Guess that mac roman names with lots of high bits are actually SJIS.

Many Japanese fonts incorrectly include names tagged as Roman encoding and
English language which are actually Japanese names in the SJIS encoding.
Guess that names with a large number of high bits set are SJIS encoded
Japanese names rather than English names.

diff --git a/src/fcfreetype.c b/src/fcfreetype.c
index f85e2f8931eae1bc8b2305829ce355a9e6a08c37..082d17bfb8865010c5ac38fdda737f572ca11f62 100644 (file)
--- a/src/fcfreetype.c
+++ b/src/fcfreetype.c
@@ -560,6 +560,28 @@ FcFontCapabilities(FT_Face face);
 #include <iconv.h>
 #endif
 
+/*
+ * A shift-JIS will have many high bits turned on
+ */
+static FcBool
+FcLooksLikeSJIS (FcChar8 *string, int len)
+{
+    int            nhigh = 0, nlow = 0;
+
+    while (len-- > 0)
+    {
+       if (*string++ & 0x80) nhigh++;
+       else nlow++;
+    }
+    /*
+     * Heuristic -- if more than 1/3 of the bytes have the high-bit set,
+     * this is likely to be SJIS and not ROMAN
+     */
+    if (nhigh * 2 > nlow)
+       return FcTrue;
+    return FcFalse;
+}
+
 static FcChar8 *
 FcSfntNameTranscode (FT_SfntName *sname)
 {
@@ -580,24 +602,35 @@ FcSfntNameTranscode (FT_SfntName *sname)
     fromcode = fcFtEncoding[i].fromcode;
 
     /*
-     * "real" Mac language IDs are all less than 150.
-     * Names using one of the MS language IDs are assumed
-     * to use an associated encoding (Yes, this is a kludge)
+     * Many names encoded for TT_PLATFORM_MACINTOSH are broken
+     * in various ways. Kludge around them.
      */
-    if (!strcmp (fromcode, FC_ENCODING_MAC_ROMAN) &&
-       sname->language_id >= 0x100)
+    if (!strcmp (fromcode, FC_ENCODING_MAC_ROMAN))
     {
-       int     f;
+       if (sname->language_id == TT_MAC_LANGID_ENGLISH &&
+           FcLooksLikeSJIS (sname->string, sname->string_len))
+       {
+           fromcode = "SJIS";
+       }
+       else if (sname->language_id >= 0x100)
+       {
+           /*
+            * "real" Mac language IDs are all less than 150.
+            * Names using one of the MS language IDs are assumed
+            * to use an associated encoding (Yes, this is a kludge)
+            */
+           int f;
 
-       fromcode = NULL;
-       for (f = 0; f < NUM_FC_MAC_ROMAN_FAKE; f++)
-           if (fcMacRomanFake[f].language_id == sname->language_id)
-           {
-               fromcode = fcMacRomanFake[f].fromcode;
-               break;
-           }
-       if (!fromcode)
-           return 0;
+           fromcode = NULL;
+           for (f = 0; f < NUM_FC_MAC_ROMAN_FAKE; f++)
+               if (fcMacRomanFake[f].language_id == sname->language_id)
+               {
+                   fromcode = fcMacRomanFake[f].fromcode;
+                   break;
+               }
+           if (!fromcode)
+               return 0;
+       }
     }
     if (!strcmp (fromcode, "UCS-2BE") || !strcmp (fromcode, "UTF-16BE"))
     {
@@ -738,10 +771,24 @@ static const FcChar8 *
 FcSfntNameLanguage (FT_SfntName *sname)
 {
     int i;
+    FT_UShort  platform_id = sname->platform_id;
+    FT_UShort  language_id = sname->language_id;
+
+    /*
+     * Many names encoded for TT_PLATFORM_MACINTOSH are broken
+     * in various ways. Kludge around them.
+     */
+    if (platform_id == TT_PLATFORM_MACINTOSH &&
+       sname->encoding_id == TT_MAC_ID_ROMAN &&
+       FcLooksLikeSJIS (sname->string, sname->string_len))
+    {
+       language_id = TT_MAC_LANGID_JAPANESE;
+    }
+    
     for (i = 0; i < NUM_FC_FT_LANGUAGE; i++)
-       if (fcFtLanguage[i].platform_id == sname->platform_id &&
+       if (fcFtLanguage[i].platform_id == platform_id &&
            (fcFtLanguage[i].language_id == TT_LANGUAGE_DONT_CARE ||
-            fcFtLanguage[i].language_id == sname->language_id))
+            fcFtLanguage[i].language_id == language_id))
        {
            if (fcFtLanguage[i].lang[0] == '\0')
              return NULL;