[OT] Unbreak Thai shaping and fallback Arabic shaping

The merger of normalizer and glyph-mapping broke shapers that
modified text stream.  Unbreak them by adding a new preprocess_text
shaping stage that happens before normalizing/cmap and disallow
setup_mask modification of actual text.

src/hb-ot-shape-complex-arabic.cc

@@ -274,12 +274,8 @@ arabic_fallback_shape (hb_font_t *font, hb_buffer_t *buffer)
 }
 
 static void
-setup_masks_arabic (const hb_ot_shape_plan_t *plan,
-		    hb_buffer_t              *buffer,
-		    hb_font_t                *font)
+arabic_joining (hb_buffer_t *buffer)
 {
-  const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data;
-
   unsigned int count = buffer->len;
   unsigned int prev = 0, state = 0;
 
@@ -305,14 +301,37 @@ setup_masks_arabic (const hb_ot_shape_plan_t *plan,
     state = entry->next_state;
   }
 
-  if (likely (!arabic_plan->do_fallback)) {
-    /* Has OpenType tables */
-    for (unsigned int i = 0; i < count; i++)
-      buffer->info[i].mask |= arabic_plan->mask_array[buffer->info[i].arabic_shaping_action()];
-  } else
+  HB_BUFFER_DEALLOCATE_VAR (buffer, arabic_shaping_action);
+}
+
+static void
+preprocess_text_arabic (const hb_ot_shape_plan_t *plan,
+			hb_buffer_t              *buffer,
+			hb_font_t                *font)
+{
+  const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data;
+
+  if (unlikely (arabic_plan->do_fallback))
+  {
+    arabic_joining (buffer);
     arabic_fallback_shape (font, buffer);
+  }
+}
 
-  HB_BUFFER_DEALLOCATE_VAR (buffer, arabic_shaping_action);
+static void
+setup_masks_arabic (const hb_ot_shape_plan_t *plan,
+		    hb_buffer_t              *buffer,
+		    hb_font_t                *font)
+{
+  const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data;
+
+  if (likely (!arabic_plan->do_fallback))
+  {
+    arabic_joining (buffer);
+    unsigned int count = buffer->len;
+    for (unsigned int i = 0; i < count; i++)
+      buffer->info[i].mask |= arabic_plan->mask_array[buffer->info[i].arabic_shaping_action()];
+  }
 }
 
 const hb_ot_complex_shaper_t _hb_ot_complex_shaper_arabic =
@@ -322,6 +341,7 @@ const hb_ot_complex_shaper_t _hb_ot_complex_shaper_arabic =
   NULL, /* override_features */
   data_create_arabic,
   data_destroy_arabic,
+  preprocess_text_arabic,
   NULL, /* normalization_preference */
   setup_masks_arabic,
   true, /* zero_width_attached_marks */

src/hb-ot-shape-complex-indic.cc

@@ -1134,6 +1134,7 @@ const hb_ot_complex_shaper_t _hb_ot_complex_shaper_indic =
   override_features_indic,
   data_create_indic,
   data_destroy_indic,
+  NULL, /* preprocess_text */
   NULL, /* normalization_preference */
   setup_masks_indic,
   false, /* zero_width_attached_marks */

src/hb-ot-shape-complex-misc.cc

 /*
- * Copyright © 2010  Google, Inc.
+ * Copyright © 2010,2012  Google, Inc.
  *
  *  This is part of HarfBuzz, a text shaping library.
  *
@@ -90,6 +90,7 @@ const hb_ot_complex_shaper_t _hb_ot_complex_shaper_default =
   NULL, /* override_features */
   NULL, /* data_create */
   NULL, /* data_destroy */
+  NULL, /* preprocess_text */
   normalization_preference_default,
   NULL, /* setup_masks */
   true, /* zero_width_attached_marks */
@@ -99,9 +100,9 @@ const hb_ot_complex_shaper_t _hb_ot_complex_shaper_default =
 /* Thai / Lao shaper */
 
 static void
-setup_masks_thai (const hb_ot_shape_plan_t *plan HB_UNUSED,
-		  hb_buffer_t              *buffer,
-		  hb_font_t                *font HB_UNUSED)
+preprocess_text_thai (const hb_ot_shape_plan_t *plan HB_UNUSED,
+		      hb_buffer_t              *buffer,
+		      hb_font_t                *font HB_UNUSED)
 {
   /* The following is NOT specified in the MS OT Thai spec, however, it seems
    * to be what Uniscribe and other engines implement.  According to Eric Muller:
@@ -200,7 +201,8 @@ const hb_ot_complex_shaper_t _hb_ot_complex_shaper_thai =
   NULL, /* override_features */
   NULL, /* data_create */
   NULL, /* data_destroy */
+  preprocess_text_thai,
   NULL, /* normalization_preference */
-  setup_masks_thai,
+  NULL, /* setup_masks */
   true, /* zero_width_attached_marks */
 };

src/hb-ot-shape-complex-private.hh

@@ -56,7 +56,6 @@ struct hb_ot_complex_shaper_t
   /* collect_features()
    * Called during shape_plan().
    * Shapers should use plan->map to add their features and callbacks.
-   * May be NULL.
    */
   void (*collect_features) (hb_ot_shape_planner_t *plan);
 
@@ -64,7 +63,6 @@ struct hb_ot_complex_shaper_t
    * Called during shape_plan().
    * Shapers should use plan->map to override features and add callbacks after
    * common features are added.
-   * May be NULL.
    */
   void (*override_features) (hb_ot_shape_planner_t *plan);
 
@@ -73,7 +71,7 @@ struct hb_ot_complex_shaper_t
    * Called at the end of shape_plan().
    * Whatever shapers return will be accessible through plan->data later.
    * If NULL is returned, means a plan failure.
-   * May be NULL. */
+   */
   void *(*data_create) (const hb_ot_shape_plan_t *plan);
 
   /* data_destroy()
@@ -83,6 +81,16 @@ struct hb_ot_complex_shaper_t
    * May be NULL. */
   void (*data_destroy) (void *data);
 
+
+  /* preprocess_text()
+   * Called during shape().
+   * Shapers can use to modify text before shaping starts.
+   */
+  void (*preprocess_text) (const hb_ot_shape_plan_t *plan,
+			   hb_buffer_t              *buffer,
+			   hb_font_t                *font);
+
+
   /* normalization_preference()
    * Called during shape().
    */
@@ -92,6 +100,7 @@ struct hb_ot_complex_shaper_t
   /* setup_masks()
    * Called during shape().
    * Shapers should use map to get feature masks and set on buffer.
+   * Shapers may NOT modify characters.
    */
   void (*setup_masks) (const hb_ot_shape_plan_t *plan,
 		       hb_buffer_t              *buffer,

src/hb-ot-shape.cc

@@ -319,6 +319,9 @@ hb_ot_map_glyphs_fast (hb_buffer_t  *buffer)
 static inline void
 hb_ot_substitute_default (hb_ot_shape_context_t *c)
 {
+  if (c->plan->shaper->preprocess_text)
+    c->plan->shaper->preprocess_text (c->plan, c->buffer, c->font);
+
   hb_ot_mirror_chars (c);
 
   HB_BUFFER_ALLOCATE_VAR (c->buffer, glyph_index);