Optimized the memory usage when writing chunks

[hhxsv5]

Hello guys, I did a test, substr() has no memory overhead and is more efficient than str_split().

$bigStr = str_repeat('abcdefghijklmnopqrstuvwxyz0123456789', 1000000);
$memStart = memory_get_usage(true);
$timeStart = microtime(true);

// ----- substr()
$len = strlen($bigStr);
$chunkLimit = 4096;
for ($offset = 0; $offset < $len; $offset += $chunkLimit) {
    $chunk = substr($bigStr, $offset, $chunkLimit);
}

// ----- str_split()
//$chunkLimit = 4096;
//foreach (str_split($bigStr, $chunkLimit) as $chunk) {
//}

$costMem = memory_get_usage(true) - $memStart;
$costTime = microtime(true) - $timeStart;
echo "Memory: {$costMem} Time: {$costTime}\n";

// substr()
Memory: 0 Time: 0.0036599636077881

// str_split()
Memory: 2097152 Time: 0.043421983718872

reference

[barryvdh]

Why is your reference using factor 0.6 upon writing? Because of gzipping?
And why is the chunkSize 2MB instead of one there?

[hhxsv5]

Swoole's buffer_output_size is 2MB by default, but it sometimes fails(maybe a Swoole bug) when writing 2MB data after testing, So probably use 0.5*2MB, I think it's a safe setting.