First pass at building a cmap

src/hb-ot-cmap-table.hh

@@ -254,6 +254,8 @@ struct CmapSubtableFormat10 : CmapSubtableTrimmed<HBUINT32 > {};
 template <typename T>
 struct CmapSubtableLongSegmented
 {
+  friend struct cmap;
+
   inline bool get_glyph (hb_codepoint_t codepoint, hb_codepoint_t *glyph) const
   {
     int i = groups.bsearch (codepoint);
@@ -269,6 +271,20 @@ struct CmapSubtableLongSegmented
     return_trace (c->check_struct (this) && groups.sanitize (c));
   }
 
+  inline bool serialize(hb_serialize_context_t *context,
+                        unsigned int group_count,
+                        Supplier<CmapSubtableLongGroup> &group_supplier)
+  {
+    TRACE_SERIALIZE (this);
+    if (unlikely(!context->extend_min (*this))) return_trace (false);
+    if (unlikely(!groups.serialize(context, group_count))) return_trace (false);
+    for (unsigned int i = 0; i < group_count; i++) {
+      const CmapSubtableLongGroup &group = group_supplier[i];
+      memcpy(&groups[i], &group, sizeof(group));
+    }
+    return true;
+  }
+
   protected:
   HBUINT16	format;		/* Subtable format; set to 12. */
   HBUINT16	reservedZ;	/* Reserved; set to 0. */
@@ -505,15 +521,15 @@ struct cmap
 		  encodingRecord.sanitize (c, this));
   }
 
-  inline bool subset (hb_subset_plan_t *plan, hb_face_t *source, hb_face_t *dest) const
+  inline void populate_groups(hb_auto_array_t<hb_codepoint_t> &codepoints,
+                              hb_auto_array_t<CmapSubtableLongGroup> *groups) const
   {
-    hb_auto_array_t<CmapSubtableLongGroup> groups;
     CmapSubtableLongGroup *group = nullptr;
-    for (unsigned int i = 0; i < plan->codepoints.len; i++) {
-      hb_codepoint_t cp = plan->codepoints[i];
+    for (unsigned int i = 0; i < codepoints.len; i++) {
+      hb_codepoint_t cp = codepoints[i];
       if (!group)
       {
-        group = groups.push();
+        group = groups->push();
         group->startCharCode.set(cp);
         group->endCharCode.set(cp);
         group->glyphID.set(i);  // index in codepoints is new gid
@@ -527,13 +543,84 @@ struct cmap
     }
 
     DEBUG_MSG(SUBSET, nullptr, "cmap");
-    for (unsigned int i = 0; i < groups.len; i++) {
-      CmapSubtableLongGroup& group = groups[i];
-      DEBUG_MSG(SUBSET, nullptr, "  %d: U+%04X-U+%04X, first gid %d", i, (uint32_t) group.startCharCode, (uint32_t) group.endCharCode, (uint32_t) group.glyphID);
+    for (unsigned int i = 0; i < groups->len; i++) {
+      CmapSubtableLongGroup& group = (*groups)[i];
+      DEBUG_MSG(SUBSET, nullptr, "  %d: U+%04X-U+%04X, gid %d-%d", i, (uint32_t) group.startCharCode, (uint32_t) group.endCharCode, (uint32_t) group.glyphID, (uint32_t) group.glyphID + ((uint32_t) group.endCharCode - (uint32_t) group.startCharCode));
+    }
+  }
+
+  hb_bool_t _subset (hb_auto_array_t<CmapSubtableLongGroup> &groups,
+                     size_t dest_sz,
+                     void *dest) const
+  {
+    hb_serialize_context_t context(dest, dest_sz);
+
+    OT::cmap *cmap = context.start_serialize<OT::cmap> ();
+    if (unlikely(!context.extend_min(*cmap)))
+    {
+      return false;
+    }
+
+    cmap->version.set(0);
+
+    if (unlikely(!cmap->encodingRecord.serialize(&context, /* numTables */ 1)))
+    {
+      return false;
     }
+
+    EncodingRecord &rec = cmap->encodingRecord[0];
+    rec.platformID.set (3); // Windows
+    rec.encodingID.set (1); // Unicode BMP
+
+    CmapSubtable &subtable = rec.subtable.serialize(&context, &rec.subtable);
+    subtable.u.format.set(12);
+
+    CmapSubtableFormat12 &format12 = subtable.u.format12;
+    format12.format.set(12);
+    format12.reservedZ.set(0);
+
+    OT::Supplier<CmapSubtableLongGroup> group_supplier  (&groups[0], groups.len, sizeof (CmapSubtableLongGroup));
+    if (unlikely(!format12.serialize(&context, groups.len, group_supplier)))
+    {
+      return false;
+    }
+
+    context.end_serialize ();
     return true;
   }
 
+  hb_blob_t * subset (hb_subset_plan_t *plan, hb_face_t *source) const
+  {
+    hb_auto_array_t<CmapSubtableLongGroup> groups;
+
+    populate_groups(plan->codepoints, &groups);
+
+    // We now know how big our blob needs to be
+    // TODO use APIs from the structs to get size?
+    size_t dest_sz = 4 // header
+                   + 8 // 1 EncodingRecord
+                   + 16 // Format 12 header
+                   + 12 * groups.len; // SequentialMapGroup records
+    void *dest = calloc(dest_sz, 1);
+    if (unlikely(!dest)) {
+      DEBUG_MSG(SUBSET, nullptr, "Unable to alloc %ld for cmap subset output", dest_sz);
+      return nullptr;
+    }
+
+    if (unlikely(!_subset(groups, dest_sz, dest)))
+    {
+      free(dest);
+      return nullptr;
+    }
+
+    // all done, write the blob into dest
+    return hb_blob_create((const char *)dest, 
+                          dest_sz,
+                          HB_MEMORY_MODE_READONLY,
+                          /* userdata */ nullptr,
+                          free);
+  }
+
   struct accelerator_t
   {
     inline void init (hb_face_t *face)

src/hb-subset.cc

@@ -108,19 +108,19 @@ hb_subset_input_destroy(hb_subset_input_t *subset_input)
 }
 
 template<typename TableType>
-hb_bool_t
-subset (hb_subset_plan_t *plan, hb_face_t *source, hb_face_t *dest)
+hb_blob_t *
+_subset (hb_subset_plan_t *plan, hb_face_t *source)
 {
     OT::Sanitizer<TableType> sanitizer;
-    hb_blob_t *table_blob = sanitizer.sanitize (source->reference_table (TableType::tableTag));
-    if (unlikely(!table_blob)) {
+    hb_blob_t *source_blob = sanitizer.sanitize (source->reference_table (TableType::tableTag));
+    if (unlikely(!source_blob)) {
       DEBUG_MSG(SUBSET, nullptr, "Failed to reference table for tag %d", TableType::tableTag);
-      return false;
+      return nullptr;
     }
-    const TableType *table = OT::Sanitizer<TableType>::lock_instance (table_blob);
-    hb_bool_t result = table->subset(plan, source, dest);
+    const TableType *table = OT::Sanitizer<TableType>::lock_instance (source_blob);
+    hb_blob_t *result = table->subset(plan, source);
 
-    hb_blob_destroy (table_blob);
+    hb_blob_destroy (source_blob);
 
     hb_tag_t tag = TableType::tableTag;
     DEBUG_MSG(SUBSET, nullptr, "Subset %c%c%c%c %s", HB_UNTAG(tag), result ? "success" : "FAILED!");
@@ -242,7 +242,6 @@ hb_subset_face_add_table (hb_face_t *face, hb_tag_t tag, hb_blob_t *blob)
     return false;
 
   hb_subset_face_data_t *data = (hb_subset_face_data_t *) face->user_data;
-
   hb_subset_face_data_t::table_entry_t *entry = data->tables.push ();
   if (unlikely (!entry))
     return false;
@@ -306,10 +305,12 @@ bool
 _subset_table (hb_subset_plan_t *plan,
                hb_face_t        *source,
                hb_tag_t          tag,
-               hb_blob_t        *table_blob,
+               hb_blob_t        *source_blob,
                hb_face_t        *dest)
 {
   // TODO (grieger): Handle updating the head table (loca format + num glyphs)
+  DEBUG_MSG(SUBSET, nullptr, "begin subset %c%c%c%c", HB_UNTAG(tag));
+  hb_blob_t *dest_blob;
   switch (tag) {
     case HB_OT_TAG_glyf:
       return _subset_glyf (plan, source, dest);
@@ -320,14 +321,16 @@ _subset_table (hb_subset_plan_t *plan,
       // SKIP loca, it's handle by glyf
       return true;
     case HB_OT_TAG_cmap:
-      // TODO(rsheeter): remove hb_subset_face_add_table
-      //                 once cmap subsetting works.
-      hb_subset_face_add_table (dest, tag, table_blob);
-      return subset<const OT::cmap> (plan, source, dest);
+      dest_blob = _subset<const OT::cmap> (plan, source);
+      break;
     default:
-      // Default action, copy table as is.
-      return hb_subset_face_add_table (dest, tag, table_blob);
+      dest_blob = source_blob;
+      break;
   } 
+  DEBUG_MSG(SUBSET, nullptr, "subset %c%c%c%c %s", HB_UNTAG(tag), dest_blob ? "ok" : "FAILED");
+  if (unlikely(!dest_blob)) return false;
+  if (unlikely(!hb_subset_face_add_table (dest, tag, dest_blob))) return false;
+  return true;
 }
 
 /**