Now we support both uppercase and lowercase lang tags, and the separator

can be both the hypen and the underscore characters.
Everything is converted to Moodle's (lowercase with hypen) langs.

And the filter returns the full text if some exception happens (bad build
lang tags). Actually it was returning an empty string on error.

Should we be a stricter allowing only the correct langs????

filter/multilang/filter.php

@@ -69,7 +69,7 @@ function multilang_filter($courseid, $text) {
     }
 
 /// Break the text into lang sections
-    $search = '/<(?:lang|span) lang="([a-zA-Z_]*)".*?>(.+?)<\/(?:lang|span)>/is';
+    $search = '/<(?:lang|span) lang="([a-zA-Z_-]*)".*?>(.+?)<\/(?:lang|span)>/is';
     preg_match_all($search,$text,$list_of_langs);
 
 /// Get the existing sections langs
@@ -77,6 +77,9 @@ function multilang_filter($courseid, $text) {
     $bestkey   = 0;
     //Iterate
     foreach ($list_of_langs[1] as $key => $lang) {
+        //Normalize: Moodle's langs are always lowercase and they use the underscore
+        //Should we be stricter?
+        $lang = strtolower(str_replace('-','_',$lang));
         $foundkey = array_search($lang, $preflangs);
         if ($foundkey !== false && $foundkey !== NULL && $foundkey < $minpref) {
             $minpref = $foundkey;
@@ -87,7 +90,10 @@ function multilang_filter($courseid, $text) {
         }
     }
 
-    $text = trim($list_of_langs[2][$bestkey]);
+    //If we have found some lang
+    if (!empty($lang)) {
+        $text = trim($list_of_langs[2][$bestkey]);
+    }
 
     return $text;
 }