====== PHP RFC: Short Closures ======
* Version: 0.2
* Date: 2015-05-01
* Author: Bob Weinand, bobwei9@hotmail.com
* Status: Declined / Withdrawn in favor of http://wiki.php.net/rfc/arrow_functions
* First Published at: http://wiki.php.net/rfc/short_closures
===== Introduction =====
Anonymous functions, also known as closures, allow the creation of functions which have no specified name. They are most useful as the value of callback parameters, but they have many other uses.
The current implementation of anonymous functions in PHP is quite verbose compared to other languages. That makes using anonymous functions be more difficult than it could be, as there is both more to type, and more importantly the current implementation makes it hard to read (and so maintain) code that uses anonymous functions.
A better syntax encourages functional code and partial applications (see the examples), which are a powerful tools people writing PHP code should be able to use as easily as they can be used elsewhere.
===== Proposal =====
This RFC proposes the introduction of the ~> operator to allow shorthand creation of anonymous functions to reduce the amount of 'boilerplate' needed to use them.
Current code:
function ($x) {
return $x * 2;
}
would be equivalent to the new syntax:
$x ~> $x * 2
Anonymous functions defined in this way will automatically ''use ()'' all of the (compiled) variables in the Closure body. See the 'Variable binding' section for more details.
==== Syntax ====
The syntax used to define a short hand anonymous function would be:
* Parameters. When the function has a single parameter the surrounding parentheses (aka round brackets) may be omitted. For functions with multiple parameters the parentheses are required.
* The new short closure operator ~>
* The body of the anonymous function. When the body of the function is a single expression the surrounding curly brackets and return keyword may be omitted. When the body of the function is not a single expression, the braces (and eventual return statement) are required.
I.e. all of the following would be equivalent:
$x ~> $x * 2
$x ~> { return $x * 2;}
($x) ~> $x * 2
($x) ~> { return $x * 2; }
Omitting the parentheses when the function has multiple parameters will result in a parse error:
$x, $y ~> {$x + $y} // Unexpected ','
($x, $y) ~> $x + $y // correct
Using the return keyword when braces have been omitted, will similarly give a parse error:
($x, $y) ~> return $x + $y; // Unexpected T_RETURN
($x, $y) ~> { return $x + $y; } // correct
In case of no parameters, an empty parenthesis pair is needed.
~> 2 * 3; // Unexpected T_TILDED_ARROW
() ~> 2 * 3; // correct, will return 6 when called
Concrete syntax is (~> is right associative with lowest possible precedence):
( parameter_list ) ~> expression
| ( parameter_list ) ~> { statements }
/* return by reference */
| &( parameter_list ) ~> expression
| &( parameter_list ) ~> { statements }
/* shorthand form for just one parameter */
| $variable ~> expression
| $variable ~> { statements }
When a bare expression is used as second parameter, its result will be the return value of the Closure.
Also, parameter_list does //not// include default values nor type hints. See also the 'Type Hints and Return Types' section at the bottom.
//Discussion Point: the { statements } syntax//
This RFC stance is that chained short Closures followed by a full Closure would look quite weird: ''$foo ~> $bar ~> function ($baz) use ($foo, $bar) { /* ... */ }''. Instead of a nicer ''$foo ~> $bar ~> $baz ~> { /* ... */ }''. Which is why they are supported. That syntax is **not** an invitation to randomly abuse it and use it in totally inappropriate places.
//Discussion Point: single parameter//
While it might appear not consistent, with any other number of parameters, a lot of languages having extra short Closures allow this. Also, Closures with just one parameter are relatively common, so this RFC author thinks it is worth supporting that.
==== Variable binding ====
The position of this RFC is that the shorthand syntax is to allow anonymous functions to be used as easily as possible. Therefore, rather than requiring individual variables be bound to the closure through the ''use ($x)'' syntax, instead all variables used in the body of the anonymous function will automatically be bound to the anonymous function closure from the defining scope.
The variable binding is always **by value**. There are no implicit references. If these are needed, the current syntax with ''use ()'' can be used.
For example:
$a = 1;
function foo(array $input, $b) {
$c = rand(0, 4);
return array_map($x ~> ($x * 2) + $b + $c, $input);
}
Variables $b and $c would be bound automatically to the anonymous function, and so be usable inside it. Variable $a is not in the scope of the function, and so is not bound, and so cannot be used inside the closure. e.g. this code will give an error:
$a = 1;
function foo(array $input, $b) {
// Notice: Undefined variable: a in %s on line %d
return array_map($x ~> ($x * 2) + $b + $a, $input);
}
If a user wants to avoid binding all variables automatically they can use the current syntax to define the anonymous function.
===== Examples =====
These examples cover some simple operations and show how the short-hand syntax is easier to read compared to the existing long-hand syntax.
==== Array sort with user function ====
Sort ''$array'' which contains objects which have a property named ''val''.
Current syntax:
usort($array,
function($a, $b) {
return $a->val <=> $b->val;
}
);
New syntax:
usort($array, ($a, $b) ~> $a->val <=> $b->val);
==== Extracting data from an array and summing it ====
Current syntax:
function sumEventScores($events, $scores) {
$types = array_map(
function($event) {
return $event['type'];
},
$events
);
return array_reduce(
$types,
function($sum, $type) use ($scores) {
return $sum + $scores[$type];
}
);
}
New syntax:
function sumEventScores($events, $scores) {
$types = array_map($event ~> $event['type'], $events);
return array_reduce($types, ($sum, $type) ~> $sum + $scores[$type]);
}
The calling code for this function would be:
$events = array(
array(
'type' =>'CreateEvent',
'date' => '2015-05-01T16:19:33+00:00'
),
array(
'type' =>'PushEvent',
'date' => '2015-05-01T16:19:54+00:00'
),
//...
);
$scores = [
'PushEvent' => 5,
'CreateEvent' => 4,
'IssuesEvent' => 3,
'CommitCommentEvent' => 2
];
sumEventScores($events, $scores);
==== Lazy evaluation ====
It may be necessary to have code only evaluated under specific conditions, like debugging code:
function runDebug(callable $func) {
/* only run under debug situations, but don't let it interrupt program flow, just log it */
if (DEBUG) {
try {
$func();
} catch (Exception $e) { /*... */ }
}
}
$myFile = "/etc/passwd";
/* Old code */
runDebug(function() use ($myFile) { /* yeah, we have to use use ($myFile) here, which isn't really helpful in this context */
if (!file_exists($myFile)) {
throw new Exception("File $myFile does not exist...");
}
});
/* New code */
runDebug(() ~> {
if (!file_exists($myFile)) {
throw new Exception("File $myFile does not exist...");
}
});
/* still continue here, unlike an assert which would unwind the stack frame here ... */
==== Partial application ====
The shorthand syntax makes it easier to write functional code like a reducer by using the ability of shorthand anonymous functions to be chained together easily.
Current syntax:
function reduce(callable $fn) {
return function($initial) use ($fn) {
return function($input) use ($fn, $initial) {
$accumulator = $initial;
foreach ($input as $value) {
$accumulator = $fn($accumulator, $value);
}
return $accumulator;
};
};
}
New syntax:
function reduce(callable $fn) {
return $initial ~> $input ~> {
$accumulator = $initial;
foreach ($input as $value) {
$accumulator = $fn($accumulator, $value);
}
return $accumulator;
};
}
===== Symbol choice =====
The symbol ''~>'' was chosen as it is a mnemonic device to help programmers understand that the variable is being brought to a function. It is also unambiguous as it has not been used elsewhere in PHP.
Currently Hack has implemented shorthand anonymous functions using the ''%%==>%%'' symbol to define them. The position of this RFC is that the ''%%==>%%'' symbol is too similar to the ''%%=>%%'' (double arrow) sign, and would cause confusion. Either through people thinking it has something to do with key-value pairs, or through a simple typo could produce valid but incorrect code. e.g.
This returns an array containing an anonymous function:
return [$x ==> $x * 2];
This returns an array if $x is already a defined variable.
return [$x => $x * 2];
Additionally, I was asked to not reuse the ''%%==>%%'' syntax (http://chat.stackoverflow.com/transcript/message/25421648#25421648) as Hack is already using it. Hence ''~>'' looks like a great alternative.
Also, Hack has some possibilities of typing here, which do not work with PHP, due to technical reasons. Regarding forward compatibility, we might have to choose another syntax than Hack here to resolve these issues. It'd end up being the same operator, with a very similar syntax, potentially confusing. Furthermore using the same syntax than Hack here might lead users to expect types working here and getting really confused.
===== Backward Incompatible Changes =====
This RFC doesn't affect backwards compatibility.
===== Proposed PHP Version(s) =====
Next PHP 7.x; actually 7.1.
===== Future Scope =====
==== Other uses for ~> operator ====
This RFC is solely for using the shorthand anonymous functions as closures. It does not cover any other usage of the shorthand function definition such as:
class Foo {
private $bar:
getBar() ~> $this->bar;
setBar($bar) ~> $this->bar = $bar;
}
Which is outside the scope of this RFC.
==== Type Hints and Return Types ====
This RFC does //not// include type hints nor return types.
Type Hints are not added due to technical problems in parser and the RFC author is not sure about whether they should be really added. If anyone achieves to solve these technical issues, he should feel free to do that in a future RFC for further discussion.
And as introducing half a typesystem would be inconsistent, the RFC proposes to not include return types either.
As an alternative, the current syntax for defining Closures still can be used here.
===== Vote =====
This RFC is a language change and as such needs a 2/3 majority.
Voting opened September 22th, 2015 and will remain open until October 2nd, 2015.
* Yes
* No
===== Patch =====
Pull request is at https://github.com/php/php-src/pull/1254