[StimulusBundle] Speed up Stimulus name normalization

[Kocal]

Q	A
Bug fix?	no
New feature?	no
Deprecations?	no
Documentation?	no
Issues	-
License	MIT

normalizeKeyName() ran three preg_* calls for every value, class and param name, and normalizeControllerName() ran a preg_replace() for every controller name. Both run on every stimulus_controller() / stimulus_action() call, on names that repeat across the whole page.

Memoize the key names (a fixed set coming from templates and controllers) and drop the regex from the controller name, which only strips a leading @.

Rendering 50k controllers with values, classes, an action and a target goes from ~414 ms to ~356 ms. On a more realistic page though, an admin CRUD page with 100 Stimulus controllers, that's 0.45 ms -> 0.42 ms per page: a 7% cut on a budget that's already under a millisecond. smnandre asked in review whether the project should have a minimum gain worth accepting, given the maintenance and history cost of a change like this. That's a fair question: this is a small, low-risk win, not a big one, and reviewers should weigh it as such rather than as more than it is.

Benchmarked from the repository root with blackfire run symfony php bench.php:

<?php
require __DIR__.'/src/StimulusBundle/vendor/autoload.php';

use Symfony\UX\StimulusBundle\Dto\StimulusAttributes;
use Twig\Environment;
use Twig\Loader\ArrayLoader;

$env = new Environment(new ArrayLoader([]));

for ($i = 0; $i < 5000; ++$i) {
    $attributes = new StimulusAttributes($env);
    $attributes->addController('my-controller', ['name' => 'ryan', 'isEnabled' => true, 'count' => 42], ['loading' => 'spinner']);
    $attributes->addAction('my-controller', 'onClick', 'click');
    $attributes->addTarget('my-controller', 'element');
    (string) $attributes;
}

Blackfire:

• before (833ms wall / 816ms CPU): https://app.blackfire.io/envs/5f4f9a62-eaa0-45ee-b7b3-a1b879f550e9/profiles/4d6d06bc-f658-4786-952a-302cc8c852ec/graph
• after (632ms wall / 628ms CPU): https://app.blackfire.io/envs/5f4f9a62-eaa0-45ee-b7b3-a1b879f550e9/profiles/c56f1954-74f2-40ad-9514-eab58bcb8912/graph
• diff: https://app.blackfire.io/envs/5f4f9a62-eaa0-45ee-b7b3-a1b879f550e9/profiles/compare/4d6d06bc-f658-4786-952a-302cc8c852ec...c56f1954-74f2-40ad-9514-eab58bcb8912/graph

Analysis, implementation and benchmarks by Claude Opus 5.

[smnandre]

Should we define a minimum gap to accept improvements ?

There is a hidden cost in maintenance, history, etc.

Genuine question, I have no opinion here :)

[Kocal]

Should we define a minimum gap to accept improvements ?

There is a hidden cost in maintenance, history, etc.

Genuine question, I have no opinion here :)

Before AI, I'd say we had to be careful with performance patches when it came to code readability, but now I think it's fine... The code is still easy to read IMHO