[5.4] Optimize UrlGenerator->isValidUrl()

[vlakoff]

UrlGenerator->to() and UrlGenerator->asset() are quite slow. Investigating about this, I found out UrlGenerator->validUrl() was a bottleneck, taking by itself about half of the total execution time.

Then, I found out this method could be made significantly faster:

$path = 'foobar';
$nb = 10000;

$t1 = microtime(true);
for ($i = $nb; $i--; ) {
    if (Str::startsWith($path, ['#', '//', 'mailto:', 'tel:', 'http://', 'https://'])) {}
}
$t2 = microtime(true);
for ($i = $nb; $i--; ) {
    if (preg_match('~^(#|//|https?://|mailto:|tel:)~', $path)) {}
}
$t3 = microtime(true);

echo $t2 - $t1, "\n", $t3 - $t2;

Before: 673 ms
After: 76 ms

[Modelizer]

it should include http as well?

[ntzm]

It does, https?:// matches https:// and http://

[vlakoff]

I investigated why the http[s] protocols had been added (ec1b1ae) while the filter_var below supposedly supports them already (and does naive checks thus is quite permissive).

So, I guess it's to allow internationalized domain names, which are not supported by filter_var():

filter_var('http://스타벅스코리아.com', FILTER_VALIDATE_URL) // false

[fernandobandeira]

It should work with php-intl enabled...

Anyway shouldn't we consider making this check more strict? Currently it seems that http://anything will pass even if theres no .tld there...

[vlakoff]

It should be fine as it is. This method is not really for proper validation, but just to determine if the argument is a route or an URL, so that the developer can pass both to e.g. url() without caring.