Avoid too much overhead in layout_init (#6716)

* Avoid initializing primitive fields in layout_init

* Avoid initializing string/bytes/message fields in layout_init

* Lazily create map when needed

* Lazily create repeated fields

* Change layout_init to only do memcpy

* Fix test for php-7.0

* Fix conformance test where default value of string/message map is not encoded

* Fix test for zts

* Clean up

* Fix comments

.gitignore

@@ -139,6 +139,8 @@ composer.lock
 php/tests/generated/
 php/tests/old_protoc
 php/tests/protobuf/
+php/tests/core
+php/tests/vgcore*
 php/ext/google/protobuf/.libs/
 php/ext/google/protobuf/Makefile.fragments
 php/ext/google/protobuf/Makefile.global

php/ext/google/protobuf/array.c

@@ -259,6 +259,19 @@ void repeated_field_push_native(RepeatedField *intern, void *value) {
   }
 }
 
+void repeated_field_ensure_created(
+    const upb_fielddef *field,
+    CACHED_VALUE *repeated_field PHP_PROTO_TSRMLS_DC) {
+  if (ZVAL_IS_NULL(CACHED_PTR_TO_ZVAL_PTR(repeated_field))) {
+    zval_ptr_dtor(repeated_field);
+#if PHP_MAJOR_VERSION < 7
+    MAKE_STD_ZVAL(CACHED_PTR_TO_ZVAL_PTR(repeated_field));
+#endif
+    repeated_field_create_with_field(repeated_field_type, field,
+                                     repeated_field PHP_PROTO_TSRMLS_CC);
+  }
+}
+
 void repeated_field_create_with_field(
     zend_class_entry *ce, const upb_fielddef *field,
     CACHED_VALUE *repeated_field PHP_PROTO_TSRMLS_DC) {

php/ext/google/protobuf/map.c

@@ -243,6 +243,18 @@ map_field_handlers->write_dimension = map_field_write_dimension;
 map_field_handlers->get_gc = map_field_get_gc;
 PHP_PROTO_INIT_CLASS_END
 
+void map_field_ensure_created(const upb_fielddef *field,
+                              CACHED_VALUE *map_field PHP_PROTO_TSRMLS_DC) {
+  if (ZVAL_IS_NULL(CACHED_PTR_TO_ZVAL_PTR(map_field))) {
+    zval_ptr_dtor(map_field);
+#if PHP_MAJOR_VERSION < 7
+    MAKE_STD_ZVAL(CACHED_PTR_TO_ZVAL_PTR(map_field));
+#endif
+    map_field_create_with_field(map_field_type, field,
+                                map_field PHP_PROTO_TSRMLS_CC);
+  }
+}
+
 void map_field_create_with_field(const zend_class_entry *ce,
                                  const upb_fielddef *field,
                                  CACHED_VALUE *map_field PHP_PROTO_TSRMLS_DC) {

php/ext/google/protobuf/message.c

@@ -178,7 +178,7 @@ static zval* message_get_property_internal(zval* object,
       zend_get_property_info(Z_OBJCE_P(object), Z_STR_P(member), true);
 #endif
   return layout_get(
-      self->descriptor->layout, message_data(self), field,
+      self->descriptor->layout, self, field,
       OBJ_PROP(Z_OBJ_P(object), property_info->offset) TSRMLS_CC);
 }
 
@@ -191,7 +191,7 @@ static void message_get_oneof_property_internal(zval* object, zval* member,
     return;
   }
 
-  layout_get(self->descriptor->layout, message_data(self), field,
+  layout_get(self->descriptor->layout, self, field,
              ZVAL_PTR_TO_CACHED_PTR(return_value) TSRMLS_CC);
 }
 
@@ -255,7 +255,6 @@ void custom_data_init(const zend_class_entry* ce,
                       MessageHeader* intern PHP_PROTO_TSRMLS_DC) {
   Descriptor* desc = UNBOX_HASHTABLE_VALUE(Descriptor, get_ce_obj(ce));
   intern->data = ALLOC_N(uint8_t, desc->layout->size);
-  memset(message_data(intern), 0, desc->layout->size);
   // We wrap first so that everything in the message object is GC-rooted in
   // case a collection happens during object creation in layout_init().
   intern->descriptor = desc;
@@ -575,9 +574,9 @@ PHP_METHOD(Message, readOneof) {
   const upb_fielddef* field = upb_msgdef_itof(msg->descriptor->msgdef, index);
 
   // Unlike singular fields, oneof fields share cached property. So we cannot
-  // let lay_get modify the cached property. Instead, we pass in the return
+  // let layout_get modify the cached property. Instead, we pass in the return
   // value directly.
-  layout_get(msg->descriptor->layout, message_data(msg), field,
+  layout_get(msg->descriptor->layout, msg, field,
              ZVAL_PTR_TO_CACHED_PTR(return_value) TSRMLS_CC);
 }
 

php/ext/google/protobuf/protobuf.h

@@ -200,7 +200,7 @@
 
 #define CACHED_VALUE zval*
 #define CACHED_TO_ZVAL_PTR(VALUE) (VALUE)
-#define CACHED_PTR_TO_ZVAL_PTR(VALUE) (*VALUE)
+#define CACHED_PTR_TO_ZVAL_PTR(VALUE) (*(CACHED_VALUE*)(VALUE))
 #define ZVAL_PTR_TO_CACHED_PTR(VALUE) (&VALUE)
 #define ZVAL_PTR_TO_CACHED_VALUE(VALUE) (VALUE)
 #define ZVAL_TO_CACHED_VALUE(VALUE) (&VALUE)
@@ -475,7 +475,7 @@ static inline int php_proto_zend_hash_get_current_data_ex(HashTable* ht,
 
 #define CACHED_VALUE zval
 #define CACHED_TO_ZVAL_PTR(VALUE) (&VALUE)
-#define CACHED_PTR_TO_ZVAL_PTR(VALUE) (VALUE)
+#define CACHED_PTR_TO_ZVAL_PTR(VALUE) ((CACHED_VALUE*)(VALUE))
 #define ZVAL_PTR_TO_CACHED_PTR(VALUE) (VALUE)
 #define ZVAL_PTR_TO_CACHED_VALUE(VALUE) (*VALUE)
 #define ZVAL_TO_CACHED_VALUE(VALUE) (VALUE)
@@ -935,6 +935,7 @@ struct MessageField {
 
 struct MessageLayout {
   const upb_msgdef* msgdef;
+  void* empty_template;  // Can memcpy() onto a layout to clear it.
   MessageField* fields;
   size_t size;
 };
@@ -948,7 +949,7 @@ PHP_PROTO_WRAP_OBJECT_END
 MessageLayout* create_layout(const upb_msgdef* msgdef);
 void layout_init(MessageLayout* layout, void* storage,
                  zend_object* object PHP_PROTO_TSRMLS_DC);
-zval* layout_get(MessageLayout* layout, const void* storage,
+zval* layout_get(MessageLayout* layout, MessageHeader* header,
                  const upb_fielddef* field, CACHED_VALUE* cache TSRMLS_DC);
 void layout_set(MessageLayout* layout, MessageHeader* header,
                 const upb_fielddef* field, zval* val TSRMLS_DC);
@@ -1089,6 +1090,8 @@ upb_value map_iter_value(MapIter* iter, int* len);
 const upb_fielddef* map_entry_key(const upb_msgdef* msgdef);
 const upb_fielddef* map_entry_value(const upb_msgdef* msgdef);
 
+void map_field_ensure_created(const upb_fielddef *field,
+                              CACHED_VALUE *map_field PHP_PROTO_TSRMLS_DC);
 void map_field_create_with_field(const zend_class_entry* ce,
                                  const upb_fielddef* field,
                                  CACHED_VALUE* map_field PHP_PROTO_TSRMLS_DC);
@@ -1147,6 +1150,9 @@ PHP_PROTO_WRAP_OBJECT_START(RepeatedFieldIter)
   long position;
 PHP_PROTO_WRAP_OBJECT_END
 
+void repeated_field_ensure_created(
+    const upb_fielddef *field,
+    CACHED_VALUE *repeated_field PHP_PROTO_TSRMLS_DC);
 void repeated_field_create_with_field(
     zend_class_entry* ce, const upb_fielddef* field,
     CACHED_VALUE* repeated_field PHP_PROTO_TSRMLS_DC);
@@ -1489,6 +1495,9 @@ size_t stringsink_string(void *_sink, const void *hd, const char *ptr,
 #define FREE(object) efree(object)
 #define PEFREE(object) pefree(object, 1)
 
+// Find corresponding zval property for the field.
+CACHED_VALUE* find_zval_property(MessageHeader* msg, const upb_fielddef* field);
+
 // String argument.
 #define STR(str) (str), strlen(str)