Change 441410 had a related patch set uploaded (by C. Scott Ananian; owner: C. Scott Ananian):
[mediawiki/core@master] Don't armor french spaces before punctuation followed by word characters

https://gerrit.wikimedia.org/r/441410

Change 441410 merged by jenkins-bot:
[mediawiki/core@master] Don't armor french spaces before punctuation followed by word characters

https://gerrit.wikimedia.org/r/441410

The code could be optimized as so:

Replace:

'/(\S) (?=[?:;!%»›](?!\w))/u' => "\\1$space"

With:

'/(?<=\S) ([?:;!%»›])(?!\w)/u' => "$space\\1"

Before: The engine matches every non-space character from the beginning, then misses a following space most of the time, then backtracks.

After: The engine matches only spaces from the beginning, which are less frequent than all the non-space characters.

Change 546178 had a related patch set uploaded (by C. Scott Ananian; owner: C. Scott Ananian):
[mediawiki/core@master] Improve efficiency of french-spacing regexp

https://gerrit.wikimedia.org/r/546178

@cscott I see you have improved it even further :)

@Od1n could you do some quick benchmarks to satisfy the reviewer on the patch above?

$nb = 10000;

$text = str_repeat( 'lorem ipsum dolor sit amet', 1000 );
$space = ' ';

$t1 = microtime( true );
for ($i = $nb; $i--; ) {
    preg_replace( '/(\S) (?=[?:;!%»›](?!\w))/u', "\\1$space", $text );
}
$t2 = microtime( true );
for ($i = $nb; $i--; ) {
    preg_replace( '/(?<=\S) (?=[?:;!%»›](?!\w))/u', "$space", $text );
}
$t3 = microtime( true );

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

With $text = str_repeat( 'lorem ipsum dolor sit amet', 1000 );
Before: 1.83 seconds
After: 0.64 seconds

With $text = str_repeat( 'lorem : ipsum : dolor : sit : amet', 1000 );
Before: 5.30 seconds
After: 2.95 seconds

With $text = str_repeat( 'lorem: ipsum: dolor: sit: amet', 1000 );
Before: 2.20 seconds
After: 0.69 seconds

@thiemowmde Your benchmarks are above :)

The lookbehind assertion is perfectly fine here: it is executed only when spaces are matched, and it looks only a character back. Actually, it's this lookbehind assertion that makes the code optimized :)

executed only when spaces are matched, and it looks only a character back.

Excellent analysis. Thanks a lot for benchmarking this!

Change 546178 merged by jenkins-bot:
[mediawiki/core@master] Improve efficiency of french-spacing regexp

https://gerrit.wikimedia.org/r/546178