Unlike preg_replace, mb_ereg_replace doesn't use separators

Exemple with preg_replace :
<?php $data = preg_replace("/[^A-Za-z0-9\.\-]/","",$data); ?>

Exemple with mb_ereg_replace :
<?php $data = mb_ereg_replace("[^A-Za-z0-9\.\-]","",$data); ?>

Notations to reference captures in the replacement string:

<?php

// (1) \\number notation: (1 to 9, not greater than 9)
echo mb_ereg_replace('(\S*) (\S*) (\S*)', '\\1 jam, \\2 juice, \\3 squash', 'apple orange lemon').'<br>'; // apple jam, orange juice, lemon squash

// (2) \k<number> notation: (also greater than 9) (also as \k'number')
echo mb_ereg_replace('(\S*) (\S*) (\S*)', '\k<1> jam, \k<2> juice, \k<3> squash', 'apple orange lemon').'<br>'; // (same as above)

// (3) \k<word> notation: (also as \k'word')
echo mb_ereg_replace('(?<word1>\S*) (?<word2>\S*) (?<word3>\S*)', '\k<word1> jam, \k<word2> juice, \k<word3> squash', 'apple orange lemon').'<br>'; // (same as above)

// Note non-named-subpatterns like "(\S*)" should not be used with named-subpatterns like "(?<word>..)" because non-named-subpatterns cannot be captured when named-subpatterns exist.

I got a pretty nasty error while trying to parse table rows(all contents were set to UTF-8) from the database for a dictionary project. The idea was to get all the rows from the first table (that is a table with bulgarian phrase in the first field, and its translation in english, french and german in the next fields). I needed to index all the bulgarian words that are found in the table to make an intelligent search. And that is where my headache started.

First of all, even with mb_strtolower() a lot of cyrillic characters went corrupted (ex: 'т,ъ,у,ф,б,г,з,ж,' etc...). After an hour of different attempts I got such a solution:

<?php

mb_internal_encoding("UTF-8");
mb_regex_encoding("UTF-8");

$rows = $db->getRows();

$contents = array();
foreach ($rows as $eachRow)
{
    $cleared = str_replace($commonWords, ' ', mb_strtolower(stripslashes($eachRow['bulgarian']), 'UTF-8' ));
    if (trim($cleared) != '') $contents[] = trim($cleared);
}    

$list = array();
foreach ($contents as $eachRow)
{
    $exploded = explode(' ', $eachRow);
    foreach ($exploded as $eachExpl)
    {
        $eachExpl = mb_ereg_replace('[^а-я ]',' ', $eachExpl);
        if (trim($eachExpl) != '') 
            if (!in_array($eachExpl, $list, true))    $list[] = trim($eachExpl);
    }
}

?>

To work properly I got to set all the internal encoding settings to UTF-8. Else the default Latin-1 got half my database with missing characters.

I am posting this solution just in case someone has encountered a similar problem. Hope it helps you in case you need something like that.

You can use \\n for capture group in replacement.
And you can NOT use $n notation (unlike preg_replace function).

Pluche's comment should REALLY be added to the documentation, preferably under the "$pattern" param description. It is crucial to using this function.

<?php
$pattern = "([あ-ん]+)[0-9]+";
$string = mb_ereg_replace($pattern, '「\\1」:\\0', $string);
?>

you can use \\n for capture group in replacement

If encoding of PHP script differs from encoding of string to be processed by mb_ereg_replace(), then you can't just write pattern in script. Both $pattern and $replacement must be converted to same encoding as string to be processed. In this example script is in UTF-8, file to be processed is in UTF-16LE encoding:

<?php
$file_encoding = 'UTF-16LE';
mb_regex_encoding( $file_encoding );

$pattern     = "aaa";
$replacement = "AAA";
$pattern_encoded     = mb_convert_encoding( $pattern,     $file_encoding, 'UTF-8' );
$replacement_encoded = mb_convert_encoding( $replacement, $file_encoding, 'UTF-8' );

$result = mb_ereg_replace( $pattern_encoded, $replacement_encoded, file_get_contents('UTF-16LE.txt') );
file_put_contents('UTF-16LE-updated.txt', $result);
?>

'i' option does not work correctly with multibyte characters. The function does not locate/replace the multibyte string if it's different case then specified on multibyte needle which is in different case.

A simple mb_str_ireplace() implementation - a faster (?) replacement for non-regexp multi-byte string replacement:

<?php
function mb_str_ireplace($co, $naCo, $wCzym)
{
    $wCzymM = mb_strtolower($wCzym);
    $coM    = mb_strtolower($co);
    $offset = 0;
    
        while(!is_bool($poz = mb_strpos($wCzymM, $coM, $offset)))
    {
        $offset = $poz + mb_strlen($naCo);
        $wCzym = mb_substr($wCzym, 0, $poz). $naCo .mb_substr($wCzym, $poz+mb_strlen($co));
        $wCzymM = mb_strtolower($wCzym);
    }
    
    return $wCzym;
}
?>

[thiago - EDITOR NOTE: This function has improvements from d-okumura [aat] fi{dot}kyd[dot]co.jp]

Since PHP 5.4, to make uppercase ou lowercase characters, or rewrite some uris, without to take care about initial encoding, the transliteration is easier (and probably the best way): see http://php.net/manual/fr/transliterator.transliterate.php and http://userguide.icu-project.org/transforms/general

For example (with create) (french text: replace all accuentued -éèàîïùç...- chars with ascii chars):
<?php
$transliterator = Transliterator::create("NFD; [:Nonspacing Mark:] Remove; NFC;");
echo $transliterator->transliterate("Héhé, ça marche !");
?>
// Result: « Hehe, ca marche ! »

To rewrite a phrase in URI (with createFromRules):
<?php
$transliterator = Transliterator::createFromRules("::Latin-ASCII; ::Lower; [^[:L:][:N:]]+ > '-';");
echo trim($transliterator->transliterate("Héhé, ça marche !"), '-');
?>
// Result : « hehe-ca-marche »

Are you looking for htmlentities() for multibyte strings? This might help you - it just replace <, >, ", '

<?php
/**
 *  Multibyte equivalent for htmlentities() [lite version :)]
 *
 * @param string $str
 * @param string $encoding
 * @return string
 **/
function mb_htmlentities($str, $encoding = 'utf-8') {
    mb_regex_encoding($encoding);
    $pattern = array('<', '>', '"', '\'');
    $replacement = array('&lt;', '&gt;', '&quot;', '&#39;');
    for ($i=0; $i<sizeof($pattern); $i++) {
        $str = mb_ereg_replace($pattern[$i], $replacement[$i], $str);
    }
    return $str;
}
?>

Regarding the mb_str_ireplace() function: I benchmarked it against mb_eregi_replace() for single-character substitution, and it was significantly slower. Despite avoiding the ereg call, I think the while loop ends slowing you down too much for this to be practical.

If you want to replace characters like "ä" or "ø" you can use mb_ereg_replace, but it is very slow. str_replace is much faster and also works with characters like "ä" or "ø"!

I think this has something to with the fact that str_replace works on byte level and does not care about characters.
I hope that can help.

To selectively uppercase parts of a string via mb_eregi_replace

    $str = mb_eregi_replace('\b([0-9]{1,4}[a-z]{1,2})\b', "strtoupper
('\\1')", $str, 'e');

Full example, how to fix an address manually typed, uppercasing the first letter of a words and keeping uppercase roman numerals and the letters A,B,C after the house number):

function ucAddress($str) {
// first lowercase all and use the default ucwords
    $str = ucwords(strtolower($str));
// let's fix the default ucwords...
// uppercase letters after house number (was lowercased by the strtolower above)
    $str = mb_eregi_replace('\b([0-9]{1,4}[a-z]{1,2})\b', "strtoupper
('\\1')", $str, 'e');
// the same for roman numerals
    $str = mb_eregi_replace('\bM{0,4}(CM|CD|D?C{0,3})(XC|XL|L?X{0,3})(IX|IV|V?I{0,3})\b', "strtoupper('\\0')", $str, 'e');
    return $str;
}

To selectively uppercase parts of a string via mb_eregi_replace

    $str = mb_eregi_replace('\b([0-9]{1,4}[a-z]{1,2})\b', "strtoupper
('\\1')", $str, 'e');

Full example, how to fix an address manually typed, uppercasing the first letter of a words and keeping uppercase roman numerals and the letters A,B,C after the house number):

function ucAddress($str) {
// first lowercase all and use the default ucwords
    $str = ucwords(strtolower($str));
// let's fix the default ucwords...
// uppercase letters after house number (was lowercased by the strtolower above)
    $str = mb_eregi_replace('\b([0-9]{1,4}[a-z]{1,2})\b', "strtoupper
('\\1')", $str, 'e');
// the same for roman numerals
    $str = mb_eregi_replace('\bM{0,4}(CM|CD|D?C{0,3})(XC|XL|L?X{0,3})(IX|IV|V?I{0,3})\b', "strtoupper('\\0')", $str, 'e');
    return $str;
}

Dr. Marco Marsala
Network Solution srl
http://www.realizzazionesitigenova.it

well, if you just calculated the length of the find and replace strings once instead of on every loop, it would likely speed it up a lot.

As trng mentioned before you can use \\n for replacement but NOT \\\\n as mentioned in preg_replace docs. So string definition will be like:
$str = '\\1';