in order to ensure that the bytecode interpretation is exactly

equivalent to the one in FT 1.4, moved some code from the
old version of FreeType in order to compute vector normalization
a bit differently

src/truetype/ttinterp.c

@@ -825,6 +825,22 @@
   }
 
 
+#ifdef FT_CONFIG_OPTION_OLD_CALCS
+
+  static TT_F26Dot6  Norm( TT_F26Dot6  X, TT_F26Dot6  Y )
+  {
+    FT_Int64       T1, T2;
+
+    MUL_64( X, X, T1 );
+    MUL_64( Y, Y, T2 );
+
+    ADD_64( T1, T2, T1 );
+
+    return (TT_F26Dot6)SQRT_64( T1 );
+  }
+#endif
+
+
   /*************************************************************************/
   /*                                                                       */
   /* Before an opcode is executed, the interpreter verifies that there are */
@@ -1194,7 +1210,7 @@
     else
     {
       TT_Long  x, y;
-#if 0
+#ifdef FT_CONFIG_OPTION_OLD_CALCS
       x = TT_MULDIV( CUR.GS.projVector.x, CUR.tt_metrics.x_ratio, 0x4000 );
       y = TT_MULDIV( CUR.GS.projVector.y, CUR.tt_metrics.y_ratio, 0x4000 );
       CUR.tt_metrics.ratio = Norm( x, y );
@@ -2126,6 +2142,98 @@
   /*    In case Vx and Vy are both zero, Normalize() returns SUCCESS, and  */
   /*    R is undefined.                                                    */
   /*                                                                       */
+
+#ifdef FT_CONFIG_OPTION_OLD_CALCS
+  static TT_Bool  Normalize( EXEC_OP_ TT_F26Dot6      Vx,
+                                      TT_F26Dot6      Vy,
+                                      TT_UnitVector*  R )
+  {
+    TT_F26Dot6  W;
+    TT_Bool     S1, S2;
+
+    if ( ABS( Vx ) < 0x10000L && ABS( Vy ) < 0x10000L )
+    {
+      Vx *= 0x100;
+      Vy *= 0x100;
+
+      W = Norm( Vx, Vy );
+
+      if ( W == 0 )
+      {
+        /* XXX : UNDOCUMENTED! It seems that it's possible to try  */
+        /*       to normalize the vector (0,0). Return immediately */
+        return SUCCESS;
+      }
+
+      R->x = (TT_F2Dot14)FT_MulDiv( Vx, 0x4000L, W );
+      R->y = (TT_F2Dot14)FT_MulDiv( Vy, 0x4000L, W );
+
+      return SUCCESS;
+    }
+
+    W = Norm( Vx, Vy );
+
+    Vx = FT_MulDiv( Vx, 0x4000L, W );
+    Vy = FT_MulDiv( Vy, 0x4000L, W );
+
+    W = Vx * Vx + Vy * Vy;
+
+    /* Now, we want that Sqrt( W ) = 0x4000 */
+    /* Or 0x1000000 <= W < 0x1004000        */
+
+    if ( Vx < 0 )
+    {
+      Vx = -Vx;
+      S1 = TRUE;
+    }
+    else
+      S1 = FALSE;
+
+    if ( Vy < 0 )
+    {
+      Vy = -Vy;
+      S2 = TRUE;
+    }
+    else
+      S2 = FALSE;
+
+    while ( W < 0x1000000L )
+    {
+      /* We need to increase W, by a minimal amount */
+      if ( Vx < Vy )
+        Vx++;
+      else
+        Vy++;
+
+      W = Vx * Vx + Vy * Vy;
+    }
+
+    while ( W >= 0x1004000L )
+    {
+      /* We need to decrease W, by a minimal amount */
+      if ( Vx < Vy )
+        Vx--;
+      else
+        Vy--;
+
+      W = Vx * Vx + Vy * Vy;
+    }
+
+    /* Note that in various cases, we can only  */
+    /* compute a Sqrt(W) of 0x3FFF, eg. Vx = Vy */
+
+    if ( S1 )
+      Vx = -Vx;
+
+    if ( S2 )
+      Vy = -Vy;
+
+    R->x = (TT_F2Dot14)Vx;   /* Type conversion */
+    R->y = (TT_F2Dot14)Vy;   /* Type conversion */
+
+    return SUCCESS;
+  }
+#else  
   static
   TT_Bool  Normalize( EXEC_OP_ TT_F26Dot6      Vx,
                                TT_F26Dot6      Vy,
@@ -2203,9 +2311,42 @@
       R->y = (TT_F2Dot14)TT_MULDIV( Vy >> shift, 0x4000, d );
     }
 
+    {
+      TT_ULong  x, y, w;
+      TT_Int    sx, sy;
+
+      sx = ( R->x >= 0 ? 1 : -1 );
+      sy = ( R->y >= 0 ? 1 : -1 );
+      x  = (TT_ULong)sx*R->x;
+      y  = (TT_ULong)sy*R->y;
+      
+      w = x*x+y*y;
+
+      /* we now want to adjust (x,y) in order to have sqrt(w) == 0x4000 */
+      /* which means 0x1000000 <= w < 0x1004000                         */
+      while ( w <= 0x10000000L )
+      {
+        /* increment the smallest coordinate */
+        if ( x < y )  x++;
+                 else y++;
+                 
+        w = x*x+y*y;
+      }
+      
+      while ( w >= 0x10040000L )
+      {
+        /* decrement the smallest coordinate */
+        if ( x < y )  x--;
+                 else y--;
+        w = x*x+y*y;
+      }
+
+      R->x = sx*x;
+      R->y = sy*y;
+    }
     return SUCCESS;
   }
-
+#endif
 
   /*************************************************************************/
   /*                                                                       */