[7.x] Optimize Str::startsWith()

[SjorsO]

This PR makes Str::startsWith around 10% faster by using strncmp.

There have been two PRs in the past that have tried to optimize this method using strpos: #30952 and #16761. Using strpos is not a good idea, this comment on a previous PR explains why.

Benchmark

Haystack	Needle	Time	%
Rows below use the current substr implementation
test	te	1193
test	no	1250
str_repeat('test', 2000)	str_repeat('test', 20)	1368
str_repeat('test', 2000)	str_repeat('nope', 20)	1344
str_repeat('test', 2000000)	test	1244
str_repeat('test', 2000000)	nope	1260
test	str_repeat('test', 2000000)	1198
Rows below use a strpos implementation
test	te	990	+17.0% ✔️
test	no	964	+22.9% ✔️
str_repeat('test', 2000)	str_repeat('test', 20)	2674	-95.4% ❌
str_repeat('test', 2000)	str_repeat('nope', 20)	7007	-421% ❌
str_repeat('test', 2000000)	test	1018	+18.1% ✔️
str_repeat('test', 2000000)	nope	99999+	-999% ❌
test	str_repeat('test', 2000000)	1003	+16.3% ✔️
Rows below use the strncmp implementation proposed in this PR
test	te	1039	+12.9% ✔️
test	no	1082	+13.4% ✔️
str_repeat('test', 2000)	str_repeat('test', 20)	1192	+12.9% ✔️
str_repeat('test', 2000)	str_repeat('nope', 20)	1205	+10.3% ✔️
str_repeat('test', 2000000)	test	1080	+13.2% ✔️
str_repeat('test', 2000000)	nope	1108	+12.1% ✔️
test	str_repeat('test', 2000000)	1181	+1.4% ✔️

Benchmark code

/** @test */
function benchmark_str_starts_with()
{
    $haystack = 'test';

    $needle = 'te';

    $speed = [];

    for ($times = 0; $times < 5; $times++) {
        $startedAt = microtime(true) * 1000;

        for ($i = 0; $i < 10000000; $i++) {
            Str::startsWith($haystack, $needle);
        }

        $speed[] = (microtime(true) * 1000) - $startedAt;
    }

    dump(
        'Haystack: '.Str::limit($haystack, 20),
        'Needle: '.Str::limit($needle, 20),
        $speed,
        'Average: '.(array_sum($speed) / count($speed))
    );
}

[taylorotwell]

Does this work with UTF-8 strings?

[SjorsO]

As far as I know, yes.

In the same way the old code used substr instead of mb_substr, we can use a non-multibyte function because we are comparing the start of a string, not a specific offset.

I've added an additional test with some Chinese characters (that are definitely UTF-8).

[vlakoff]

The endsWith() method could be optimized similarly, by using substr_compare() with a negative index argument:

substr_compare($haystack, $needle, -strlen($needle)) === 0

props Will Hudgins, I discovered this code in a RFC for PHP 8 he has authored.

[vlakoff]

I have run some benchmarks with the above code for endsWith(), and it isn't faster. It seems to be a bit slower actually.

According to the RFC for PHP 8, it should be more memory efficient, still, but I think the more important is execution speed.