Can you run some bench comparing with/without LRU, also maybe without cache at all?

Can you run some bench comparing with/without LRU, also maybe without cache at all?

Just a a quick reminder why the cache was implemented: IPv6 checking is extremely slow (in comparison to IPv4).

PHP 8.2.4

Average after 3 runs each and 1 million IP addresses / cache entries:

• With LRU: 3.778 seconds, 2.097152 mb
• Max Items Cache: 3.615 seconds, 2.097152 mb
• Current Cache: 2.833 seconds, 100.663296 mb

Other Tests:

• Keeping count of the items stored and not using count: 3.593 seconds, 2.097152 mb

• Keeping count of the items stored and not using count (LRU): 3.653 seconds, 2.097152 mb

• Without any Cache (IPv4): 2.484 seconds, 2.097152 mb

• Without any Cache (IPv6): 5.741 seconds, 2.097152 mb

An option to improve a tiny bit would be to only cache IPv6 results since IPv4 checking seems to be a little bit faster than ANY caching.

<?php

require './vendor/autoload.php';

use Symfony\Component\HttpFoundation\IpUtils;

$start = microtime(true);

for ($i = 0; $i < 1_000_000; ++$i) {
    IpUtils::checkIp4(random_int(1, 254).random_int(0, 254).random_int(0, 254).random_int(0, 255), '192.168.1.1');
    // IpUtils::checkIp6('::' . $i, 'fe80::/65');
}

$end = microtime(true);

var_dump($end - $start, get_peak_memory_usage());

When only checking about 100 IP addresses the difference becomes basically indistinguishable:

• With LRU: 0.00061893463134766 seconds, 2.097152 mb
• Max Items Cache: 0.00063395500183105 seconds, 2.097152 mb
• Current Cache: 0.00051212310791016 seconds, 2.097152 mb
• Without any Cache (IPv4): 0.00046300888061523 seconds, 2.097152 mb
• Without any Cache (IPv6): 0.00075387954711914 seconds, 2.097152 mb

An option to improve a tiny bit would be to only cache IPv6 results since IPv4 checking seems to be a little bit faster than ANY caching.

let's do this yes! (with an inline comment reminding us about this)
let's drop the -v6 suffix while doing so

With LRU: 3.778 seconds, 2.097152 mb
Without any Cache (IPv6): 5.741 seconds, 2.097152 mb

The difference is not that big actually.

The cache has been introduced in #23098
What about dropping it altogether? The diff doesn't look much worth the added complexity to me.

the benchmark being done seems to be testing the performance in case of cache misses though.

the cache was added to improve performance for the common case of checking the trusted ips many times, which will do cache hits for most checks.

Thank you @danielburger1337.