👎 also sorry, as this is not a perf critical code path, and the CS change doesn't improve readability.

Regarding the readability, it will if we continue to add builtin processors #25627 (imo it already improves it). It's also the appropriate structure for this type of case.

Well, here is a benchmark... and switch is actually slower than if: (see below, switch is faster).

<?php

require 'vendor/autoload.php';

$container = new \Symfony\Component\DependencyInjection\ContainerBuilder();

class Test {
    private $arg;

    public function __construct(string $arg)
    {
        $this->arg = $arg;
    }

    public function getArg(): string
    {
        return $this->arg;
    }
}

$container
    ->register(Test::class, Test::class)
    ->addArgument('%env(string:FLAG)%')
    ->setShared(false);

$start = microtime(true);
for ($i = 0; $i < 1000000; $i++) {
    $container->get(Test::class)->getArg();
}
$end = microtime(true);

echo sprintf("Elapsed: %f\n", $end - $start);

Average with switch: 8.350835
Average with if: 6.927684

My patch targets the 3.4 🍎 branch. And I was comparing with master 🍐.

Comparing 3.4 and my patch, the results are of course more coherent:

if (avg): 8.954663
switch (avg): 8.880407

The good thing is: master is way faster than 3.4!

Regarding those new numbers, I think we should merge this patch: it's indeed a micro-optim, but promoting new PHP features, performance best practices and the appropriate construct (even if we can debate that) is a thing Symfony should do.

I don't think it makes sense to make such a change. We are not promoting anything. Nobody will notice that. Let's close.