rfc:argument_unpacking

PHP RFC: Argument Unpacking

Introduction

This RFC complements the variadics RFC. It introduces a syntax for unpacking arrays and Traversables into argument lists (also known as “splat operator”, “scatter operator” or “spread operator”).

As a usage example, consider a variadic method public function query($query, ...$params). You are provided a $query and an array of $params and want to call the method using these. Currently this is possible using call_user_func_array():

call_user_func_array([$db, 'query'], array_merge(array($query), $params));

This RFC proposes a syntax for unpacking arguments directly in the call syntax:

$db->query($query, ...$params);

Proposal

An argument in a function call that is prefixed by ... will be “unpacked”: Instead of passing the argument itself to the function the elements it contains will be passed (as individual arguments). This works both for arrays and Traversables.

As such all of the following function calls are equivalent:

function test(...$args) { var_dump($args); }
 
test(1, 2, 3);                         // [1, 2, 3]
test(...[1, 2, 3]);                    // [1, 2, 3]
test(...new ArrayIterator([1, 2, 3])); // [1, 2, 3]
 
// Note: It doesn't really make sense to unpack a constant array like [1, 2, 3].
//       Normally these would unpack some variable like ...$args

It's possible to use ... multiple times in a call and it is possible to use normal arguments before argument unpacking:

$args1 = [1, 2, 3];
$args2 = [4, 5, 6];
test(...$args1, ...$args2); // [1, 2, 3, 4, 5, 6]
test(1, 2, 3, ...$args2);   // [1, 2, 3, 4, 5, 6]

However, it is not possible to use normal arguments after argument unpacking was used. Both of the following are invalid:

test(...$args1, 4, 5, 6);
test(...$args1, 4, 5, 6, ...$args2);

The ... operator works in all argument lists, including new expressions:

fn(...$args);
$fn(...$args);
$obj->fn(...$args);
ClassName::fn(...$args);
new ClassName(...$args);

Argument unpacking is not limited to variadic functions, it can also be used on “normal” functions:

function test($arg1, $arg2, $arg3 = null) {
    var_dump($arg1, $arg2, $arg3);
}
 
test(...[1, 2]);       // 1, 2
test(...[1, 2, 3]);    // 1, 2, 3
test(...[1, 2, 3, 4]); // 1, 2, 3 (remaining arg is not captured by the function declaration)

If you try to unpack something that is not an array or Traversable a warning is thrown, but apart from that the call continues as usual:

var_dump(1, 2, ...null, ...[3, 4]);
// Warning: Only arrays and Traversables can be unpacked
// int(1) int(2) int(3) int(4)

By-reference passing

If an array is unpacked the elements will by passed by-value/by-reference according to the function definition:

function test($val1, $val2, &...$refs) {
    foreach ($refs as &$ref) ++$ref;
}
 
$array = [1, 2, 3, 4, 5];
test(...$array);
var_dump($array); // [1, 2, 4, 5, 6]

By-reference passing will not work if the unpacked entity is a Traversable. Instead an E_WARNING level error is thrown and the argument is passed by-value instead:

test(...new ArrayIterator([1, 2, 3, 4, 5]));
// Warning: Cannot pass by-reference argument 3 of test() by unpacking a Traversable, passing by-value instead

The reasons why we can't pass by-reference from a Traversable are two-fold:

  • It's not possible to determine the number of elements in a Traversable ahead of time. As such we can not know whether unpacking the Traversable will or will not hit a by-reference argument.
  • It's not possible to determine if a Traversable has support for by-reference iteration or if it will trigger an error if this is requested.

String keys

In order to ensure forward-compatibility with named arguments the unpacking operator does not support string keys. If a string key is encountered during unpacking a recoverable error is thrown. If the error is ignored using a custom error handler, no further arguments will be unpacked but the call still happens.

Backward Compatibility

This change does not break userland or internal compatibility.

Advantages over call_user_func_array

Usage of call_user_func_array becomes complicated if you need to pass fixed arguments as well. Compare:

call_user_func_array([$db, 'query'], array_merge(array($query), $params));
// vs
$db->query($query, ...$params);

call_user_func_array requires a callback. So even if the called function/method is known, you still need to use a dynamic string/array callback. This usually precludes any IDE support.

call_user_func_array does not work for constructors. Instead ReflectionClass::newInstanceArgs() has to be used:

(new ReflectionClass('ClassName'))->newInstanceArgs($args);
// vs
new ClassName(...$args);

Futhermore call_user_func_array has a rather large performance impact. If a large number of calls go through it, this can make a signficant difference. For this reason projects 1) often replace particularly common call_user_func_array calls with a switch statement of the following form:

switch (count($args)) {
    case 0: $func(); break;
    case 1: $func($args[0]); break;
    case 2: $func($args[0], $args[1]); break;
    case 3: $func($args[0], $args[1], $args[2]); break;
    case 4: $func($args[0], $args[1], $args[2], $args[3]); break;
    case 5: $func($args[0], $args[1], $args[2], $args[3], $args[4]); break;
    default: call_user_func_array($func, $args); break;
}

The ... argument unpacking syntax is about 3.5 to 4 times faster than call_user_func_args. This solves the performance issue. Benchmark code and results.

Lastly, it seems that people naturally expect that this syntax is present if the variadics syntax is present. So if we implement variadics, it's probably best to include this as well.

Examples

The code samples in the “Proposal” section are rather technical and not code you would actually write. This section contains a few more practical examples of this feature.

Extending variadic functions: forwarding

The introduction already mentioned $db->query($query, ...$params) as an example. At this point you could wonder: Why would I want to write code like that? Why should I have the parameters only as an array?

One case where this occurs is when extending variadic functions:

class MySqlWithLogging extends MySql {
    protected $logger;
    public function query($query, ...$params) {
        $this->logger->log(
            'Running query "%s" with parameters [%s]',
            $query, implode(', ', $params)
        );
 
        return parent::query($query, ...$params);
    }
}

The above code sample extends the variadic query() method with logging and needs to forward all arguments to the parent function.

Partial application: multiple unpacks

Some people were wondering on what occasion you would ever want to unpack *two* arguments in one function call. An example of such a usage is “partial application”.

If you are not familiar with the concept, partial application allows you to “bind” arguments to a function:

$arrayToLower = bind('array_map', 'strtolower');
 
$arrayToLower(['Foo', 'BAR', 'baZ']); // returns ['foo', 'bar', 'baz']
 
// The above $arrayToLower call resolves to:
// array_map('strtolower', ['Foo', 'BAR', 'baZ'])

This is a common functional paradigm, but rather rarely used in PHP. Anyway, an “old-style” (no variadic syntax, no argument unpacking) definition of the bind() function would look like this:

function bind(callable $function) {
    $boundArgs = array_slice(func_get_args(), 1);
    return function() use ($function, $boundArgs) {
        return call_user_func_array(
            $function, array_merge($boundArgs, func_get_args())
        );
    }
}

And the “new-style” definition (with variadic syntax and argument unpacking) looks like this:

function bind(callable $function, ...$boundArgs) {
    return function(...$args) use($function, $boundArgs) {
        return $function(...$boundArgs, ...$args);
    }
}

Patch

The patch for this features is available as a PR: https://github.com/php/php-src/pull/477

Vote

As this is a language change a two third majority is required.

Voting started 2013-12-21 and ended 2014-01-11.

Implement argument unpacking in PHP 5.6?
Real name Yes No
ajf (ajf)  
auroraeosrose (auroraeosrose)  
bukka (bukka)  
bwoebi (bwoebi)  
chobieeee (chobieeee)  
colder (colder)  
cpriest (cpriest)  
daverandom (daverandom)  
davey (davey)  
dm (dm)  
Fabien Potencier (fabpot)  
indeyets (indeyets)  
jpauli (jpauli)  
jwage (jwage)  
kalle (kalle)  
krakjoe (krakjoe)  
kriscraig (kriscraig)  
levim (levim)  
lstrojny (lstrojny)  
mbeccati (mbeccati)  
mfonda (mfonda)  
mike (mike)  
mrook (mrook)  
nikic (nikic)  
peehaa (peehaa)  
philstu (philstu)  
pollita (pollita)  
rdlowrey (rdlowrey)  
reeze (reeze)  
stas (stas)  
treffynnon (treffynnon)  
tyrael (tyrael)  
willfitch (willfitch)  
yohgaki (yohgaki)  
Final result: 32 2
This poll has been closed.

Support in other languages

This feature is supported by many languages. Some of the more important ones being:

  • Python using the *args syntax
  • Ruby using Python's syntax
  • Java supports this, but only for variadic parameters and without any special syntax (type based)
  • JavaScript (ECMAScript Harmony) using the same syntax proposed here
1)
I've seen this used at least in Laravel and Drupal and a bunch of other code
rfc/argument_unpacking.txt · Last modified: 2017/09/22 13:28 by 127.0.0.1