|
php.net | support | documentation | report a bug | advanced search | search howto | statistics | random bug | login |
[2016-07-14 15:45 UTC] dominikschilling+php at gmail dot com
Description: ------------ I noticed a weird issue when running the unit tests of WordPress with PHP 7.1.0. Our ticket: https://core.trac.wordpress.org/ticket/37295 The failing test: https://travis-ci.org/aaronjorbin/develop.wordpress/jobs/142322444#L387-L396 It looks like array_slice() is removing the reference of a variable under certain conditions. The test script is a short form of what WordPress is doing here. Test script: --------------- <?php global $wp_filter; $wp_filter[ 'pre_get_posts' ][ 9 ][ 1 ] = [ 'function' => 'function_which_expects_a_reference', 'accepted_args' => 1 ]; $wp_filter[ 'pre_get_posts' ][ 9 ][ 2 ] = [ 'function' => 'function_which_expects_a_reference', 'accepted_args' => 1 ]; class FooBar { function __construct() { do_action_ref_array( 'pre_get_posts', [ &$this ] ); } } function do_action_ref_array( $tag, $args ) { global $wp_filter; reset( $wp_filter[ $tag ] ); do { foreach ( current( $wp_filter[ $tag ] ) as $the_ ) { $a = array_slice( $args, 0, (int) $the_['accepted_args'] ); xdebug_debug_zval( 'a' ); call_user_func_array( $the_['function'], $a ); } } while ( next( $wp_filter[ $tag ] ) !== false ); } function function_which_expects_a_reference( &$variable ) {} new FooBar(); Expected result: ---------------- a: (refcount=1, is_ref=0)=array (0 => (refcount=3, is_ref=1)=class FooBar { }) a: (refcount=1, is_ref=0)=array (0 => (refcount=3, is_ref=1)=class FooBar { }) Actual result: -------------- a: (refcount=1, is_ref=0)=array (0 => (refcount=1, is_ref=1)=class FooBar { }) a: (refcount=1, is_ref=0)=array (0 => (refcount=4, is_ref=0)=class FooBar { }) PHP Warning: Parameter 1 to function_which_expects_a_reference() expected to be a reference, value given in /php71.php on line 21 PHP Stack trace: PHP 1. {main}() /php71.php:0 PHP 2. FooBar->__construct() /php71.php:28 PHP 3. do_action_ref_array() /php71.php:8 Warning: Parameter 1 to function_which_expects_a_reference() expected to be a reference, value given in /php71.php on line 21 Call Stack: 0.0006 361384 1. {main}() /php71.php:0 0.0030 362576 2. FooBar->__construct() /php71.php:28 0.0030 362976 3. do_action_ref_array() /php71.php:8 PatchesPull RequestsHistoryAllCommentsChangesGit/SVN commits
|
|||||||||||||||||||||||||||
Copyright © 2001-2025 The PHP GroupAll rights reserved. |
Last updated: Mon Oct 27 03:00:02 2025 UTC |
(That it needs two iterations is another bug in array_slice). The general issue here is the refcount of the reference being 1, as &$this doesn't make $this a reference; it merely fetches the $this value (due to some optimization) and references that value, resulting in a reference of refcount 1... I have no idea how to fix that without leaking the reference or storing it on the object and checking for it in each dtor... Simpler repro: function ref(&$ref) {} new class { function __construct() { $args = [&$this]; for ($i = 0; $i < 2; $i++) { $a = array_slice($args, 0, 1); call_user_func_array('ref', $a); } } };Thanks for link to the RFC. I think I'm missing the point why "re-assign $this" == "modification of $this" which is basically "$this = 'something'" vs. "$this->something = 'something'". For example the following script is still working and also a modification of $this, although it's another context: class C { public $b; function foo(){ $this->b = 'a'; } } $x = new C; $x->foo(); var_dump($x->b); // Prints a > That it needs two iterations is another bug in array_slice Does that now mean that it should already fail on the first iteration?The problem is not related to $this handling at all. Test script: --------------- <?php function ref(&$a) { var_dump($a); } $b = 1; $args = [&$b]; unset($b); for ($i = 0; $i < 2; $i++) { $a = array_slice($args, 0, 1); call_user_func_array('ref', $a); } ?> Expected result: ---------------- int(1) int(1) Actual result: -------------- int(1) PHP Warning: Parameter 1 to ref() expected to be a reference, value given in %s on line %d I think this warning doesn't make a lot of sense at all, or at least it shouldn't prevent execution of the called function. Fixing this is a new minor BC break.