MyBB

When you added \b to the badword filter you broke unicode support because PCRE "wordbreaks" doesn't support unicode - That is why I suggested using ^ and $ instead. Maybe I didn't make myself through though, it shouldn't been instead, not addition to.

Anyway this is a regression of that previous related bug report then.

First, I didn't add the \b's here. I added them somewhere else (that is bug #1028 and has nothing to do with this one), but if we figure this is the cause of the current bug, I'll remove them from there too.

Second, removing \b's doesn't solve the problem. Changing that line of code to:

$message = preg_replace("#^".$badword['badword']."$#i", "\\1".$badword['replacement']."\\2", $message);

will make the code even unable to remove English badwords.

Besides, \b is a totally different pattern from $ and ^.

\b is a word boundary (supposed to match at the beginning or end of a word) as described here: http://www.regular-expressions.info/wordboundaries.html

^ and $ match the beginning and the end of a string (and not a word) respectively) as mentioned here: http://www.regular-expressions.info/anchors.html

We can't replace one for another, unless we're working with one-word sentences.

Ah - sorry, thought we were talking about the changes to your commit. That is probably still a regression but just separate.

Do you have your PCRE compiled with unicode support?

I need to check that. From what I know, there are many other users using famous shared hosts having this problem. So either they all don't compile PCRE with Unicode support, or this has nothing to do with the problem.

For the record, using mb_eregi_replace did not solve the issue.

But if you have yours compiled with Unicode support, I can give you a test code and we can see if it works.

The trick is to have a working regex expression that doesn't care about unicode characters.

Technically we're just matching bytes to bytes with badwords so as long as you don't have anything in the regex that incurs logic that could break this (i.e. \b - word boundaries) then it would work fine.

Word boundaries break unicode support. I've seen it, it makes them go all wonky if you have multibyte characters in your regex. That's why we don't use them. If you want to check for a word boundary then check for a space on both sides.

There's a little regression in the r5081:
In my test-installation I'm replacing the word "test" with "bla".
I wrote a post with the text "test test test test test". The output was the following: "bla test bla test bla". The output with r5080 and before was this: "bla bla bla bla bla".

And there's another little problem:
This revision introduces the support of wildcards. But they didn't work without any problems. I replaced "My*BB" with "Forum". The result of a post with "My_BB" was "Forum_". I think the "_" shouldn't be there... ;-)

If this one could be tested thoroughly and with high priority, I'd appreciate it