google-protobuf 3.2.0 → 3.9.1

data/ext/google/protobuf_c/protobuf.c

@@ -41,6 +41,7 @@ VALUE upb_def_to_ruby_obj_map;
 
 VALUE cError;
 VALUE cParseError;
+VALUE cTypeError;
 
 void add_def_obj(const void* def, VALUE value) {
   rb_hash_aset(upb_def_to_ruby_obj_map, ULL2NUM((intptr_t)def), value);
@@ -90,19 +91,24 @@ void Init_protobuf_c() {
   descriptor_instancevar_interned = rb_intern(kDescriptorInstanceVar);
   DescriptorPool_register(protobuf);
   Descriptor_register(protobuf);
+  FileDescriptor_register(protobuf);
   FieldDescriptor_register(protobuf);
   OneofDescriptor_register(protobuf);
   EnumDescriptor_register(protobuf);
   MessageBuilderContext_register(internal);
   OneofBuilderContext_register(internal);
   EnumBuilderContext_register(internal);
+  FileBuilderContext_register(internal);
   Builder_register(internal);
   RepeatedField_register(protobuf);
   Map_register(protobuf);
 
   cError = rb_const_get(protobuf, rb_intern("Error"));
   cParseError = rb_const_get(protobuf, rb_intern("ParseError"));
+  cTypeError = rb_const_get(protobuf, rb_intern("TypeError"));
 
+  rb_define_singleton_method(protobuf, "discard_unknown",
+                             Google_Protobuf_discard_unknown, 1);
   rb_define_singleton_method(protobuf, "deep_copy",
                              Google_Protobuf_deep_copy, 1);
 
@@ -110,8 +116,6 @@ void Init_protobuf_c() {
   kRubyStringASCIIEncoding = rb_usascii_encoding();
   kRubyString8bitEncoding = rb_ascii8bit_encoding();
 
-  upb_def_to_ruby_obj_map = rb_hash_new();
   rb_gc_register_address(&upb_def_to_ruby_obj_map);
-  map_parse_frames = rb_ary_new();
-  rb_gc_register_address(&map_parse_frames);
+  upb_def_to_ruby_obj_map = rb_hash_new();
 }

data/ext/google/protobuf_c/protobuf.h

@@ -40,6 +40,7 @@
 // Forward decls.
 struct DescriptorPool;
 struct Descriptor;
+struct FileDescriptor;
 struct FieldDescriptor;
 struct EnumDescriptor;
 struct MessageLayout;
@@ -47,10 +48,12 @@ struct MessageField;
 struct MessageHeader;
 struct MessageBuilderContext;
 struct EnumBuilderContext;
+struct FileBuilderContext;
 struct Builder;
 
 typedef struct DescriptorPool DescriptorPool;
 typedef struct Descriptor Descriptor;
+typedef struct FileDescriptor FileDescriptor;
 typedef struct FieldDescriptor FieldDescriptor;
 typedef struct OneofDescriptor OneofDescriptor;
 typedef struct EnumDescriptor EnumDescriptor;
@@ -60,6 +63,7 @@ typedef struct MessageHeader MessageHeader;
 typedef struct MessageBuilderContext MessageBuilderContext;
 typedef struct OneofBuilderContext OneofBuilderContext;
 typedef struct EnumBuilderContext EnumBuilderContext;
+typedef struct FileBuilderContext FileBuilderContext;
 typedef struct Builder Builder;
 
 /*
@@ -116,10 +120,10 @@ struct Descriptor {
   const upb_handlers* pb_serialize_handlers;
   const upb_handlers* json_serialize_handlers;
   const upb_handlers* json_serialize_handlers_preserve;
-  // Handlers hold type class references for sub-message fields directly in some
-  // cases. We need to keep these rooted because they might otherwise be
-  // collected.
-  VALUE typeclass_references;
+};
+
+struct FileDescriptor {
+  const upb_filedef* filedef;
 };
 
 struct FieldDescriptor {
@@ -149,24 +153,32 @@ struct EnumBuilderContext {
   VALUE enumdesc;
 };
 
+struct FileBuilderContext {
+  VALUE pending_list;
+  VALUE file_descriptor;
+  VALUE builder;
+};
+
 struct Builder {
   VALUE pending_list;
+  VALUE default_file_descriptor;
   upb_def** defs;  // used only while finalizing
 };
 
 extern VALUE cDescriptorPool;
 extern VALUE cDescriptor;
+extern VALUE cFileDescriptor;
 extern VALUE cFieldDescriptor;
 extern VALUE cEnumDescriptor;
 extern VALUE cMessageBuilderContext;
 extern VALUE cOneofBuilderContext;
 extern VALUE cEnumBuilderContext;
+extern VALUE cFileBuilderContext;
 extern VALUE cBuilder;
 
 extern VALUE cError;
 extern VALUE cParseError;
-
-extern VALUE map_parse_frames;
+extern VALUE cTypeError;
 
 // We forward-declare all of the Ruby method implementations here because we
 // sometimes call the methods directly across .c files, rather than going
@@ -180,15 +192,18 @@ VALUE DescriptorPool_alloc(VALUE klass);
 void DescriptorPool_register(VALUE module);
 DescriptorPool* ruby_to_DescriptorPool(VALUE value);
 VALUE DescriptorPool_add(VALUE _self, VALUE def);
-VALUE DescriptorPool_build(VALUE _self);
+VALUE DescriptorPool_build(int argc, VALUE* argv, VALUE _self);
 VALUE DescriptorPool_lookup(VALUE _self, VALUE name);
 VALUE DescriptorPool_generated_pool(VALUE _self);
 
+extern VALUE generated_pool;
+
 void Descriptor_mark(void* _self);
 void Descriptor_free(void* _self);
 VALUE Descriptor_alloc(VALUE klass);
 void Descriptor_register(VALUE module);
 Descriptor* ruby_to_Descriptor(VALUE value);
+VALUE Descriptor_initialize(VALUE _self, VALUE file_descriptor_rb);
 VALUE Descriptor_name(VALUE _self);
 VALUE Descriptor_name_set(VALUE _self, VALUE str);
 VALUE Descriptor_each(VALUE _self);
@@ -198,8 +213,19 @@ VALUE Descriptor_add_oneof(VALUE _self, VALUE obj);
 VALUE Descriptor_each_oneof(VALUE _self);
 VALUE Descriptor_lookup_oneof(VALUE _self, VALUE name);
 VALUE Descriptor_msgclass(VALUE _self);
+VALUE Descriptor_file_descriptor(VALUE _self);
 extern const rb_data_type_t _Descriptor_type;
 
+void FileDescriptor_mark(void* _self);
+void FileDescriptor_free(void* _self);
+VALUE FileDescriptor_alloc(VALUE klass);
+void FileDescriptor_register(VALUE module);
+FileDescriptor* ruby_to_FileDescriptor(VALUE value);
+VALUE FileDescriptor_initialize(int argc, VALUE* argv, VALUE _self);
+VALUE FileDescriptor_name(VALUE _self);
+VALUE FileDescriptor_syntax(VALUE _self);
+VALUE FileDescriptor_syntax_set(VALUE _self, VALUE syntax);
+
 void FieldDescriptor_mark(void* _self);
 void FieldDescriptor_free(void* _self);
 VALUE FieldDescriptor_alloc(VALUE klass);
@@ -209,6 +235,8 @@ VALUE FieldDescriptor_name(VALUE _self);
 VALUE FieldDescriptor_name_set(VALUE _self, VALUE str);
 VALUE FieldDescriptor_type(VALUE _self);
 VALUE FieldDescriptor_type_set(VALUE _self, VALUE type);
+VALUE FieldDescriptor_default(VALUE _self);
+VALUE FieldDescriptor_default_set(VALUE _self, VALUE default_value);
 VALUE FieldDescriptor_label(VALUE _self);
 VALUE FieldDescriptor_label_set(VALUE _self, VALUE label);
 VALUE FieldDescriptor_number(VALUE _self);
@@ -216,6 +244,8 @@ VALUE FieldDescriptor_number_set(VALUE _self, VALUE number);
 VALUE FieldDescriptor_submsg_name(VALUE _self);
 VALUE FieldDescriptor_submsg_name_set(VALUE _self, VALUE value);
 VALUE FieldDescriptor_subtype(VALUE _self);
+VALUE FieldDescriptor_has(VALUE _self, VALUE msg_rb);
+VALUE FieldDescriptor_clear(VALUE _self, VALUE msg_rb);
 VALUE FieldDescriptor_get(VALUE _self, VALUE msg_rb);
 VALUE FieldDescriptor_set(VALUE _self, VALUE msg_rb, VALUE value);
 upb_fieldtype_t ruby_to_fieldtype(VALUE type);
@@ -236,6 +266,8 @@ void EnumDescriptor_free(void* _self);
 VALUE EnumDescriptor_alloc(VALUE klass);
 void EnumDescriptor_register(VALUE module);
 EnumDescriptor* ruby_to_EnumDescriptor(VALUE value);
+VALUE EnumDescriptor_initialize(VALUE _self, VALUE file_descriptor_rb);
+VALUE EnumDescriptor_file_descriptor(VALUE _self);
 VALUE EnumDescriptor_name(VALUE _self);
 VALUE EnumDescriptor_name_set(VALUE _self, VALUE str);
 VALUE EnumDescriptor_add_value(VALUE _self, VALUE name, VALUE number);
@@ -277,11 +309,23 @@ EnumBuilderContext* ruby_to_EnumBuilderContext(VALUE value);
 VALUE EnumBuilderContext_initialize(VALUE _self, VALUE enumdesc);
 VALUE EnumBuilderContext_value(VALUE _self, VALUE name, VALUE number);
 
+void FileBuilderContext_mark(void* _self);
+void FileBuilderContext_free(void* _self);
+VALUE FileBuilderContext_alloc(VALUE klass);
+void FileBuilderContext_register(VALUE module);
+VALUE FileBuilderContext_initialize(VALUE _self, VALUE file_descriptor,
+				    VALUE builder);
+VALUE FileBuilderContext_add_message(VALUE _self, VALUE name);
+VALUE FileBuilderContext_add_enum(VALUE _self, VALUE name);
+VALUE FileBuilderContext_pending_descriptors(VALUE _self);
+
 void Builder_mark(void* _self);
 void Builder_free(void* _self);
 VALUE Builder_alloc(VALUE klass);
 void Builder_register(VALUE module);
 Builder* ruby_to_Builder(VALUE value);
+VALUE Builder_initialize(VALUE _self);
+VALUE Builder_add_file(int argc, VALUE *argv, VALUE _self);
 VALUE Builder_add_message(VALUE _self, VALUE name);
 VALUE Builder_add_enum(VALUE _self, VALUE name);
 VALUE Builder_finalize_to_pool(VALUE _self, VALUE pool_rb);
@@ -293,14 +337,16 @@ VALUE Builder_finalize_to_pool(VALUE _self, VALUE pool_rb);
 #define NATIVE_SLOT_MAX_SIZE sizeof(uint64_t)
 
 size_t native_slot_size(upb_fieldtype_t type);
-void native_slot_set(upb_fieldtype_t type,
+void native_slot_set(const char* name,
+                     upb_fieldtype_t type,
                      VALUE type_class,
                      void* memory,
                      VALUE value);
 // Atomically (with respect to Ruby VM calls) either update the value and set a
 // oneof case, or do neither. If |case_memory| is null, then no case value is
 // set.
-void native_slot_set_value_and_case(upb_fieldtype_t type,
+void native_slot_set_value_and_case(const char* name,
+                                    upb_fieldtype_t type,
                                     VALUE type_class,
                                     void* memory,
                                     VALUE value,
@@ -316,7 +362,7 @@ void native_slot_deep_copy(upb_fieldtype_t type, void* to, void* from);
 bool native_slot_eq(upb_fieldtype_t type, void* mem1, void* mem2);
 
 VALUE native_slot_encode_and_freeze_string(upb_fieldtype_t type, VALUE value);
-void native_slot_check_int_range_precision(upb_fieldtype_t type, VALUE value);
+void native_slot_check_int_range_precision(const char* name, upb_fieldtype_t type, VALUE value);
 
 extern rb_encoding* kRubyStringUtf8Encoding;
 extern rb_encoding* kRubyStringASCIIEncoding;
@@ -397,6 +443,7 @@ typedef struct {
   upb_fieldtype_t key_type;
   upb_fieldtype_t value_type;
   VALUE value_type_class;
+  VALUE parse_frame;
   upb_strtable table;
 } Map;
 
@@ -405,6 +452,7 @@ void Map_free(void* self);
 VALUE Map_alloc(VALUE klass);
 VALUE Map_init(int argc, VALUE* argv, VALUE self);
 void Map_register(VALUE module);
+VALUE Map_set_frame(VALUE self, VALUE val);
 
 extern const rb_data_type_t Map_type;
 extern VALUE cMap;
@@ -424,6 +472,7 @@ VALUE Map_dup(VALUE _self);
 VALUE Map_deep_copy(VALUE _self);
 VALUE Map_eq(VALUE _self, VALUE _other);
 VALUE Map_hash(VALUE _self);
+VALUE Map_to_h(VALUE _self);
 VALUE Map_inspect(VALUE _self);
 VALUE Map_merge(VALUE _self, VALUE hashmap);
 VALUE Map_merge_into_self(VALUE _self, VALUE hashmap);
@@ -444,10 +493,12 @@ VALUE Map_iter_value(Map_iter* iter);
 // -----------------------------------------------------------------------------
 
 #define MESSAGE_FIELD_NO_CASE ((size_t)-1)
+#define MESSAGE_FIELD_NO_HASBIT ((size_t)-1)
 
 struct MessageField {
   size_t offset;
   size_t case_offset;  // for oneofs, a uint32. Else, MESSAGE_FIELD_NO_CASE.
+  size_t hasbit;
 };
 
 struct MessageLayout {
@@ -458,6 +509,9 @@ struct MessageLayout {
 
 MessageLayout* create_layout(const upb_msgdef* msgdef);
 void free_layout(MessageLayout* layout);
+bool field_contains_hasbit(MessageLayout* layout,
+                 const upb_fielddef* field);
+VALUE layout_get_default(const upb_fielddef* field);
 VALUE layout_get(MessageLayout* layout,
                  const void* storage,
                  const upb_fielddef* field);
@@ -465,6 +519,12 @@ void layout_set(MessageLayout* layout,
                 void* storage,
                 const upb_fielddef* field,
                 VALUE val);
+VALUE layout_has(MessageLayout* layout,
+                 const void* storage,
+                 const upb_fielddef* field);
+void layout_clear(MessageLayout* layout,
+                 const void* storage,
+                 const upb_fielddef* field);
 void layout_init(MessageLayout* layout, void* storage);
 void layout_mark(MessageLayout* layout, void* storage);
 void layout_dup(MessageLayout* layout, void* to, void* from);
@@ -477,8 +537,20 @@ VALUE layout_inspect(MessageLayout* layout, void* storage);
 // Message class creation.
 // -----------------------------------------------------------------------------
 
+// This should probably be factored into a common upb component.
+
+typedef struct {
+  upb_byteshandler handler;
+  upb_bytessink sink;
+  char *ptr;
+  size_t len, size;
+} stringsink;
+
+void stringsink_uninit(stringsink *sink);
+
 struct MessageHeader {
-  Descriptor* descriptor;  // kept alive by self.class.descriptor reference.
+  Descriptor* descriptor;      // kept alive by self.class.descriptor reference.
+  stringsink* unknown_fields;  // store unknown fields in decoding.
   // Data comes after this.
 };
 
@@ -496,14 +568,16 @@ VALUE Message_deep_copy(VALUE _self);
 VALUE Message_eq(VALUE _self, VALUE _other);
 VALUE Message_hash(VALUE _self);
 VALUE Message_inspect(VALUE _self);
+VALUE Message_to_h(VALUE _self);
 VALUE Message_index(VALUE _self, VALUE field_name);
 VALUE Message_index_set(VALUE _self, VALUE field_name, VALUE value);
 VALUE Message_descriptor(VALUE klass);
 VALUE Message_decode(VALUE klass, VALUE data);
 VALUE Message_encode(VALUE klass, VALUE msg_rb);
-VALUE Message_decode_json(VALUE klass, VALUE data);
+VALUE Message_decode_json(int argc, VALUE* argv, VALUE klass);
 VALUE Message_encode_json(int argc, VALUE* argv, VALUE klass);
 
+VALUE Google_Protobuf_discard_unknown(VALUE self, VALUE msg_rb);
 VALUE Google_Protobuf_deep_copy(VALUE self, VALUE obj);
 
 VALUE build_module_from_enumdesc(EnumDescriptor* enumdef);

data/ext/google/protobuf_c/repeated_field.c

@@ -178,7 +178,7 @@ VALUE RepeatedField_index_set(VALUE _self, VALUE _index, VALUE val) {
   }
 
   memory = RepeatedField_memoryat(self, index, element_size);
-  native_slot_set(field_type, field_type_class, memory, val);
+  native_slot_set("", field_type, field_type_class, memory, val);
   return Qnil;
 }
 
@@ -217,12 +217,18 @@ VALUE RepeatedField_push(VALUE _self, VALUE val) {
 
   RepeatedField_reserve(self, self->size + 1);
   memory = (void *) (((uint8_t *)self->elements) + self->size * element_size);
-  native_slot_set(field_type, self->field_type_class, memory, val);
+  native_slot_set("", field_type, self->field_type_class, memory, val);
   // native_slot_set may raise an error; bump size only after set.
   self->size++;
   return _self;
 }
 
+VALUE RepeatedField_push_vararg(VALUE _self, VALUE args) {
+  for (int i = 0; i < RARRAY_LEN(args); i++) {
+    RepeatedField_push(_self, rb_ary_entry(args, i));
+  }
+  return _self;
+}
 
 // Used by parsing handlers.
 void RepeatedField_push_native(VALUE _self, void* data) {
@@ -626,8 +632,8 @@ void RepeatedField_register(VALUE module) {
   VALUE klass = rb_define_class_under(
       module, "RepeatedField", rb_cObject);
   rb_define_alloc_func(klass, RepeatedField_alloc);
-  cRepeatedField = klass;
   rb_gc_register_address(&cRepeatedField);
+  cRepeatedField = klass;
 
   rb_define_method(klass, "initialize",
                    RepeatedField_init, -1);
@@ -635,7 +641,7 @@ void RepeatedField_register(VALUE module) {
   rb_define_method(klass, "[]", RepeatedField_index, -1);
   rb_define_method(klass, "at", RepeatedField_index, -1);
   rb_define_method(klass, "[]=", RepeatedField_index_set, 2);
-  rb_define_method(klass, "push", RepeatedField_push, 1);
+  rb_define_method(klass, "push", RepeatedField_push_vararg, -2);
   rb_define_method(klass, "<<", RepeatedField_push, 1);
   rb_define_private_method(klass, "pop_one", RepeatedField_pop_one, 0);
   rb_define_method(klass, "replace", RepeatedField_replace, 1);

data/ext/google/protobuf_c/storage.c

@@ -38,6 +38,8 @@
 // Ruby <-> native slot management.
 // -----------------------------------------------------------------------------
 
+#define CHARPTR_AT(msg, ofs) ((char*)msg + ofs)
+#define DEREF_OFFSET(msg, ofs, type) *(type*)CHARPTR_AT(msg, ofs)
 #define DEREF(memory, type) *(type*)(memory)
 
 size_t native_slot_size(upb_fieldtype_t type) {
@@ -57,46 +59,16 @@ size_t native_slot_size(upb_fieldtype_t type) {
   }
 }
 
-static VALUE value_from_default(const upb_fielddef *field) {
-  switch (upb_fielddef_type(field)) {
-    case UPB_TYPE_FLOAT:   return DBL2NUM(upb_fielddef_defaultfloat(field));
-    case UPB_TYPE_DOUBLE:  return DBL2NUM(upb_fielddef_defaultdouble(field));
-    case UPB_TYPE_BOOL:
-      return upb_fielddef_defaultbool(field) ? Qtrue : Qfalse;
-    case UPB_TYPE_MESSAGE: return Qnil;
-    case UPB_TYPE_ENUM: {
-      const upb_enumdef *enumdef = upb_fielddef_enumsubdef(field);
-      int32_t num = upb_fielddef_defaultint32(field);
-      const char *label = upb_enumdef_iton(enumdef, num);
-      if (label) {
-        return ID2SYM(rb_intern(label));
-      } else {
-        return INT2NUM(num);
-      }
-    }
-    case UPB_TYPE_INT32:   return INT2NUM(upb_fielddef_defaultint32(field));
-    case UPB_TYPE_INT64:   return LL2NUM(upb_fielddef_defaultint64(field));;
-    case UPB_TYPE_UINT32:  return UINT2NUM(upb_fielddef_defaultuint32(field));
-    case UPB_TYPE_UINT64:  return ULL2NUM(upb_fielddef_defaultuint64(field));
-    case UPB_TYPE_STRING:
-    case UPB_TYPE_BYTES: {
-      size_t size;
-      const char *str = upb_fielddef_defaultstr(field, &size);
-      return rb_str_new(str, size);
-    }
-    default: return Qnil;
-  }
-}
-
 static bool is_ruby_num(VALUE value) {
   return (TYPE(value) == T_FLOAT ||
           TYPE(value) == T_FIXNUM ||
           TYPE(value) == T_BIGNUM);
 }
 
-void native_slot_check_int_range_precision(upb_fieldtype_t type, VALUE val) {
+void native_slot_check_int_range_precision(const char* name, upb_fieldtype_t type, VALUE val) {
   if (!is_ruby_num(val)) {
-    rb_raise(rb_eTypeError, "Expected number type for integral field.");
+    rb_raise(cTypeError, "Expected number type for integral field '%s' (given %s).",
+             name, rb_class2name(CLASS_OF(val)));
   }
 
   // NUM2{INT,UINT,LL,ULL} macros do the appropriate range checks on upper
@@ -106,13 +78,15 @@ void native_slot_check_int_range_precision(upb_fieldtype_t type, VALUE val) {
     double dbl_val = NUM2DBL(val);
     if (floor(dbl_val) != dbl_val) {
       rb_raise(rb_eRangeError,
-               "Non-integral floating point value assigned to integer field.");
+               "Non-integral floating point value assigned to integer field '%s' (given %s).",
+               name, rb_class2name(CLASS_OF(val)));
     }
   }
   if (type == UPB_TYPE_UINT32 || type == UPB_TYPE_UINT64) {
     if (NUM2DBL(val) < 0) {
       rb_raise(rb_eRangeError,
-               "Assigning negative value to unsigned integer field.");
+               "Assigning negative value to unsigned integer field '%s' (given %s).",
+               name, rb_class2name(CLASS_OF(val)));
     }
   }
 }
@@ -137,12 +111,14 @@ VALUE native_slot_encode_and_freeze_string(upb_fieldtype_t type, VALUE value) {
   return value;
 }
 
-void native_slot_set(upb_fieldtype_t type, VALUE type_class,
+void native_slot_set(const char* name,
+                     upb_fieldtype_t type, VALUE type_class,
                      void* memory, VALUE value) {
-  native_slot_set_value_and_case(type, type_class, memory, value, NULL, 0);
+  native_slot_set_value_and_case(name, type, type_class, memory, value, NULL, 0);
 }
 
-void native_slot_set_value_and_case(upb_fieldtype_t type, VALUE type_class,
+void native_slot_set_value_and_case(const char* name,
+                                    upb_fieldtype_t type, VALUE type_class,
                                     void* memory, VALUE value,
                                     uint32_t* case_memory,
                                     uint32_t case_number) {
@@ -153,13 +129,15 @@ void native_slot_set_value_and_case(upb_fieldtype_t type, VALUE type_class,
   switch (type) {
     case UPB_TYPE_FLOAT:
       if (!is_ruby_num(value)) {
-        rb_raise(rb_eTypeError, "Expected number type for float field.");
+        rb_raise(cTypeError, "Expected number type for float field '%s' (given %s).",
+                 name, rb_class2name(CLASS_OF(value)));
       }
       DEREF(memory, float) = NUM2DBL(value);
       break;
     case UPB_TYPE_DOUBLE:
       if (!is_ruby_num(value)) {
-        rb_raise(rb_eTypeError, "Expected number type for double field.");
+        rb_raise(cTypeError, "Expected number type for double field '%s' (given %s).",
+                 name, rb_class2name(CLASS_OF(value)));
       }
       DEREF(memory, double) = NUM2DBL(value);
       break;
@@ -170,15 +148,27 @@ void native_slot_set_value_and_case(upb_fieldtype_t type, VALUE type_class,
       } else if (value == Qfalse) {
         val = 0;
       } else {
-        rb_raise(rb_eTypeError, "Invalid argument for boolean field.");
+        rb_raise(cTypeError, "Invalid argument for boolean field '%s' (given %s).",
+                 name, rb_class2name(CLASS_OF(value)));
       }
       DEREF(memory, int8_t) = val;
       break;
     }
     case UPB_TYPE_STRING:
+      if (CLASS_OF(value) == rb_cSymbol) {
+        value = rb_funcall(value, rb_intern("to_s"), 0);
+      } else if (CLASS_OF(value) != rb_cString) {
+        rb_raise(cTypeError, "Invalid argument for string field '%s' (given %s).",
+                 name, rb_class2name(CLASS_OF(value)));
+      }
+
+      DEREF(memory, VALUE) = native_slot_encode_and_freeze_string(type, value);
+      break;
+
     case UPB_TYPE_BYTES: {
       if (CLASS_OF(value) != rb_cString) {
-        rb_raise(rb_eTypeError, "Invalid argument for string field.");
+        rb_raise(cTypeError, "Invalid argument for bytes field '%s' (given %s).",
+                 name, rb_class2name(CLASS_OF(value)));
       }
 
       DEREF(memory, VALUE) = native_slot_encode_and_freeze_string(type, value);
@@ -188,29 +178,61 @@ void native_slot_set_value_and_case(upb_fieldtype_t type, VALUE type_class,
       if (CLASS_OF(value) == CLASS_OF(Qnil)) {
         value = Qnil;
       } else if (CLASS_OF(value) != type_class) {
-        rb_raise(rb_eTypeError,
-                 "Invalid type %s to assign to submessage field.",
-                 rb_class2name(CLASS_OF(value)));
+        // check for possible implicit conversions
+        VALUE converted_value = NULL;
+        char* field_type_name = rb_class2name(type_class);
+
+        if (strcmp(field_type_name, "Google::Protobuf::Timestamp") == 0 &&
+            rb_obj_is_kind_of(value, rb_cTime)) {
+          // Time -> Google::Protobuf::Timestamp
+          VALUE hash = rb_hash_new();
+          rb_hash_aset(hash, rb_str_new2("seconds"), rb_funcall(value, rb_intern("to_i"), 0));
+          rb_hash_aset(hash, rb_str_new2("nanos"), rb_funcall(value, rb_intern("nsec"), 0));
+          VALUE args[1] = { hash };
+          converted_value = rb_class_new_instance(1, args, type_class);
+        } else if (strcmp(field_type_name, "Google::Protobuf::Duration") == 0 &&
+                   rb_obj_is_kind_of(value, rb_cNumeric)) {
+          // Numeric -> Google::Protobuf::Duration
+          VALUE hash = rb_hash_new();
+          rb_hash_aset(hash, rb_str_new2("seconds"), rb_funcall(value, rb_intern("to_i"), 0));
+          VALUE n_value = rb_funcall(value, rb_intern("remainder"), 1, INT2NUM(1));
+          n_value = rb_funcall(n_value, rb_intern("*"), 1, INT2NUM(1000000000));
+          n_value = rb_funcall(n_value, rb_intern("round"), 0);
+          rb_hash_aset(hash, rb_str_new2("nanos"), n_value);
+          VALUE args[1] = { hash };
+          converted_value = rb_class_new_instance(1, args, type_class);
+        }
+
+        // raise if no suitable conversaion could be found
+        if (converted_value == NULL) {
+          rb_raise(cTypeError,
+                   "Invalid type %s to assign to submessage field '%s'.",
+                  rb_class2name(CLASS_OF(value)), name);
+        } else {
+          value = converted_value;
+        }
       }
       DEREF(memory, VALUE) = value;
       break;
     }
     case UPB_TYPE_ENUM: {
       int32_t int_val = 0;
-      if (!is_ruby_num(value) && TYPE(value) != T_SYMBOL) {
-        rb_raise(rb_eTypeError,
-                 "Expected number or symbol type for enum field.");
+      if (TYPE(value) == T_STRING) {
+        value = rb_funcall(value, rb_intern("to_sym"), 0);
+      } else if (!is_ruby_num(value) && TYPE(value) != T_SYMBOL) {
+        rb_raise(cTypeError,
+                 "Expected number or symbol type for enum field '%s'.", name);
       }
       if (TYPE(value) == T_SYMBOL) {
         // Ensure that the given symbol exists in the enum module.
         VALUE lookup = rb_funcall(type_class, rb_intern("resolve"), 1, value);
         if (lookup == Qnil) {
-          rb_raise(rb_eRangeError, "Unknown symbol value for enum field.");
+          rb_raise(rb_eRangeError, "Unknown symbol value for enum field '%s'.", name);
         } else {
           int_val = NUM2INT(lookup);
         }
       } else {
-        native_slot_check_int_range_precision(UPB_TYPE_INT32, value);
+        native_slot_check_int_range_precision(name, UPB_TYPE_INT32, value);
         int_val = NUM2INT(value);
       }
       DEREF(memory, int32_t) = int_val;
@@ -220,7 +242,7 @@ void native_slot_set_value_and_case(upb_fieldtype_t type, VALUE type_class,
     case UPB_TYPE_INT64:
     case UPB_TYPE_UINT32:
     case UPB_TYPE_UINT64:
-      native_slot_check_int_range_precision(type, value);
+      native_slot_check_int_range_precision(name, type, value);
       switch (type) {
       case UPB_TYPE_INT32:
         DEREF(memory, int32_t) = NUM2INT(value);
@@ -393,7 +415,12 @@ const upb_msgdef *map_entry_msgdef(const upb_fielddef* field) {
 }
 
 bool is_map_field(const upb_fielddef *field) {
-  return tryget_map_entry_msgdef(field) != NULL;
+  const upb_msgdef* subdef = tryget_map_entry_msgdef(field);
+  if (subdef == NULL) return false;
+
+  // Map fields are a proto3 feature.
+  // If we're using proto2 syntax we need to fallback to the repeated field.
+  return upb_msgdef_syntax(subdef) == UPB_SYNTAX_PROTO3;
 }
 
 const upb_fielddef* map_field_key(const upb_fielddef* field) {
@@ -422,6 +449,12 @@ const upb_fielddef* map_entry_value(const upb_msgdef* msgdef) {
 // Memory layout management.
 // -----------------------------------------------------------------------------
 
+bool field_contains_hasbit(MessageLayout* layout,
+                            const upb_fielddef* field) {
+  return layout->fields[upb_fielddef_index(field)].hasbit !=
+      MESSAGE_FIELD_NO_HASBIT;
+}
+
 static size_t align_up_to(size_t offset, size_t granularity) {
   // Granularity must be a power of two.
   return (offset + granularity - 1) & ~(granularity - 1);
@@ -436,6 +469,23 @@ MessageLayout* create_layout(const upb_msgdef* msgdef) {
 
   layout->fields = ALLOC_N(MessageField, nfields);
 
+  size_t hasbit = 0;
+  for (upb_msg_field_begin(&it, msgdef);
+       !upb_msg_field_done(&it);
+       upb_msg_field_next(&it)) {
+    const upb_fielddef* field = upb_msg_iter_field(&it);
+    if (upb_fielddef_haspresence(field)) {
+      layout->fields[upb_fielddef_index(field)].hasbit = hasbit++;
+    } else {
+      layout->fields[upb_fielddef_index(field)].hasbit =
+	  MESSAGE_FIELD_NO_HASBIT;
+    }
+  }
+
+  if (hasbit != 0) {
+    off += (hasbit + 8 - 1) / 8;
+  }
+
   for (upb_msg_field_begin(&it, msgdef);
        !upb_msg_field_done(&it);
        upb_msg_field_next(&it)) {
@@ -558,6 +608,137 @@ static uint32_t* slot_oneof_case(MessageLayout* layout,
       layout->fields[upb_fielddef_index(field)].case_offset);
 }
 
+static void slot_set_hasbit(MessageLayout* layout,
+                            const void* storage,
+                            const upb_fielddef* field) {
+  size_t hasbit = layout->fields[upb_fielddef_index(field)].hasbit;
+  assert(hasbit != MESSAGE_FIELD_NO_HASBIT);
+
+  ((uint8_t*)storage)[hasbit / 8] |= 1 << (hasbit % 8);
+}
+
+static void slot_clear_hasbit(MessageLayout* layout,
+                              const void* storage,
+                              const upb_fielddef* field) {
+  size_t hasbit = layout->fields[upb_fielddef_index(field)].hasbit;
+  assert(hasbit != MESSAGE_FIELD_NO_HASBIT);
+  ((uint8_t*)storage)[hasbit / 8] &= ~(1 << (hasbit % 8));
+}
+
+static bool slot_is_hasbit_set(MessageLayout* layout,
+                            const void* storage,
+                            const upb_fielddef* field) {
+  size_t hasbit = layout->fields[upb_fielddef_index(field)].hasbit;
+  if (hasbit == MESSAGE_FIELD_NO_HASBIT) {
+    return false;
+  }
+
+  return DEREF_OFFSET(
+      (uint8_t*)storage, hasbit / 8, char) & (1 << (hasbit % 8));
+}
+
+VALUE layout_has(MessageLayout* layout,
+                 const void* storage,
+                 const upb_fielddef* field) {
+  assert(field_contains_hasbit(layout, field));
+  return slot_is_hasbit_set(layout, storage, field) ? Qtrue : Qfalse;
+}
+
+void layout_clear(MessageLayout* layout,
+                 const void* storage,
+                 const upb_fielddef* field) {
+  void* memory = slot_memory(layout, storage, field);
+  uint32_t* oneof_case = slot_oneof_case(layout, storage, field);
+
+  if (field_contains_hasbit(layout, field)) {
+    slot_clear_hasbit(layout, storage, field);
+  }
+
+  if (upb_fielddef_containingoneof(field)) {
+    memset(memory, 0, NATIVE_SLOT_MAX_SIZE);
+    *oneof_case = ONEOF_CASE_NONE;
+  } else if (is_map_field(field)) {
+    VALUE map = Qnil;
+
+    const upb_fielddef* key_field = map_field_key(field);
+    const upb_fielddef* value_field = map_field_value(field);
+    VALUE type_class = field_type_class(value_field);
+
+    if (type_class != Qnil) {
+      VALUE args[3] = {
+        fieldtype_to_ruby(upb_fielddef_type(key_field)),
+        fieldtype_to_ruby(upb_fielddef_type(value_field)),
+        type_class,
+      };
+      map = rb_class_new_instance(3, args, cMap);
+    } else {
+      VALUE args[2] = {
+        fieldtype_to_ruby(upb_fielddef_type(key_field)),
+        fieldtype_to_ruby(upb_fielddef_type(value_field)),
+      };
+      map = rb_class_new_instance(2, args, cMap);
+    }
+
+    DEREF(memory, VALUE) = map;
+  } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) {
+    VALUE ary = Qnil;
+
+    VALUE type_class = field_type_class(field);
+
+    if (type_class != Qnil) {
+      VALUE args[2] = {
+        fieldtype_to_ruby(upb_fielddef_type(field)),
+        type_class,
+      };
+      ary = rb_class_new_instance(2, args, cRepeatedField);
+    } else {
+      VALUE args[1] = { fieldtype_to_ruby(upb_fielddef_type(field)) };
+      ary = rb_class_new_instance(1, args, cRepeatedField);
+    }
+
+    DEREF(memory, VALUE) = ary;
+  } else {
+    native_slot_set(upb_fielddef_name(field),
+                    upb_fielddef_type(field), field_type_class(field),
+                    memory, layout_get_default(field));
+  }
+}
+
+VALUE layout_get_default(const upb_fielddef *field) {
+  switch (upb_fielddef_type(field)) {
+    case UPB_TYPE_FLOAT:   return DBL2NUM(upb_fielddef_defaultfloat(field));
+    case UPB_TYPE_DOUBLE:  return DBL2NUM(upb_fielddef_defaultdouble(field));
+    case UPB_TYPE_BOOL:
+      return upb_fielddef_defaultbool(field) ? Qtrue : Qfalse;
+    case UPB_TYPE_MESSAGE: return Qnil;
+    case UPB_TYPE_ENUM: {
+      const upb_enumdef *enumdef = upb_fielddef_enumsubdef(field);
+      int32_t num = upb_fielddef_defaultint32(field);
+      const char *label = upb_enumdef_iton(enumdef, num);
+      if (label) {
+        return ID2SYM(rb_intern(label));
+      } else {
+        return INT2NUM(num);
+      }
+    }
+    case UPB_TYPE_INT32:   return INT2NUM(upb_fielddef_defaultint32(field));
+    case UPB_TYPE_INT64:   return LL2NUM(upb_fielddef_defaultint64(field));;
+    case UPB_TYPE_UINT32:  return UINT2NUM(upb_fielddef_defaultuint32(field));
+    case UPB_TYPE_UINT64:  return ULL2NUM(upb_fielddef_defaultuint64(field));
+    case UPB_TYPE_STRING:
+    case UPB_TYPE_BYTES: {
+      size_t size;
+      const char *str = upb_fielddef_defaultstr(field, &size);
+      VALUE str_rb = rb_str_new(str, size);
+
+      rb_enc_associate(str_rb, (upb_fielddef_type(field) == UPB_TYPE_BYTES) ?
+                 kRubyString8bitEncoding : kRubyStringUtf8Encoding);
+      rb_obj_freeze(str_rb);
+      return str_rb;
+    }
+    default: return Qnil;
+  }
+}
 
 VALUE layout_get(MessageLayout* layout,
                  const void* storage,
@@ -565,15 +746,24 @@ VALUE layout_get(MessageLayout* layout,
   void* memory = slot_memory(layout, storage, field);
   uint32_t* oneof_case = slot_oneof_case(layout, storage, field);
 
+  bool field_set;
+  if (field_contains_hasbit(layout, field)) {
+    field_set = slot_is_hasbit_set(layout, storage, field);
+  } else {
+    field_set = true;
+  }
+
   if (upb_fielddef_containingoneof(field)) {
     if (*oneof_case != upb_fielddef_number(field)) {
-      return value_from_default(field);
+      return layout_get_default(field);
     }
     return native_slot_get(upb_fielddef_type(field),
                            field_type_class(field),
                            memory);
   } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) {
     return *((VALUE *)memory);
+  } else if (!field_set) {
+    return layout_get_default(field);
   } else {
     return native_slot_get(upb_fielddef_type(field),
                            field_type_class(field),
@@ -587,20 +777,28 @@ static void check_repeated_field_type(VALUE val, const upb_fielddef* field) {
 
   if (!RB_TYPE_P(val, T_DATA) || !RTYPEDDATA_P(val) ||
       RTYPEDDATA_TYPE(val) != &RepeatedField_type) {
-    rb_raise(rb_eTypeError, "Expected repeated field array");
+    rb_raise(cTypeError, "Expected repeated field array");
   }
 
   self = ruby_to_RepeatedField(val);
   if (self->field_type != upb_fielddef_type(field)) {
-    rb_raise(rb_eTypeError, "Repeated field array has wrong element type");
+    rb_raise(cTypeError, "Repeated field array has wrong element type");
   }
 
-  if (self->field_type == UPB_TYPE_MESSAGE ||
-      self->field_type == UPB_TYPE_ENUM) {
+  if (self->field_type == UPB_TYPE_MESSAGE) {
     if (self->field_type_class !=
-        get_def_obj(upb_fielddef_subdef(field))) {
-      rb_raise(rb_eTypeError,
-               "Repeated field array has wrong message/enum class");
+        Descriptor_msgclass(get_def_obj(upb_fielddef_subdef(field)))) {
+      rb_raise(cTypeError,
+               "Repeated field array has wrong message class");
+    }
+  }
+
+
+  if (self->field_type == UPB_TYPE_ENUM) {
+    if (self->field_type_class !=
+        EnumDescriptor_enummodule(get_def_obj(upb_fielddef_subdef(field)))) {
+      rb_raise(cTypeError,
+               "Repeated field array has wrong enum class");
     }
   }
 }
@@ -612,21 +810,21 @@ static void check_map_field_type(VALUE val, const upb_fielddef* field) {
 
   if (!RB_TYPE_P(val, T_DATA) || !RTYPEDDATA_P(val) ||
       RTYPEDDATA_TYPE(val) != &Map_type) {
-    rb_raise(rb_eTypeError, "Expected Map instance");
+    rb_raise(cTypeError, "Expected Map instance");
   }
 
   self = ruby_to_Map(val);
   if (self->key_type != upb_fielddef_type(key_field)) {
-    rb_raise(rb_eTypeError, "Map key type does not match field's key type");
+    rb_raise(cTypeError, "Map key type does not match field's key type");
   }
   if (self->value_type != upb_fielddef_type(value_field)) {
-    rb_raise(rb_eTypeError, "Map value type does not match field's value type");
+    rb_raise(cTypeError, "Map value type does not match field's value type");
   }
   if (upb_fielddef_type(value_field) == UPB_TYPE_MESSAGE ||
       upb_fielddef_type(value_field) == UPB_TYPE_ENUM) {
     if (self->value_type_class !=
         get_def_obj(upb_fielddef_subdef(value_field))) {
-      rb_raise(rb_eTypeError,
+      rb_raise(cTypeError,
                "Map value type has wrong message/enum class");
     }
   }
@@ -659,6 +857,7 @@ void layout_set(MessageLayout* layout,
       // use native_slot_set_value_and_case(), which ensures that both the value
       // and case number are altered atomically (w.r.t. the Ruby VM).
       native_slot_set_value_and_case(
+          upb_fielddef_name(field),
           upb_fielddef_type(field), field_type_class(field),
           memory, val,
           oneof_case, upb_fielddef_number(field));
@@ -670,67 +869,25 @@ void layout_set(MessageLayout* layout,
     check_repeated_field_type(val, field);
     DEREF(memory, VALUE) = val;
   } else {
-    native_slot_set(upb_fielddef_type(field), field_type_class(field),
+    native_slot_set(upb_fielddef_name(field),
+                    upb_fielddef_type(field), field_type_class(field),
                     memory, val);
   }
+
+  if (layout->fields[upb_fielddef_index(field)].hasbit !=
+      MESSAGE_FIELD_NO_HASBIT) {
+    slot_set_hasbit(layout, storage, field);
+  }
 }
 
 void layout_init(MessageLayout* layout,
                  void* storage) {
+
   upb_msg_field_iter it;
   for (upb_msg_field_begin(&it, layout->msgdef);
        !upb_msg_field_done(&it);
        upb_msg_field_next(&it)) {
-    const upb_fielddef* field = upb_msg_iter_field(&it);
-    void* memory = slot_memory(layout, storage, field);
-    uint32_t* oneof_case = slot_oneof_case(layout, storage, field);
-
-    if (upb_fielddef_containingoneof(field)) {
-      memset(memory, 0, NATIVE_SLOT_MAX_SIZE);
-      *oneof_case = ONEOF_CASE_NONE;
-    } else if (is_map_field(field)) {
-      VALUE map = Qnil;
-
-      const upb_fielddef* key_field = map_field_key(field);
-      const upb_fielddef* value_field = map_field_value(field);
-      VALUE type_class = field_type_class(value_field);
-
-      if (type_class != Qnil) {
-        VALUE args[3] = {
-          fieldtype_to_ruby(upb_fielddef_type(key_field)),
-          fieldtype_to_ruby(upb_fielddef_type(value_field)),
-          type_class,
-        };
-        map = rb_class_new_instance(3, args, cMap);
-      } else {
-        VALUE args[2] = {
-          fieldtype_to_ruby(upb_fielddef_type(key_field)),
-          fieldtype_to_ruby(upb_fielddef_type(value_field)),
-        };
-        map = rb_class_new_instance(2, args, cMap);
-      }
-
-      DEREF(memory, VALUE) = map;
-    } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) {
-      VALUE ary = Qnil;
-
-      VALUE type_class = field_type_class(field);
-
-      if (type_class != Qnil) {
-        VALUE args[2] = {
-          fieldtype_to_ruby(upb_fielddef_type(field)),
-          type_class,
-        };
-        ary = rb_class_new_instance(2, args, cRepeatedField);
-      } else {
-        VALUE args[1] = { fieldtype_to_ruby(upb_fielddef_type(field)) };
-        ary = rb_class_new_instance(1, args, cRepeatedField);
-      }
-
-      DEREF(memory, VALUE) = ary;
-    } else {
-      native_slot_init(upb_fielddef_type(field), memory);
-    }
+    layout_clear(layout, storage, upb_msg_iter_field(&it));
   }
 }
 
@@ -777,6 +934,11 @@ void layout_dup(MessageLayout* layout, void* to, void* from) {
     } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) {
       DEREF(to_memory, VALUE) = RepeatedField_dup(DEREF(from_memory, VALUE));
     } else {
+      if (field_contains_hasbit(layout, field)) {
+        if (!slot_is_hasbit_set(layout, from, field)) continue;
+        slot_set_hasbit(layout, to, field);
+      }
+
       native_slot_dup(upb_fielddef_type(field), to_memory, from_memory);
     }
   }
@@ -806,6 +968,11 @@ void layout_deep_copy(MessageLayout* layout, void* to, void* from) {
       DEREF(to_memory, VALUE) =
           RepeatedField_deep_copy(DEREF(from_memory, VALUE));
     } else {
+      if (field_contains_hasbit(layout, field)) {
+        if (!slot_is_hasbit_set(layout, from, field)) continue;
+        slot_set_hasbit(layout, to, field);
+      }
+
       native_slot_deep_copy(upb_fielddef_type(field), to_memory, from_memory);
     }
   }
@@ -842,8 +1009,10 @@ VALUE layout_eq(MessageLayout* layout, void* msg1, void* msg2) {
         return Qfalse;
       }
     } else {
-      if (!native_slot_eq(upb_fielddef_type(field),
-                          msg1_memory, msg2_memory)) {
+      if (slot_is_hasbit_set(layout, msg1, field) !=
+	  slot_is_hasbit_set(layout, msg2, field) ||
+          !native_slot_eq(upb_fielddef_type(field),
+			  msg1_memory, msg2_memory)) {
         return Qfalse;
       }
     }