Evaluate the replacement as PHP code, bring G2 hook_filter implementation up to speed with Drupal 6

Needs to be converted to the "new" format no longer depending on invalid UTF8 sequences.

See #267484: Incorrect filter documentation for details.

It probably means something like this:

     case 'prepare':
-      // Note: we use the bytes 0xFE and 0xFF to replace < > during the filtering process.
-      // These bytes are not valid in UTF-8 data and thus least likely to cause problems.
-      $text = preg_replace('@<dfn>(.+?)</dfn>@s', "\xFEdfn\xFF\\1\xFE/dfn\xFF", $text);
+      $text = preg_replace('@<dfn>(.+?)</dfn>@s', "[g2-dfn]\\1[/g2-dfn]", $text);
       return $text;
 
     case "process":
-      $text = preg_replace('@\xFEdfn\xFF(.+?)\xFE/dfn\xFF@se', "_g2_filter_process('$1')", $text);
+      $text = preg_replace('@\[g2-dfn\](.+?)\[/g2-dfn\]@se', "_g2_filter_process('$1')", $text);
       return $text;
 

Or maybe a simpler form in just one step, without prepare. The process callback seems to be unavoidable, though.

Looks good to me. I've included it in the dev. version (off-CVS for now).

Included in today's release, with a zillion other changes.