Bug #63217 [Asn]: Behavior of implementing the ArrayAccess interface has been changed in php 5.4

From: Date: Tue, 24 Nov 2015 15:56:57 +0000
Subject: Bug #63217 [Asn]: Behavior of implementing the ArrayAccess interface has been changed in php 5.4
References: 1  Groups: php.bugs 
Request: Send a blank email to php-bugs+get-197396@lists.php.net to get a copy of this message
Edit report at https://bugs.php.net/bug.php?id=63217&edit=1

 ID:                 63217
 Updated by:         ajf@php.net
 Reported by:        kmsheng at pixnet dot tw
 Summary:            Behavior of implementing the ArrayAccess interface
                     has been changed in php 5.4
 Status:             Assigned
 Type:               Bug
 Package:            Arrays related
 Operating System:   freebsd
 PHP Version:        5.4Git-2012-10-04 (Git)
 Assigned To:        ajf
 Block user comment: N
 Private report:     N

 New Comment:

To clarify why this happens, it wasn't due to wanting to make ArrayAccess like arrays. Rather,
PHP has an optimisation for array indexing with a constant (a literal like "123", 123,
true, false, etc, not the const/define() kind) string (e.g. $_POST["password"]) where it
will check at compile-time if it is numeric and replace it then if so (so $foobar["123"]
becomes $foobar[123], but $foobar["bar"] stays the same). This means we don't have to
run the numeric string check every time that line of code is executed. Arrays in PHP consider
$foo["1"] and $foo[1] to be the same, so otherwise we'd need to check when we run the
code if the string is a number.

The problem is that $foobar in this case might actually be an object with ArrayAccess and not an
array, and so what apparently would be a transparent optimisation ends up causing this bug.

The solution is to remove this optimisation. This won't necessarily cause a performance hit,
because there's other ways we could avoid doing the numeric string check at runtime.


Previous Comments:
------------------------------------------------------------------------
[2015-11-24 15:10:48] levim@php.net

Changing this back to a bug status. Here is an example where I hope it is clear it is a bug (on
3v4l: https://3v4l.org/XGtVh):

<?php
class Dictionary implements ArrayAccess {
	function offsetExists($offset) {}
	function offsetGet($offset) {}
	function offsetUnset($offset) {}

	function offsetSet($offset, $value) {
		if (!is_string($offset)) {
			throw new InvalidArgumentException();
		}
	}
}

try {
    $Dictionary = new Dictionary();
    $Dictionary["12"] = 0xDEADBEEF;
    echo "No Exception for \"12\"\n";
} catch (InvalidArgumentException $e) {
    echo "Caught Exception for \"12\"\n";
}

try {
    $str = "12";
    $Dictionary[$str] = 0xDEADBEEF;
    echo "No Exception for \$variable = \"12\"\n";
} catch (InvalidArgumentException $e) {
    echo "Caught Exception for \$variable = \"12\"\n";
}

?>

------------------------------------------------------------------------
[2015-11-24 05:22:51] rudolf dot theunissen at gmail dot com

It makes sense for array keys to be coerced to intergers, ie. 1, 1.0, and "1" will attempt
to access the same index. However, this does not make sense for ArrayAccess. The argument for
consistency is poor, because float and object keys are handled differently already (floats stay
floats and objects don't raise warnings).  One example use case is SplObjectStorage, which
makes use of object keys via ArrayAccess.

With the introduction of strict types and scalar typehints, it makes sense for the implementation to
be responsible for handling various key types.

ArrayAccess should only provide array syntax, not array behaviour.

Test script:
https://3v4l.org/0BMYh
---------------
<?php

class Test implements ArrayAccess
{
    public function offsetGet($offset){
        return gettype($offset);
    }
    
    public function offsetSet($offset, $value){}
    public function offsetUnset($offset){}
    public function offsetExists($offset){}
}

$a = array('a', 'b', 'c');
$o = new Test();
$s = '1';

echo $o[1],   "\n"; // 'integer'
echo $o['1'], "\n"; // 'integer' !?
echo $o[1.0], "\n"; // 'double'
echo $o[$s],  "\n"; // 'string'
echo $o[$o],  "\n"; // 'object'

echo $a[1],   "\n"; // 'integer'
echo $a['1'], "\n"; // 'integer'
echo $a[1.0], "\n"; // 'integer'
echo $a[$s],  "\n"; // 'integer'
echo $a[$o],  "\n"; // Warning: Illegal offset type

Output:
---------------
integer
integer
double
string
object
b
b
b
b

Warning: Illegal offset type in...

------------------------------------------------------------------------
[2013-04-07 17:44:34] david at grudl dot com

This is really buggy.

$test['0']; // prints int(0)

but

$key = '0'; 
$test[$key]; // prints string(1) "0"

------------------------------------------------------------------------
[2012-10-05 04:54:32] laruence@php.net

numeric string key will cast to number, I think this is a improvement, to be 
consistent with the internal dim fetch.

but yeah, we need a document of such change, change to doc problem.

------------------------------------------------------------------------
[2012-10-04 10:14:49] kmsheng at pixnet dot tw

Description:
------------
Execute the test script provided below in php 5.3.10 and php 5.4.0-3 may get different results which
are string(1) "0" and int(0).

I don't know if this is a bug or a patch.

Test script:
---------------
<?php
class Test implements ArrayAccess {

    public function offsetExists($off)
    {
    }

    public function offsetUnset($off)
    {
    }

    public function offsetSet($off, $el)
    {
    }

    public function offsetGet($off)
    {
        var_dump($off);
    }
}

$test = new Test();

$test['0'];

Expected result:
----------------
string(1) "0"

Actual result:
--------------
int(0)


------------------------------------------------------------------------



--
Edit this bug report at https://bugs.php.net/bug.php?id=63217&edit=1


Thread (7 messages)

« previous php.bugs (#197396) next »