`memory_limit` is ignored

[mvorisek]

Description

The following code:

<?php

$logFx = function ($v, $clear = false) {
    $p = __DIR__ . '/../../log';

    ob_start();
    var_dump($v);
    $str = ob_get_clean();
    file_put_contents($p, $str, $clear ? 0 : FILE_APPEND);
};
$logFx('', true);

$logFx(ini_get('memory_limit'));
$logFx('x');

$logFx(memory_get_usage(true));
$logFx(memory_get_usage(false));

ini_set('memory_limit', '16M');

$str = '';
for ($i = 0; $i < 4 * 1024; ++$i) {
    $str .= random_bytes(16 * 1024);
    $logFx(memory_get_usage(true));
}

$logFx(memory_get_usage(true));
$logFx(memory_get_usage(false));
$logFx(ini_get('memory_limit'));
$logFx('y');

repro repo: https://github.com/atk4/ui/tree/refs/heads/php_memory_limit_ignored

CI output: https://github.com/atk4/ui/actions/runs/24887405656/job/72870734637#step:10:4112

Resulted in this output:

int(200)
string(57473) "string(0) ""
string(4) "128M"
string(1) "x"
int(18874368)
int(1950048)
...
int(83824640)
int(83841024)
int(83857408)
int(83873792)
int(83890176)
int(83890176)
int(69063032)
string(3) "16M"
string(1) "y"

But I expected this output instead:

int(200)
string(57473) "string(0) ""
string(4) "128M"
string(1) "x"
int(18874368)
int(1950048)
...

must terminate before this is output:
string(1) "y"

PHP, randomly ignores memory_limit and it seems it allocates a new memory page even when real memory usage is already over the limit.

PHP Version

PHP 8.5.5 (cli) (built: Apr 22 2026 11:22:58) (NTS)
Copyright (c) The PHP Group
Built by https://github.com/docker-library/php
Zend Engine v4.5.5, Copyright (c) Zend Technologies
    with Xdebug v3.5.1, Copyright (c) 2002-2026, by Derick Rethans
    with Zend OPcache v8.5.5, Copyright (c), by Zend Technologies

Operating System

linux

[iliaal]

@mvorisek Does this happen without Xdebug, I tried to reproduce locally, but hit memory limit as expected.

[mvorisek]

It hapens only with xdebug, more precisely with coverage mode. But not always.

I think the reason is not xdebug itself, but coverage dumps which are memory heavy and the allocated memory chunks are larger than the new limit and the error is emit only when a new memory chunk is needed.

If someone knows what could be the issue please share. I will try to put this into self contained Dockerfile to make it more localy reproducible.

[iluuu1994]

I cannot reproduce this. Please find a reproducer without Xdebug or report over there.