Zend: Fix iterator relocation at the current element during rehash

[LamentXU123]

Found when I am implementing #23585

<?php
$a = [0, 1, 2];
foreach ($a as &$outer) {
    foreach ($a as &$value) {
        echo $value, "\n";
        if ($value === 1) {
            unset($a[0], $a[1]);
            $a['x'] = 3; 
        }
    }
    break;
}

Return

0
1

But I expect:

0
1
2
3

[devnexen]

nit: while at it, you can also fix zend_array_dup_elements()

[LamentXU123]

Right. and I think this is the similar fix to the existing one so we don't need to add more NEWS entry.

[LamentXU123]

I am testing my new agent workflow to help me branching written PR to remote and obviously it glitched so this is unintentionally closed and reopened. Sorry for the noise.

[iliaal]

Two more relocation gaps in this function, both pre-existing and both already handled in zend_hash_rehash():

• No one-past-the-end migration. Rehash does it at :1434, and this loop cannot reach that position under either bound, since lower_pos() returns nNumUsed as both its sentinel and a real cursor there. A by-ref foreach on the last element then loses elements appended after a COW dup.
• :2411 and :2424 still use source->nInternalPointer == idx, the form 5d40592 replaced in rehash with the (j, i] range at :1389. A cursor parked on a hole is never remapped, so key()/next() skips live elements. Test target-> rather than source->, at both sites.

Follow-up is fine, neither is yours.

[iliaal]

$copy is write-only and is the only reason this write separates through zend_array_dup_elements() instead of zend_hash_rehash(). Delete it and the test still passes, but it becomes a duplicate of rehash_multiple_iterators.phpt case 2 and :2431 loses its only pin.

echo 'copy: ', implode(' ', array_keys($copy)), "\n"; with copy: c d e f g h in --EXPECT-- makes the line undeletable. Optional, nothing is wrong today.

[iliaal]

Fix looks right. The bound now agrees with the nInternalPointer range sitting three lines up at :1389, which does the same job for the array's own cursor. Both tests pin distinct sites: apply either bound alone and the other test still fails.

zend_array_dup_elements() has two more relocation gaps that rehash already handles, both pre-existing. Inline.

[iliaal]

LGTM, thanks for making the changes!