google-protobuf 3.0.0 → 3.20.0

data/ext/google/protobuf_c/protobuf.h

@@ -31,509 +31,90 @@
 #ifndef __GOOGLE_PROTOBUF_RUBY_PROTOBUF_H__
 #define __GOOGLE_PROTOBUF_RUBY_PROTOBUF_H__
 
+#include <ruby/encoding.h>
 #include <ruby/ruby.h>
 #include <ruby/vm.h>
-#include <ruby/encoding.h>
-
-#include "upb.h"
-
-// Forward decls.
-struct DescriptorPool;
-struct Descriptor;
-struct FieldDescriptor;
-struct EnumDescriptor;
-struct MessageLayout;
-struct MessageField;
-struct MessageHeader;
-struct MessageBuilderContext;
-struct EnumBuilderContext;
-struct Builder;
-
-typedef struct DescriptorPool DescriptorPool;
-typedef struct Descriptor Descriptor;
-typedef struct FieldDescriptor FieldDescriptor;
-typedef struct OneofDescriptor OneofDescriptor;
-typedef struct EnumDescriptor EnumDescriptor;
-typedef struct MessageLayout MessageLayout;
-typedef struct MessageField MessageField;
-typedef struct MessageHeader MessageHeader;
-typedef struct MessageBuilderContext MessageBuilderContext;
-typedef struct OneofBuilderContext OneofBuilderContext;
-typedef struct EnumBuilderContext EnumBuilderContext;
-typedef struct Builder Builder;
-
-/*
- It can be a bit confusing how the C structs defined below and the Ruby
- objects interact and hold references to each other. First, a few principles:
-
- - Ruby's "TypedData" abstraction lets a Ruby VALUE hold a pointer to a C
-   struct (or arbitrary memory chunk), own it, and free it when collected.
-   Thus, each struct below will have a corresponding Ruby object
-   wrapping/owning it.
-
- - To get back from an underlying upb {msg,enum}def to the Ruby object, we
-   keep a global hashmap, accessed by get_def_obj/add_def_obj below.
-
- The in-memory structure is then something like:
-
-   Ruby                        |      upb
-                               |
-   DescriptorPool  ------------|-----------> upb_symtab____________________
-                               |                | (message types)          \
-                               |                v                           \
-   Descriptor   ---------------|-----------> upb_msgdef         (enum types)|
-    |--> msgclass              |                |   ^                       |
-    |    (dynamically built)   |                |   | (submsg fields)       |
-    |--> MessageLayout         |                |   |                       /
-    |--------------------------|> decoder method|   |                      /
-    \--------------------------|> serialize     |   |                     /
-                               |  handlers      v   |                    /
-   FieldDescriptor  -----------|-----------> upb_fielddef               /
-                               |                    |                  /
-                               |                    v (enum fields)   /
-   EnumDescriptor  ------------|-----------> upb_enumdef  <----------'
-                               |
-                               |
-               ^               |               \___/
-               `---------------|-----------------'    (get_def_obj map)
- */
-
-// -----------------------------------------------------------------------------
-// Ruby class structure definitions.
-// -----------------------------------------------------------------------------
-
-struct DescriptorPool {
-  upb_symtab* symtab;
-};
-
-struct Descriptor {
-  const upb_msgdef* msgdef;
-  MessageLayout* layout;
-  VALUE klass;  // begins as nil
-  const upb_handlers* fill_handlers;
-  const upb_pbdecodermethod* fill_method;
-  const upb_json_parsermethod* json_fill_method;
-  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 FieldDescriptor {
-  const upb_fielddef* fielddef;
-};
-
-struct OneofDescriptor {
-  const upb_oneofdef* oneofdef;
-};
 
-struct EnumDescriptor {
-  const upb_enumdef* enumdef;
-  VALUE module;  // begins as nil
-};
-
-struct MessageBuilderContext {
-  VALUE descriptor;
-  VALUE builder;
-};
-
-struct OneofBuilderContext {
-  VALUE descriptor;
-  VALUE builder;
-};
-
-struct EnumBuilderContext {
-  VALUE enumdesc;
-};
-
-struct Builder {
-  VALUE pending_list;
-  upb_def** defs;  // used only while finalizing
-};
-
-extern VALUE cDescriptorPool;
-extern VALUE cDescriptor;
-extern VALUE cFieldDescriptor;
-extern VALUE cEnumDescriptor;
-extern VALUE cMessageBuilderContext;
-extern VALUE cOneofBuilderContext;
-extern VALUE cEnumBuilderContext;
-extern VALUE cBuilder;
-
-extern VALUE cError;
-extern VALUE cParseError;
-
-// We forward-declare all of the Ruby method implementations here because we
-// sometimes call the methods directly across .c files, rather than going
-// through Ruby's method dispatching (e.g. during message parse). It's cleaner
-// to keep the list of object methods together than to split them between
-// static-in-file definitions and header declarations.
-
-void DescriptorPool_mark(void* _self);
-void DescriptorPool_free(void* _self);
-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_lookup(VALUE _self, VALUE name);
-VALUE DescriptorPool_generated_pool(VALUE _self);
-
-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_name(VALUE _self);
-VALUE Descriptor_name_set(VALUE _self, VALUE str);
-VALUE Descriptor_each(VALUE _self);
-VALUE Descriptor_lookup(VALUE _self, VALUE name);
-VALUE Descriptor_add_field(VALUE _self, VALUE obj);
-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);
-extern const rb_data_type_t _Descriptor_type;
-
-void FieldDescriptor_mark(void* _self);
-void FieldDescriptor_free(void* _self);
-VALUE FieldDescriptor_alloc(VALUE klass);
-void FieldDescriptor_register(VALUE module);
-FieldDescriptor* ruby_to_FieldDescriptor(VALUE value);
-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_label(VALUE _self);
-VALUE FieldDescriptor_label_set(VALUE _self, VALUE label);
-VALUE FieldDescriptor_number(VALUE _self);
-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_get(VALUE _self, VALUE msg_rb);
-VALUE FieldDescriptor_set(VALUE _self, VALUE msg_rb, VALUE value);
-upb_fieldtype_t ruby_to_fieldtype(VALUE type);
-VALUE fieldtype_to_ruby(upb_fieldtype_t type);
-
-void OneofDescriptor_mark(void* _self);
-void OneofDescriptor_free(void* _self);
-VALUE OneofDescriptor_alloc(VALUE klass);
-void OneofDescriptor_register(VALUE module);
-OneofDescriptor* ruby_to_OneofDescriptor(VALUE value);
-VALUE OneofDescriptor_name(VALUE _self);
-VALUE OneofDescriptor_name_set(VALUE _self, VALUE value);
-VALUE OneofDescriptor_add_field(VALUE _self, VALUE field);
-VALUE OneofDescriptor_each(VALUE _self, VALUE field);
-
-void EnumDescriptor_mark(void* _self);
-void EnumDescriptor_free(void* _self);
-VALUE EnumDescriptor_alloc(VALUE klass);
-void EnumDescriptor_register(VALUE module);
-EnumDescriptor* ruby_to_EnumDescriptor(VALUE value);
-VALUE EnumDescriptor_name(VALUE _self);
-VALUE EnumDescriptor_name_set(VALUE _self, VALUE str);
-VALUE EnumDescriptor_add_value(VALUE _self, VALUE name, VALUE number);
-VALUE EnumDescriptor_lookup_name(VALUE _self, VALUE name);
-VALUE EnumDescriptor_lookup_value(VALUE _self, VALUE number);
-VALUE EnumDescriptor_each(VALUE _self);
-VALUE EnumDescriptor_enummodule(VALUE _self);
-extern const rb_data_type_t _EnumDescriptor_type;
-
-void MessageBuilderContext_mark(void* _self);
-void MessageBuilderContext_free(void* _self);
-VALUE MessageBuilderContext_alloc(VALUE klass);
-void MessageBuilderContext_register(VALUE module);
-MessageBuilderContext* ruby_to_MessageBuilderContext(VALUE value);
-VALUE MessageBuilderContext_initialize(VALUE _self,
-                                       VALUE descriptor,
-                                       VALUE builder);
-VALUE MessageBuilderContext_optional(int argc, VALUE* argv, VALUE _self);
-VALUE MessageBuilderContext_required(int argc, VALUE* argv, VALUE _self);
-VALUE MessageBuilderContext_repeated(int argc, VALUE* argv, VALUE _self);
-VALUE MessageBuilderContext_map(int argc, VALUE* argv, VALUE _self);
-VALUE MessageBuilderContext_oneof(VALUE _self, VALUE name);
-
-void OneofBuilderContext_mark(void* _self);
-void OneofBuilderContext_free(void* _self);
-VALUE OneofBuilderContext_alloc(VALUE klass);
-void OneofBuilderContext_register(VALUE module);
-OneofBuilderContext* ruby_to_OneofBuilderContext(VALUE value);
-VALUE OneofBuilderContext_initialize(VALUE _self,
-                                     VALUE descriptor,
-                                     VALUE builder);
-VALUE OneofBuilderContext_optional(int argc, VALUE* argv, VALUE _self);
-
-void EnumBuilderContext_mark(void* _self);
-void EnumBuilderContext_free(void* _self);
-VALUE EnumBuilderContext_alloc(VALUE klass);
-void EnumBuilderContext_register(VALUE module);
-EnumBuilderContext* ruby_to_EnumBuilderContext(VALUE value);
-VALUE EnumBuilderContext_initialize(VALUE _self, VALUE enumdesc);
-VALUE EnumBuilderContext_value(VALUE _self, VALUE name, VALUE number);
-
-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_add_message(VALUE _self, VALUE name);
-VALUE Builder_add_enum(VALUE _self, VALUE name);
-VALUE Builder_finalize_to_pool(VALUE _self, VALUE pool_rb);
-
-// -----------------------------------------------------------------------------
-// Native slot storage abstraction.
-// -----------------------------------------------------------------------------
-
-#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,
-                     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,
-                                    VALUE type_class,
-                                    void* memory,
-                                    VALUE value,
-                                    uint32_t* case_memory,
-                                    uint32_t case_number);
-VALUE native_slot_get(upb_fieldtype_t type,
-                      VALUE type_class,
-                      const void* memory);
-void native_slot_init(upb_fieldtype_t type, void* memory);
-void native_slot_mark(upb_fieldtype_t type, void* memory);
-void native_slot_dup(upb_fieldtype_t type, void* to, void* from);
-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);
-
-extern rb_encoding* kRubyStringUtf8Encoding;
-extern rb_encoding* kRubyStringASCIIEncoding;
-extern rb_encoding* kRubyString8bitEncoding;
-
-VALUE field_type_class(const upb_fielddef* field);
-
-#define MAP_KEY_FIELD 1
-#define MAP_VALUE_FIELD 2
-
-// Oneof case slot value to indicate that no oneof case is set. The value `0` is
-// safe because field numbers are used as case identifiers, and no field can
-// have a number of 0.
-#define ONEOF_CASE_NONE 0
+#include "defs.h"
+#include "ruby-upb.h"
 
 // These operate on a map field (i.e., a repeated field of submessages whose
 // submessage type is a map-entry msgdef).
-bool is_map_field(const upb_fielddef* field);
-const upb_fielddef* map_field_key(const upb_fielddef* field);
-const upb_fielddef* map_field_value(const upb_fielddef* field);
-
-// These operate on a map-entry msgdef.
-const upb_fielddef* map_entry_key(const upb_msgdef* msgdef);
-const upb_fielddef* map_entry_value(const upb_msgdef* msgdef);
+const upb_FieldDef* map_field_key(const upb_FieldDef* field);
+const upb_FieldDef* map_field_value(const upb_FieldDef* field);
 
 // -----------------------------------------------------------------------------
-// Repeated field container type.
+// Arena
 // -----------------------------------------------------------------------------
 
-typedef struct {
-  upb_fieldtype_t field_type;
-  VALUE field_type_class;
-  void* elements;
-  int size;
-  int capacity;
-} RepeatedField;
-
-void RepeatedField_mark(void* self);
-void RepeatedField_free(void* self);
-VALUE RepeatedField_alloc(VALUE klass);
-VALUE RepeatedField_init(int argc, VALUE* argv, VALUE self);
-void RepeatedField_register(VALUE module);
-
-extern const rb_data_type_t RepeatedField_type;
-extern VALUE cRepeatedField;
-
-RepeatedField* ruby_to_RepeatedField(VALUE value);
+// A Ruby object that wraps an underlying upb_Arena.  Any objects that are
+// allocated from this arena should reference the Arena in rb_gc_mark(), to
+// ensure that the object's underlying memory outlives any Ruby object that can
+// reach it.
 
-VALUE RepeatedField_each(VALUE _self);
-VALUE RepeatedField_index(int argc, VALUE* argv, VALUE _self);
-void* RepeatedField_index_native(VALUE _self, int index);
-int RepeatedField_size(VALUE _self);
-VALUE RepeatedField_index_set(VALUE _self, VALUE _index, VALUE val);
-void RepeatedField_reserve(RepeatedField* self, int new_size);
-VALUE RepeatedField_push(VALUE _self, VALUE val);
-void RepeatedField_push_native(VALUE _self, void* data);
-VALUE RepeatedField_pop_one(VALUE _self);
-VALUE RepeatedField_insert(int argc, VALUE* argv, VALUE _self);
-VALUE RepeatedField_replace(VALUE _self, VALUE list);
-VALUE RepeatedField_clear(VALUE _self);
-VALUE RepeatedField_length(VALUE _self);
-VALUE RepeatedField_dup(VALUE _self);
-VALUE RepeatedField_deep_copy(VALUE _self);
-VALUE RepeatedField_to_ary(VALUE _self);
-VALUE RepeatedField_eq(VALUE _self, VALUE _other);
-VALUE RepeatedField_hash(VALUE _self);
-VALUE RepeatedField_inspect(VALUE _self);
-VALUE RepeatedField_plus(VALUE _self, VALUE list);
-
-// Defined in repeated_field.c; also used by Map.
-void validate_type_class(upb_fieldtype_t type, VALUE klass);
-
-// -----------------------------------------------------------------------------
-// Map container type.
-// -----------------------------------------------------------------------------
+VALUE Arena_new();
+upb_Arena* Arena_get(VALUE arena);
 
-typedef struct {
-  upb_fieldtype_t key_type;
-  upb_fieldtype_t value_type;
-  VALUE value_type_class;
-  upb_strtable table;
-} Map;
+// Fuses this arena to another, throwing a Ruby exception if this is not
+// possible.
+void Arena_fuse(VALUE arena, upb_Arena* other);
 
-void Map_mark(void* self);
-void Map_free(void* self);
-VALUE Map_alloc(VALUE klass);
-VALUE Map_init(int argc, VALUE* argv, VALUE self);
-void Map_register(VALUE module);
-
-extern const rb_data_type_t Map_type;
-extern VALUE cMap;
-
-Map* ruby_to_Map(VALUE value);
-
-VALUE Map_each(VALUE _self);
-VALUE Map_keys(VALUE _self);
-VALUE Map_values(VALUE _self);
-VALUE Map_index(VALUE _self, VALUE key);
-VALUE Map_index_set(VALUE _self, VALUE key, VALUE value);
-VALUE Map_has_key(VALUE _self, VALUE key);
-VALUE Map_delete(VALUE _self, VALUE key);
-VALUE Map_clear(VALUE _self);
-VALUE Map_length(VALUE _self);
-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_inspect(VALUE _self);
-VALUE Map_merge(VALUE _self, VALUE hashmap);
-VALUE Map_merge_into_self(VALUE _self, VALUE hashmap);
-
-typedef struct {
-  Map* self;
-  upb_strtable_iter it;
-} Map_iter;
-
-void Map_begin(VALUE _self, Map_iter* iter);
-void Map_next(Map_iter* iter);
-bool Map_done(Map_iter* iter);
-VALUE Map_iter_key(Map_iter* iter);
-VALUE Map_iter_value(Map_iter* iter);
+// Pins this Ruby object to the lifetime of this arena, so that as long as the
+// arena is alive this object will not be collected.
+//
+// We use this to guarantee that the "frozen" bit on the object will be
+// remembered, even if the user drops their reference to this precise object.
+void Arena_Pin(VALUE arena, VALUE obj);
 
 // -----------------------------------------------------------------------------
-// Message layout / storage.
+// ObjectCache
 // -----------------------------------------------------------------------------
 
-#define MESSAGE_FIELD_NO_CASE ((size_t)-1)
-
-struct MessageField {
-  size_t offset;
-  size_t case_offset;  // for oneofs, a uint32. Else, MESSAGE_FIELD_NO_CASE.
-};
+// Global object cache from upb array/map/message/symtab to wrapper object.
+//
+// This is a conceptually "weak" cache, in that it does not prevent "val" from
+// being collected (though in Ruby <2.7 is it effectively strong, due to
+// implementation limitations).
 
-struct MessageLayout {
-  const upb_msgdef* msgdef;
-  MessageField* fields;
-  size_t size;
-};
+// Adds an entry to the cache. The "arena" parameter must give the arena that
+// "key" was allocated from.  In Ruby <2.7.0, it will be used to remove the key
+// from the cache when the arena is destroyed.
+void ObjectCache_Add(const void* key, VALUE val);
 
-MessageLayout* create_layout(const upb_msgdef* msgdef);
-void free_layout(MessageLayout* layout);
-VALUE layout_get(MessageLayout* layout,
-                 const void* storage,
-                 const upb_fielddef* field);
-void layout_set(MessageLayout* layout,
-                void* storage,
-                const upb_fielddef* field,
-                VALUE val);
-void layout_init(MessageLayout* layout, void* storage);
-void layout_mark(MessageLayout* layout, void* storage);
-void layout_dup(MessageLayout* layout, void* to, void* from);
-void layout_deep_copy(MessageLayout* layout, void* to, void* from);
-VALUE layout_eq(MessageLayout* layout, void* msg1, void* msg2);
-VALUE layout_hash(MessageLayout* layout, void* storage);
-VALUE layout_inspect(MessageLayout* layout, void* storage);
+// Returns the cached object for this key, if any. Otherwise returns Qnil.
+VALUE ObjectCache_Get(const void* key);
 
 // -----------------------------------------------------------------------------
-// Message class creation.
+// StringBuilder, for inspect
 // -----------------------------------------------------------------------------
 
-struct MessageHeader {
-  Descriptor* descriptor;  // kept alive by self.class.descriptor reference.
-  // Data comes after this.
-};
+struct StringBuilder;
+typedef struct StringBuilder StringBuilder;
 
-extern rb_data_type_t Message_type;
+StringBuilder* StringBuilder_New();
+void StringBuilder_Free(StringBuilder* b);
+void StringBuilder_Printf(StringBuilder* b, const char* fmt, ...);
+VALUE StringBuilder_ToRubyString(StringBuilder* b);
 
-VALUE build_class_from_descriptor(Descriptor* descriptor);
-void* Message_data(void* msg);
-void Message_mark(void* self);
-void Message_free(void* self);
-VALUE Message_alloc(VALUE klass);
-VALUE Message_method_missing(int argc, VALUE* argv, VALUE _self);
-VALUE Message_initialize(int argc, VALUE* argv, VALUE _self);
-VALUE Message_dup(VALUE _self);
-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_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_encode_json(int argc, VALUE* argv, VALUE klass);
-
-VALUE Google_Protobuf_deep_copy(VALUE self, VALUE obj);
-
-VALUE build_module_from_enumdesc(EnumDescriptor* enumdef);
-VALUE enum_lookup(VALUE self, VALUE number);
-VALUE enum_resolve(VALUE self, VALUE sym);
-
-const upb_pbdecodermethod *new_fillmsg_decodermethod(
-    Descriptor* descriptor, const void *owner);
-
-// Maximum depth allowed during encoding, to avoid stack overflows due to
-// cycles.
-#define ENCODE_MAX_NESTING 63
-
-// -----------------------------------------------------------------------------
-// Global map from upb {msg,enum}defs to wrapper Descriptor/EnumDescriptor
-// instances.
-// -----------------------------------------------------------------------------
-void add_def_obj(const void* def, VALUE value);
-VALUE get_def_obj(const void* def);
+void StringBuilder_PrintMsgval(StringBuilder* b, upb_MessageValue val,
+                               TypeInfo info);
 
 // -----------------------------------------------------------------------------
 // Utilities.
 // -----------------------------------------------------------------------------
 
-void check_upb_status(const upb_status* status, const char* msg);
+extern VALUE cTypeError;
+
+#ifdef NDEBUG
+#define PBRUBY_ASSERT(expr) \
+  do {                      \
+  } while (false && (expr))
+#else
+#define PBRUBY_ASSERT(expr) assert(expr)
+#endif
 
-#define CHECK_UPB(code, msg) do {                                             \
-    upb_status status = UPB_STATUS_INIT;                                      \
-    code;                                                                     \
-    check_upb_status(&status, msg);                                           \
-} while (0)
+#define PBRUBY_MAX(x, y) (((x) > (y)) ? (x) : (y))
 
-extern ID descriptor_instancevar_interned;
+#define UPB_UNUSED(var) (void)var
 
 #endif  // __GOOGLE_PROTOBUF_RUBY_PROTOBUF_H__