Benchmarks between 2 implementations of Collator::replace and Regexp::replaceCallback

collator_replace_results.md

// #define UTF8_REPLACEMENT 1

Changes:

1. convert replacement from UTF-8 to UTF-16 (before processing)
2. copy (strdup) UTF-16 version of suject as result (before processing)
3. convert back result from UTF-16 to UTF-8 (before returning)
4. we directly work on UTF-16 code unit offset (internal for replacements)

$string = str_repeat('Hayır İyi', 1);

1 iteration

real 0m0.022s user 0m0.018s sys 0m0.004s

1 000 000 iterations

real 0m18.278s user 0m18.261s sys 0m0.010s

$string = str_repeat('Hayır İyi', 10000);

1 iteration

real 0m0.093s user 0m0.085s sys 0m0.008s

10 000 iterations

real 17m4.348s user 16m59.614s sys 0m4.092s

#define UTF8_REPLACEMENT 1

Changes:

1. we don't need to convert replacement in UTF-16
2. copy (strdup) UTF-8 version of suject as result (before processing)
3. result is already in UTF-8 (before returning)
4. we need to convert length in code points (UTF-16 => UTF-8 to map offsets) (internal for replacements)

$string = str_repeat('Hayır İyi', 1);

1 iteration

real 0m0.022s user 0m0.014s sys 0m0.008s

1 000 000 iteration

real 0m21.291s user 0m21.263s sys 0m0.015s

$string = str_repeat('Hayır İyi', 10000);

1 iteration

real 0m4.960s user 0m4.951s sys 0m0.006s

10 000 iteration

too long (more than 13h ?)

collator_replace_script.php

<?php
$coll = new Collator('tr_tr');
$coll->setStrength(Collator::SECONDARY);

$search = 'I';
$string = str_repeat('Hayır İyi', 1);

for ($i = 0; $i < 1; $i++) {
        $coll->replace($string, $search, '<replaced>');
}

regexp_callback_script.php

<?php
$re = new Regexp('i', 'i');
$string = str_repeat('Hayır İyi', 1);

for ($i = 0; $i < 1; $i++) {
    $re->replaceCallback(
        $string,
        function ($matches) {
            static $m = 0;
            return '<replacement n°' . $m++ . '>';
        }
    );
}

regexp_replaceCallback_results.md

// #define UTF8_REPLACEMENT 1

Changes:

1. copy (strdup) UTF-16 version of suject as result (before processing)
2. for each match, convert replacement (callback returned value) from UTF-8 to UTF-16 (while processing)
3. convert back result from UTF-16 to UTF-8 (before returning)
4. we directly work on UTF-16 code unit offset (internal for replacements)

$string = str_repeat('Hayır İyi', 1);

1 iteration

real 0m0.023s user 0m0.015s sys 0m0.007s

1 000 000 iterations

real 0m14.406s user 0m14.394s sys 0m0.006s

$string = str_repeat('Hayır İyi', 10000);

1 iteration

real 0m0.140s user 0m0.131s sys 0m0.009s

10 000 iterations

real 28m54.813s user 28m47.893s sys 0m5.892s

#define UTF8_REPLACEMENT 1

Changes:

1. we don't need to convert replacement in UTF-16
2. copy (strdup) UTF-8 version of suject as result (before processing)
3. for each match, do inline replacement into result (UTF-8 returned by the callback and result is in UTF-8, no conversion)
4. result is already in UTF-8 (before returning)

$string = str_repeat('Hayır İyi', 1);

1 iteration

real 0m0.021s user 0m0.014s sys 0m0.006s

1 000 000 iteration

real 0m14.365s user 0m14.351s sys 0m0.005s

$string = str_repeat('Hayır İyi', 10000);

1 iteration

real 0m5.595s user 0m5.585s sys 0m0.006s

10 000 iteration

too long (more than 16h ?)