Optimization in FcLangSetIndex was broken, occasionally returning a pointer

    to the wrong location on miss

diff --git a/src/fclang.c b/src/fclang.c
index b0a0894aa9997f53d9315cc4711f68fb6b80fef3..53f2b5982b47400ca6a2ce986fb91b0953cddb17 100644 (file)
--- a/src/fclang.c
+++ b/src/fclang.c
@@ -262,6 +262,7 @@ FcLangSetIndex (const FcChar8 *lang)
     int            low, high, mid = 0;
     int            cmp = 0;
     FcChar8 firstChar = FcToLower(lang[0]); 
+    FcChar8 secondChar = firstChar ? FcToLower(lang[1]) : '\0';
     
     if (firstChar < 'a')
     {
@@ -290,22 +291,14 @@ FcLangSetIndex (const FcChar8 *lang)
        else
        {   /* fast path for resolving 2-letter languages (by far the most common) after
             * finding the first char (probably already true because of the hash table) */
-           FcChar8 secondChar = FcToLower(lang[1]);
-           if (fcLangCharSets[mid].lang[1] > secondChar) /* check second chars */
+           cmp = fcLangCharSets[mid].lang[1] - secondChar;
+           if (cmp == 0 && 
+               (fcLangCharSets[mid].lang[2] != '\0' || 
+                lang[2] != '\0'))
            {
-               high = mid - 1;
-               continue;
+               cmp = FcStrCmpIgnoreCase(fcLangCharSets[mid].lang+2, 
+                                        lang+2);
            }
-           else if (fcLangCharSets[mid].lang[1] < secondChar)
-           {
-               low = mid + 1;
-               continue;
-           }
-           else if (fcLangCharSets[mid].lang[2] == '\0' && lang[2] == '\0')
-               return mid;
-
-           else /* identical through the first two charcters, but at least one string didn't end there */
-               cmp = FcStrCmpIgnoreCase(fcLangCharSets[mid].lang+2, lang+2);
        }
        if (cmp == 0)
            return mid;