Development

...
for ($i = 0; $i <= mb_strlen($str, 'UTF-8'); $i += 10000)
...
mb_substr($str, $i, 1000, 'UTF-8'),
...

?

Just to note, the problem with this solution is it requires mbstring extension, which isn't enabled by default in a stock PHP install.

Test:
Post text 661 200 bytes
utf8_bad_strip() - script was killed after 30s
my variant utf8_bad_strip_improved() - [ Generated in 3.587 seconds, 7 queries executed - Memory usage: 15.75 MiB (Peak: 18.8 MiB) ]

Since this affects php-utf8, please also make an issue at https://github.com/FSX/php-utf8

new utf8_bad_clean == old utf8_bad_strip
The problem remains.

I haven't looked into it yet. I just said: create an issue at the php-utf8 repository.

So I did some work on this today...

@Pierre: While your solution is much more efficient than the current one, it relies on the mb_string extension, which we cannot assume is available. The native utf8_substr function in the utf8 library fails, as it fails when presented with invalid unicode. I'm also seeing segfaults (?!) when attempting to use that function on a non-invalid ASCII-only string.

@Visman: Where did that function come from? I don't see any direct reference to it on the linked wikipedia article. It appears to work much better, and doesn't rely on any extensions.

In the below "improved" is Pierre's solution, "improved (compatible)" is Pierre's solution using utf8 library instead of mb_string, and "improved (test)" is Visman's solution.

19:35:59 jamie@penguin ~ php test.php
Original took 499.747825 seconds
Improved took 3.285618 seconds, result passed
Improved (compatible) took 5.084029 seconds, result failed
Improved (test) took 0.702509 seconds, result passed

Here's a slightly tidied up version, but I don't want to commit anything yet if the licensing is going to be an issue...

function utf8_bad_strip_improved_test($str, $replace = '') { // (C) SiMM, based on ru.wikipedia.org/wiki/Unicode
	$result = '';

	$strlen = strlen($str);
	for ($i = 0; $i < $strlen;) {
		$tmp = $str[$i++];
		$ch = ord($tmp);

		if ($ch > 0x7F) {
			if ($ch < 0xC0) {
				$result .= $replace;
				continue;
			}
			else if ($ch < 0xE0) $di = 1;
			else if ($ch < 0xF0) $di = 2;
			else if ($ch < 0xF8) $di = 3;
			else if ($ch < 0xFC) $di = 4;
			else if ($ch < 0xFE) $di = 5;
			else {
				$result .= $replace;
				continue;
			}

			for ($j = 0; $j < $di; $j++) {
				$tmp .= $ch = $str[$i + $j];
				$ch = ord($ch);

				if ($ch < 0x80 || $ch > 0xBF) {
					$result .= $replace;
					continue 2;
				}
			}

			$i += $di;
		}

		$result .= $tmp;
	}

	return $result;
}

Here's a version without the replace argument. It also passes the tests:

function badClean($str)
{
	$UTF8_GOOD =
		'[\x00-\x7F]'.                           # ASCII (including control chars)
		'|[\xC2-\xDF][\x80-\xBF]'.               # Non-overlong 2-byte
		'|\xE0[\xA0-\xBF][\x80-\xBF]'.           # Excluding overlongs
		'|[\xE1-\xEC\xEE\xEF][\x80-\xBF]{2}'.    # Straight 3-byte
		'|\xED[\x80-\x9F][\x80-\xBF]'.           # Excluding surrogates
		'|\xF0[\x90-\xBF][\x80-\xBF]{2}'.        # Planes 1-3
		'|[\xF1-\xF3][\x80-\xBF]{3}'.            # Planes 4-15
		'|\xF4[\x80-\x8F][\x80-\xBF]{2}';        # Plane 16

	preg_match_all('%(?:'.$UTF8_GOOD.')+%', $str, $clean_pieces);
	return implode('', $clean_pieces[0]);
}

I was also thinking about mb_strlen. It doesn't count invalid characters so chopping the text in pieces and process it piece by piece is not going to work properly, because you aren't taking the invalid characters into account.

I'm also getting a segfault on that when giving it a large clean ASCII string.

Also it's memory consumption is insane - using a string of size 20,000,000 it is uses 2.6Gb memory and takes 14 seconds. The modified one Visman posted takes 14 seconds and uses 30Mb memory.

Can you show your test code and data?

I don't think you will find direct reference to it on the linked wikipedia article. Googled "SiMM, based on ru.wikipedia.org/wiki/Unicode" and found following postings by SiMM:
http://phpclub.ru/talk/threads/iconv-%D … f-8.24038/
http://forum.ru-board.com/topic.cgi?for … 0&start=20

I've implemented a version of SiMM's code from scratch and commit it. I'll pull it to fluxbb.org now. Can someone please give it a look over and testing when they get a chance?

Cheers.