RFC: Operator functions

Background

Over time, PHP has gradually acquired more facilities that enable functional programming. A frequent pattern in functional programming is the use of higher-order functions (functions that take or return functions, e.g. array_map()). With higher-order functions, small operations can be composed together to make more complex ones (or even an entire program). PHP's standard library comprises a large set of operations which can potentially be used with higher-order functions. However, PHP's most fundamental set of operations, the operators, are not functions and therefore cannot be directly used with higher-order functions. This means that either wrapper functions for operators must be written by PHP users, or otherwise-generic code which operates on functions must have specific code paths for the operators.

Proposal

Basics

This RFC proposes that, for each of PHP's built-in operators that operate solely on expressions (no assignment operators), a corresponding function with the same symbol would be added to the PHP standard library in the root namespace (\). So, for the + operator there would be a corresponding '+' function, for the === operator a corresponding '===' function, and so on.

These functions could then be passed as arguments to higher-order functions:

// Convert numbers from strings to ints or floats as appropriate
$numbers = array_map('+', $_GET["numbers"]);
// Adds the numbers in $terms together (equivalent to array_sum())
$sum = array_reduce($terms, '+', 0);
// Multiplies the numbers in $terms together (equivalent to array_product())
$product = array_reduce($terms, '*', 1);
// Union of the arrays (*NOT* the same as array_merge())
$merged = array_reduce($arrays, '+', []);
class Data
{
    public $values;
    public function __construct(array $values) {
        $this->values = $values;
    }
    public function sort(callable $function) {
        usort($this->values, $function);
    }
}
 
$data = new Data([1, 22, 3]);
 
// Sorts using standard comparison rules
// (equivalent to sort(), but now doesn't need its own code-path)
$data->sort('<=>');
 
// Sorts using string comparison rules
$data->sort('strcmp');

This is particularly useful when combined with partial application and function composition (primitives PHP currently does not yet have built-in, but can be written manually):

// Double all the numbers in the array
$doubled = array_map(partialApply('*', 2), $terms);
// Select only the positive numbers
$positiveSubset = array_filter($numbers, partialApply('<', 0));

An example working partial application implementation would be:

partialApply.php
function partialApply(callable $c, ...$args) {
    return function (...$args2) use ($c, $args) {
        return $c(...$args, ...$args2);
    };
}

Detail

Because operators have symbols that aren't valid identifiers in PHP source code (e.g. +), these functions cannot be called directly in the same manner as a normal function (i.e. +(1, 2)). However, a function with any name can be called by specifying its name as a string (e.g. '+'(1, 1) or "+"(1, 1)). Therefore, you could technically use these functions in place of operators:

// An excessively verbose version of sqrt(($x1 - $x2) ** 2 + ($y1 - $y2) ** 2)
$distance = sqrt('+'('**'('-'($x1, $x2), 2), '**'('-'($y1, $y2), 2)));

Of course, there is no practical reason to do this. The usefulness of this proposal is in composing operators with higher-order functions.

The table below lists the new functions that would be added to the root namespace (\). Each is named the same as its corresponding operator, including any aliases (for the sake of consistency).

Function signature Corresponding operation Notes
'+'($a[, $b]) +$a, $a + $b
'-'($a[, $b]) -$a, $a - $b
'*'($a, $b) $a * $b
'/'($a, $b) $a / $b
'%'($a, $b) $a % $b
'**'($a, $b) $a ** $b
'&'($a, $b) $a & $b
'|'($a, $b) $a | $b
'^'($a, $b) $a ^ $b
'~'($a) ~$a
'<<'($a, $b) $a << $b
'>>'($a, $b) $a >> $b
'=='($a, $b) $a == $b
'==='($a, $b) $a === $b
'!='($a, $b) $a != $b
'<>'($a, $b) $a <> $b
'!=='($a, $b) $a !== $b
'<'($a, $b) $a < $b
'>'($a, $b) $a > $b
'<='($a, $b) $a <= $b
'>='($a, $b) $a >= $b
'<=>'($a, $b) $a <=> $b
'&&'($a, $b) $a && $b Can't fully short-circuit.
'and'($a, $b) $a and $b Can't fully short-circuit.
'||'($a, $b) $a || $b Can't fully short-circuit.
'or'($a, $b) $a or $b Can't fully short-circuit.
'xor'($a, $b) $a xor $b
'!'($a) !$a
'.'($a, $b) $a . $b

Since pow() already exists and behaves identically to how '**'() would, '**'() is simply an alias of it.

These functions do not perform any extra type checking on their arguments beyond that normally performed by the operators they correspond to.

Missing operators

The table above (like the patch) currently contains all the operators in the Operators section of the PHP Manual, minus instanceof, `backticks` and the assignment operators. Whether these should have functions too is a matter to debate; instanceof doesn't take arbitrary expressions and already has a functional counterpart (is_a). As for the assignment operators, references mean they could be done, but from a functional programming perspective they have limited utility.

PHP also has some other constructs that could be classed as operators but aren't considered such by the manual. A (possibly non-exhaustive) list is:

  • ?? (Can't isset(). Can't short-circuit.)
  • ?: (Could be '?:'($a, $b[, $c]) and map to $a ?: $b or $a ? $b : $c depending on parameter count. Can't short-circuit.)
  • @ (Could not be made a function without changing it to act on a callable.)
  • (int), (string) etc. (Note intval() etc already exist.)
  • clone
  • print (This always returns 1, so we might as well make echo a function too even though it's a statement.)
  • -> (How do you distinguish between property lookup and method calls? Are identifiers replaced with strings?)
  • [] (Array indexing.)
  • () (Function invocation. call_user_func exists already.)
  • eval (Probably not a good rabbit hole to go down, this requires frowned-upon stack gymnastics due to affecting the current scope.)
  • include, require, include_once, require_once
  • yield (Like eval, would require dubious stack gymnastics. It is a control-flow expression, not merely manipulating values.)

Of these, ->, (), @ and eval are the most dubious.

Backward Incompatible Changes

All of these operator functions create no backwards-compatibility break, since they have names that cannot be used for userland functions, and thus they cannot conflict with function names in existing code (hypothetically this may not be true if using exotic extensions like runkit).

Proposed PHP Version(s)

This would go in the next PHP 7.x, most likely 7.3.

RFC Impact

To Existing Extensions

Because '**'() aliases it, pow() is moved out of ext/standard and into Zend. This is merely an organisational change and has no user impact.

To Opcache

The patch passes its test under OPcache.

Open Issues

See “Missing operators” section.

Unaffected PHP Functionality

The existing operators themselves behave the same as ever.

Being able to quote function names in function calls (e.g. '+'(1, 1)) is not a new idea introduced by this RFC, it has been possible since Uniform Variable Syntax in PHP 7.0.

Future Scope

Operator functions would fit well with built-in partial application and function composition. These could be added as functions, methods on \Closure, or both.

If built-in operators can have corresponding functions, then user functions could have corresponding operators in future, i.e. user-defined operators. This is possible in Haskell, for example, where new operators can be defined as functions.

Proposed Voting Choices

This is technically a standard library addition, so may only require a 50%+1 majority. It would be a straight Yes/No vote on whether to accept the RFC and merge the patch for PHP 7.3.

Patches and Tests

A complete patch for php-src, including test, can be found here: https://github.com/php/php-src/pull/2738

There may be some merit to adding this to the language specification, even though it otherwise doesn't cover built-in functions. There is no patch for this at present.

Implementation

After the project is implemented, this section should contain

  1. the version(s) it was merged to
  2. a link to the git commit(s)
  3. a link to the PHP manual entry for the feature
  4. a link to the language specification section (if any)

References

Rejected Features

Keep this updated with features that were discussed on the mail lists.

Changelog

  • v1.0.2 - add subsection discussing omissions
  • v1.0.1 - acknowledge why certain operators are excluded
  • v1.0 - first public non-draft version
rfc/operator_functions.txt · Last modified: 2017/09/22 13:28 by 127.0.0.1