Modifying a copied by-ref iterated array resets the array position

[survik1]

Description

Hello, we are finally pushing our code to PHP 8 and I have encountered this BC break which seems more like bug than feature to me. This behavior is still present in 8.2 and the change was not documented (unless i missed it ofc).

https://3v4l.org/aNejW

<?php

$_ENV = [
    'ZASIS_UI_BACKGROUNDCOLOR' => '#FFF1EC',
    'ZASIS_UI_INFOPANEL_ENABLE' => '1',
    'ZASIS_UI_INFOPANEL_TEXT' => 'DEBUG IS ON',
    'ZASIS_UI_INFOPANEL_BACKGROUNDCOLOR' => '#F44',
];

$arr = [
    'ui' => [
        'backgroundColor' => 'ZASIS_UI_BACKGROUNDCOLOR',
        'infoPanel' => [
            'enable' => 'ZASIS_UI_INFOPANEL_ENABLE',
            'text' => 'ZASIS_UI_INFOPANEL_TEXT',
            'fontColor' => 'ZASIS_UI_INFOPANEL_FONTCOLOR',
            'backgroundColor' => 'ZASIS_UI_INFOPANEL_BACKGROUNDCOLOR',
        ]
    ]
];

function loadEnv(&$data): void
{
    if (is_array($data)) {
        foreach ($data as $key => &$value) {
            try {
                loadEnv($value);
            } catch (\InvalidArgumentException $e) {
                unset($data[$key]);
            }
        }
    } elseif (array_key_exists($data, $_ENV)) {
        $data = $_ENV[$data];
    } else {
        throw new \InvalidArgumentException();
    }
}
loadEnv($arr);
print_r($arr);

Output in PHP >=8.0:

Array
(
    [ui] => Array
        (
            [backgroundColor] => #FFF1EC
            [infoPanel] => Array
                (
                    [backgroundColor] => #F44
                )

        )

)

Output in PHP >=7.1 <8.0:

Array
(
    [ui] => Array
        (
            [backgroundColor] => #FFF1EC
            [infoPanel] => Array
                (
                    [enable] => 1
                    [text] => DEBUG IS ON
                    [backgroundColor] => #F44
                )

        )

)

PHP Version

PHP 8.2.6

Operating System

No response

[KapitanOczywisty]

Minimal reproducible example: https://3v4l.org/oU2Wj

$data = range(0,3);

function foo($data): void
{
    foreach ($data as $key => &$value) {
        echo "{$value}\n";
        try {
            if($value === 2) {
                throw new \Exception;
            }
        } catch (\Exception $e) {
            echo "remove {$value}\n";
            unset($data[$key]);
        }
    }
}

foo($data);

Output 8.0+

0
1
2
remove 2
0
1
3

Output before 8.0

0
1
2
remove 2
3

@survik1 It's best to avoid using references whenever possible, they were always big source of problems. Just removing reference from the foreach fixes this particular issue: https://3v4l.org/Otugh

[bwoebi]

A simpler example:

$data = [0, 1, 2];

foreach ($data as $key => &$value) {
    echo "{$value}\n";
    if ($value === 1) {
        $cow_copy = $data;
        echo "unset $value\n";
        unset($data[$key]);
    }
}

giving

0
1
unset 1
0 // restarting iteration here at the position of the internal array pointer
2

The problem is modifying a by-ref iterated array whose RC > 1.

This causes duplication of the array, assigning a new array to $data. The foreach iterator then thinks it's a new array and starts over...

This is not a new issue, but an issue existing since PHP 7: https://3v4l.org/VoceP. It just happens that since PHP 8 exceptions do a CoW copy (reference counter increase) of the underlying value instead of the reference, exhibiting the side effect of the referenced array needing to be duplicated upon modification, causing this issue. Which should be an implementation detail without visible side effects, except in this case where it apparently surfaces this issue.