Add bin and edit workflow

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/Rakefile

@@ -0,0 +1,3 @@
+import '../../../lib/google/tasks/ffi.rake'
+
+task default: ['ffi-protobuf:default']

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/convert.c

@@ -0,0 +1,342 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.  All rights reserved.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file or at
+// https://developers.google.com/open-source/licenses/bsd
+
+// -----------------------------------------------------------------------------
+// Ruby <-> upb data conversion functions.
+//
+// This file Also contains a few other assorted algorithms on upb_MessageValue.
+//
+// None of the algorithms in this file require any access to the internal
+// representation of Ruby or upb objects.
+// -----------------------------------------------------------------------------
+
+#include "convert.h"
+
+#include "message.h"
+#include "protobuf.h"
+#include "shared_convert.h"
+
+static upb_StringView Convert_StringData(VALUE str, upb_Arena* arena) {
+  upb_StringView ret;
+  if (arena) {
+    char* ptr = upb_Arena_Malloc(arena, RSTRING_LEN(str));
+    memcpy(ptr, RSTRING_PTR(str), RSTRING_LEN(str));
+    ret.data = ptr;
+  } else {
+    // Data is only needed temporarily (within map lookup).
+    ret.data = RSTRING_PTR(str);
+  }
+  ret.size = RSTRING_LEN(str);
+  return ret;
+}
+
+static bool is_ruby_num(VALUE value) {
+  return (TYPE(value) == T_FLOAT || TYPE(value) == T_FIXNUM ||
+          TYPE(value) == T_BIGNUM);
+}
+
+static void Convert_CheckInt(const char* name, upb_CType type, VALUE val) {
+  if (!is_ruby_num(val)) {
+    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
+  // bound; we just need to do precision checks (i.e., disallow rounding) and
+  // check for < 0 on unsigned types.
+  if (TYPE(val) == T_FLOAT) {
+    double dbl_val = NUM2DBL(val);
+    if (floor(dbl_val) != dbl_val) {
+      rb_raise(rb_eRangeError,
+               "Non-integral floating point value assigned to integer field "
+               "'%s' (given %s).",
+               name, rb_class2name(CLASS_OF(val)));
+    }
+  }
+  if (type == kUpb_CType_UInt32 || type == kUpb_CType_UInt64) {
+    if (NUM2DBL(val) < 0) {
+      rb_raise(
+          rb_eRangeError,
+          "Assigning negative value to unsigned integer field '%s' (given %s).",
+          name, rb_class2name(CLASS_OF(val)));
+    }
+  }
+}
+
+static int32_t Convert_ToEnum(VALUE value, const char* name,
+                              const upb_EnumDef* e) {
+  int32_t val;
+
+  switch (TYPE(value)) {
+    case T_FLOAT:
+    case T_FIXNUM:
+    case T_BIGNUM:
+      Convert_CheckInt(name, kUpb_CType_Int32, value);
+      val = NUM2INT(value);
+      break;
+    case T_STRING: {
+      const upb_EnumValueDef* ev = upb_EnumDef_FindValueByNameWithSize(
+          e, RSTRING_PTR(value), RSTRING_LEN(value));
+      if (!ev) goto unknownval;
+      val = upb_EnumValueDef_Number(ev);
+      break;
+    }
+    case T_SYMBOL: {
+      const upb_EnumValueDef* ev =
+          upb_EnumDef_FindValueByName(e, rb_id2name(SYM2ID(value)));
+      if (!ev) goto unknownval;
+      val = upb_EnumValueDef_Number(ev);
+      break;
+    }
+    default:
+      rb_raise(cTypeError,
+               "Expected number or symbol type for enum field '%s'.", name);
+  }
+
+  return val;
+
+unknownval:
+  rb_raise(rb_eRangeError, "Unknown symbol value for enum field '%s'.", name);
+}
+
+VALUE Convert_CheckStringUtf8(VALUE str) {
+  VALUE utf8 = rb_enc_from_encoding(rb_utf8_encoding());
+
+  if (rb_obj_encoding(str) == utf8) {
+    // Note: Just because a string is marked as having UTF-8 encoding does
+    // not mean that it is *valid* UTF-8.  We have to check separately
+    // whether it is valid.
+    if (rb_enc_str_coderange(str) == ENC_CODERANGE_BROKEN) {
+      // TODO: For now
+      // we only warn for this case.  We will remove the warning and throw an
+      // exception below in the 30.x release
+
+      rb_warn(
+          "String is invalid UTF-8. This will be an error in a future "
+          "version.");
+      // VALUE exc = rb_const_get_at(
+      //     rb_cEncoding, rb_intern("InvalidByteSequenceError"));
+      // rb_raise(exc, "String is invalid UTF-8");
+    }
+  } else {
+    // Note: this will not duplicate underlying string data unless
+    // necessary.
+    //
+    // This will throw an exception if the conversion cannot be performed:
+    // - Encoding::UndefinedConversionError if certain characters cannot be
+    //   converted to UTF-8.
+    // - Encoding::InvalidByteSequenceError if certain characters were invalid
+    //   in the source encoding.
+    str = rb_str_encode(str, utf8, 0, Qnil);
+    PBRUBY_ASSERT(rb_enc_str_coderange(str) != ENC_CODERANGE_BROKEN);
+  }
+
+  return str;
+}
+
+upb_MessageValue Convert_RubyToUpb(VALUE value, const char* name,
+                                   TypeInfo type_info, upb_Arena* arena) {
+  upb_MessageValue ret;
+
+  switch (type_info.type) {
+    case kUpb_CType_Float:
+      if (!is_ruby_num(value)) {
+        rb_raise(cTypeError,
+                 "Expected number type for float field '%s' (given %s).", name,
+                 rb_class2name(CLASS_OF(value)));
+      }
+      ret.float_val = NUM2DBL(value);
+      break;
+    case kUpb_CType_Double:
+      if (!is_ruby_num(value)) {
+        rb_raise(cTypeError,
+                 "Expected number type for double field '%s' (given %s).", name,
+                 rb_class2name(CLASS_OF(value)));
+      }
+      ret.double_val = NUM2DBL(value);
+      break;
+    case kUpb_CType_Bool: {
+      if (value == Qtrue) {
+        ret.bool_val = 1;
+      } else if (value == Qfalse) {
+        ret.bool_val = 0;
+      } else {
+        rb_raise(cTypeError,
+                 "Invalid argument for boolean field '%s' (given %s).", name,
+                 rb_class2name(CLASS_OF(value)));
+      }
+      break;
+    }
+    case kUpb_CType_String:
+      if (rb_obj_class(value) == rb_cSymbol) {
+        value = rb_funcall(value, rb_intern("to_s"), 0);
+      } else if (!rb_obj_is_kind_of(value, rb_cString)) {
+        rb_raise(cTypeError,
+                 "Invalid argument for string field '%s' (given %s).", name,
+                 rb_class2name(CLASS_OF(value)));
+      }
+
+      value = Convert_CheckStringUtf8(value);
+      ret.str_val = Convert_StringData(value, arena);
+      break;
+    case kUpb_CType_Bytes: {
+      VALUE bytes = rb_enc_from_encoding(rb_ascii8bit_encoding());
+      if (rb_obj_class(value) != rb_cString) {
+        rb_raise(cTypeError,
+                 "Invalid argument for bytes field '%s' (given %s).", name,
+                 rb_class2name(CLASS_OF(value)));
+      }
+
+      if (rb_obj_encoding(value) != bytes) {
+        // Note: this will not duplicate underlying string data unless
+        // necessary.
+        // TODO: is this really necessary to get raw bytes?
+        value = rb_str_encode(value, bytes, 0, Qnil);
+      }
+
+      ret.str_val = Convert_StringData(value, arena);
+      break;
+    }
+    case kUpb_CType_Message:
+      ret.msg_val =
+          Message_GetUpbMessage(value, type_info.def.msgdef, name, arena);
+      break;
+    case kUpb_CType_Enum:
+      ret.int32_val = Convert_ToEnum(value, name, type_info.def.enumdef);
+      break;
+    case kUpb_CType_Int32:
+    case kUpb_CType_Int64:
+    case kUpb_CType_UInt32:
+    case kUpb_CType_UInt64:
+      Convert_CheckInt(name, type_info.type, value);
+      switch (type_info.type) {
+        case kUpb_CType_Int32:
+          ret.int32_val = NUM2INT(value);
+          break;
+        case kUpb_CType_Int64:
+          ret.int64_val = NUM2LL(value);
+          break;
+        case kUpb_CType_UInt32:
+          ret.uint32_val = NUM2UINT(value);
+          break;
+        case kUpb_CType_UInt64:
+          ret.uint64_val = NUM2ULL(value);
+          break;
+        default:
+          rb_raise(cTypeError, "Convert_RubyToUpb(): Unexpected type %d",
+                   (int)type_info.type);
+      }
+      break;
+    default:
+      rb_raise(cTypeError,
+                "Convert_RubyToUpb(): Unexpected type %d", (int)type_info.type);
+  }
+
+  return ret;
+}
+
+VALUE Convert_UpbToRuby(upb_MessageValue upb_val, TypeInfo type_info,
+                        VALUE arena) {
+  switch (type_info.type) {
+    case kUpb_CType_Float:
+      return DBL2NUM(upb_val.float_val);
+    case kUpb_CType_Double:
+      return DBL2NUM(upb_val.double_val);
+    case kUpb_CType_Bool:
+      return upb_val.bool_val ? Qtrue : Qfalse;
+    case kUpb_CType_Int32:
+      return INT2NUM(upb_val.int32_val);
+    case kUpb_CType_Int64:
+      return LL2NUM(upb_val.int64_val);
+    case kUpb_CType_UInt32:
+      return UINT2NUM(upb_val.uint32_val);
+    case kUpb_CType_UInt64:
+      return ULL2NUM(upb_val.int64_val);
+    case kUpb_CType_Enum: {
+      const upb_EnumValueDef* ev = upb_EnumDef_FindValueByNumber(
+          type_info.def.enumdef, upb_val.int32_val);
+      if (ev) {
+        return ID2SYM(rb_intern(upb_EnumValueDef_Name(ev)));
+      } else {
+        return INT2NUM(upb_val.int32_val);
+      }
+    }
+    case kUpb_CType_String: {
+      VALUE str_rb = rb_str_new(upb_val.str_val.data, upb_val.str_val.size);
+      rb_enc_associate(str_rb, rb_utf8_encoding());
+      rb_obj_freeze(str_rb);
+      return str_rb;
+    }
+    case kUpb_CType_Bytes: {
+      VALUE str_rb = rb_str_new(upb_val.str_val.data, upb_val.str_val.size);
+      rb_enc_associate(str_rb, rb_ascii8bit_encoding());
+      rb_obj_freeze(str_rb);
+      return str_rb;
+    }
+    case kUpb_CType_Message:
+      return Message_GetRubyWrapper((upb_Message*)upb_val.msg_val,
+                                    type_info.def.msgdef, arena);
+    default:
+      rb_raise(rb_eRuntimeError, "Convert_UpbToRuby(): Unexpected type %d",
+               (int)type_info.type);
+  }
+}
+
+upb_MessageValue Msgval_DeepCopy(upb_MessageValue msgval, TypeInfo type_info,
+                                 upb_Arena* arena) {
+  upb_MessageValue new_msgval;
+
+  switch (type_info.type) {
+    default:
+      memcpy(&new_msgval, &msgval, sizeof(msgval));
+      break;
+    case kUpb_CType_String:
+    case kUpb_CType_Bytes: {
+      size_t n = msgval.str_val.size;
+      char* mem = upb_Arena_Malloc(arena, n);
+      new_msgval.str_val.data = mem;
+      new_msgval.str_val.size = n;
+      memcpy(mem, msgval.str_val.data, n);
+      break;
+    }
+    case kUpb_CType_Message:
+      new_msgval.msg_val =
+          Message_deep_copy(msgval.msg_val, type_info.def.msgdef, arena);
+      break;
+  }
+
+  return new_msgval;
+}
+
+bool Msgval_IsEqual(upb_MessageValue val1, upb_MessageValue val2,
+                    TypeInfo type_info) {
+  upb_Status status;
+  upb_Status_Clear(&status);
+  bool return_value = shared_Msgval_IsEqual(val1, val2, type_info.type,
+                                            type_info.def.msgdef, &status);
+  if (upb_Status_IsOk(&status)) {
+    return return_value;
+  } else {
+    rb_raise(rb_eRuntimeError, "Msgval_IsEqual(): %s",
+             upb_Status_ErrorMessage(&status));
+  }
+}
+
+uint64_t Msgval_GetHash(upb_MessageValue val, TypeInfo type_info,
+                        uint64_t seed) {
+  upb_Status status;
+  upb_Status_Clear(&status);
+  uint64_t return_value = shared_Msgval_GetHash(
+      val, type_info.type, type_info.def.msgdef, seed, &status);
+  if (upb_Status_IsOk(&status)) {
+    return return_value;
+  } else {
+    rb_raise(rb_eRuntimeError, "Msgval_GetHash(): %s",
+             upb_Status_ErrorMessage(&status));
+  }
+}

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/convert.h

@@ -0,0 +1,50 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.  All rights reserved.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file or at
+// https://developers.google.com/open-source/licenses/bsd
+
+#ifndef RUBY_PROTOBUF_CONVERT_H_
+#define RUBY_PROTOBUF_CONVERT_H_
+
+#include "protobuf.h"
+#include "ruby-upb.h"
+
+// Converts |ruby_val| to a upb_MessageValue according to |type_info|.
+//
+// The |arena| parameter indicates the lifetime of the container where this
+// value will be assigned. It is used as follows:
+// - If type is string or bytes, the string data will be copied into |arena|.
+// - If type is message, and we need to auto-construct a message due to implicit
+//   conversions (eg. Time -> Google::Protobuf::Timestamp), the new message
+//   will be created in |arena|.
+// - If type is message and the Ruby value is a message instance, we will fuse
+//   the message's arena into |arena|, to ensure that this message outlives the
+//   container.
+upb_MessageValue Convert_RubyToUpb(VALUE ruby_val, const char *name,
+                                   TypeInfo type_info, upb_Arena *arena);
+
+// Converts |upb_val| to a Ruby VALUE according to |type_info|. This may involve
+// creating a Ruby wrapper object.
+//
+// The |arena| parameter indicates the arena that owns the lifetime of
+// |upb_val|. Any Ruby wrapper object that is created will reference |arena|
+// and ensure it outlives the wrapper.
+VALUE Convert_UpbToRuby(upb_MessageValue upb_val, TypeInfo type_info,
+                        VALUE arena);
+
+// Creates a deep copy of |msgval| in |arena|.
+upb_MessageValue Msgval_DeepCopy(upb_MessageValue msgval, TypeInfo type_info,
+                                 upb_Arena *arena);
+
+// Returns true if |val1| and |val2| are equal. Their type is given by
+// |type_info|.
+bool Msgval_IsEqual(upb_MessageValue val1, upb_MessageValue val2,
+                    TypeInfo type_info);
+
+// Returns a hash value for the given upb_MessageValue.
+uint64_t Msgval_GetHash(upb_MessageValue val, TypeInfo type_info,
+                        uint64_t seed);
+
+#endif  // RUBY_PROTOBUF_CONVERT_H_

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/defs.h

@@ -0,0 +1,82 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.  All rights reserved.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file or at
+// https://developers.google.com/open-source/licenses/bsd
+
+#ifndef RUBY_PROTOBUF_DEFS_H_
+#define RUBY_PROTOBUF_DEFS_H_
+
+#include "protobuf.h"
+#include "ruby-upb.h"
+
+// -----------------------------------------------------------------------------
+// TypeInfo
+// -----------------------------------------------------------------------------
+
+// This bundles a upb_CType and msgdef/enumdef when appropriate. This is
+// convenient for functions that need type information but cannot necessarily
+// assume a upb_FieldDef will be available.
+//
+// For example, Google::Protobuf::Map and Google::Protobuf::RepeatedField can
+// be constructed with type information alone:
+//
+//   # RepeatedField will internally store the type information in a TypeInfo.
+//   Google::Protobuf::RepeatedField.new(:message, FooMessage)
+
+typedef struct {
+  upb_CType type;
+  union {
+    const upb_MessageDef* msgdef;  // When type == kUpb_CType_Message
+    const upb_EnumDef* enumdef;    // When type == kUpb_CType_Enum
+  } def;
+} TypeInfo;
+
+static inline TypeInfo TypeInfo_get(const upb_FieldDef* f) {
+  TypeInfo ret = {upb_FieldDef_CType(f), {NULL}};
+  switch (ret.type) {
+    case kUpb_CType_Message:
+      ret.def.msgdef = upb_FieldDef_MessageSubDef(f);
+      break;
+    case kUpb_CType_Enum:
+      ret.def.enumdef = upb_FieldDef_EnumSubDef(f);
+      break;
+    default:
+      break;
+  }
+  return ret;
+}
+
+TypeInfo TypeInfo_FromClass(int argc, VALUE* argv, int skip_arg,
+                            VALUE* type_class, VALUE* init_arg);
+
+static inline TypeInfo TypeInfo_from_type(upb_CType type) {
+  TypeInfo ret = {type};
+  assert(type != kUpb_CType_Message && type != kUpb_CType_Enum);
+  return ret;
+}
+
+// -----------------------------------------------------------------------------
+// Other utilities
+// -----------------------------------------------------------------------------
+
+VALUE Descriptor_DefToClass(const upb_MessageDef* m);
+
+// Returns the underlying msgdef, enumdef, or symtab (respectively) for the
+// given Descriptor, EnumDescriptor, or DescriptorPool Ruby object.
+const upb_EnumDef* EnumDescriptor_GetEnumDef(VALUE enum_desc_rb);
+const upb_DefPool* DescriptorPool_GetSymtab(VALUE desc_pool_rb);
+const upb_MessageDef* Descriptor_GetMsgDef(VALUE desc_rb);
+
+// Returns a upb field type for the given Ruby symbol
+// (eg. :float => kUpb_CType_Float).
+upb_CType ruby_to_fieldtype(VALUE type);
+
+// The singleton generated pool (a DescriptorPool object).
+extern VALUE generated_pool;
+
+// Call at startup to register all types in this module.
+void Defs_register(VALUE module);
+
+#endif  // RUBY_PROTOBUF_DEFS_H_

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/extconf.rb

@@ -0,0 +1,28 @@
+#!/usr/bin/ruby
+
+require 'mkmf'
+
+ext_name = "google/protobuf_c"
+
+dir_config(ext_name)
+
+if RUBY_PLATFORM =~ /darwin/ || RUBY_PLATFORM =~ /linux/ || RUBY_PLATFORM =~ /freebsd/
+  $CFLAGS += " -std=gnu99 -O3 -DNDEBUG -fvisibility=hidden -Wall -Wsign-compare -Wno-declaration-after-statement"
+else
+  $CFLAGS += " -std=gnu99 -O3 -DNDEBUG"
+end
+
+if RUBY_PLATFORM =~ /linux/
+  # Instruct the linker to point memcpy calls at our __wrap_memcpy wrapper.
+  $LDFLAGS += " -Wl,-wrap,memcpy"
+end
+
+$VPATH << "$(srcdir)/third_party/utf8_range"
+$INCFLAGS += " -I$(srcdir)/third_party/utf8_range"
+
+$srcs = ["protobuf.c", "convert.c", "defs.c", "message.c",
+         "repeated_field.c", "map.c", "ruby-upb.c", "wrap_memcpy.c",
+         "utf8_range.c", "shared_convert.c",
+         "shared_message.c"]
+
+create_makefile(ext_name)

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/glue.c

@@ -0,0 +1,72 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2023 Google Inc.  All rights reserved.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file or at
+// https://developers.google.com/open-source/licenses/bsd
+
+// -----------------------------------------------------------------------------
+// Exposing inlined UPB functions. Strictly free of dependencies on
+// Ruby interpreter internals.
+
+#include "ruby-upb.h"
+
+upb_Arena* Arena_create() { return upb_Arena_Init(NULL, 0, &upb_alloc_global); }
+
+google_protobuf_FileDescriptorProto* FileDescriptorProto_parse(
+    const char* serialized_file_proto, size_t length, upb_Arena* arena) {
+  return google_protobuf_FileDescriptorProto_parse(serialized_file_proto,
+                                                   length, arena);
+}
+
+char* EnumDescriptor_serialized_options(const upb_EnumDef* enumdef,
+                                        size_t* size, upb_Arena* arena) {
+  const google_protobuf_EnumOptions* opts = upb_EnumDef_Options(enumdef);
+  char* serialized = google_protobuf_EnumOptions_serialize(opts, arena, size);
+  return serialized;
+}
+
+char* FileDescriptor_serialized_options(const upb_FileDef* filedef,
+                                        size_t* size, upb_Arena* arena) {
+  const google_protobuf_FileOptions* opts = upb_FileDef_Options(filedef);
+  char* serialized = google_protobuf_FileOptions_serialize(opts, arena, size);
+  return serialized;
+}
+
+char* Descriptor_serialized_options(const upb_MessageDef* msgdef, size_t* size,
+                                    upb_Arena* arena) {
+  const google_protobuf_MessageOptions* opts = upb_MessageDef_Options(msgdef);
+  char* serialized =
+      google_protobuf_MessageOptions_serialize(opts, arena, size);
+  return serialized;
+}
+
+char* OneOfDescriptor_serialized_options(const upb_OneofDef* oneofdef,
+                                         size_t* size, upb_Arena* arena) {
+  const google_protobuf_OneofOptions* opts = upb_OneofDef_Options(oneofdef);
+  char* serialized = google_protobuf_OneofOptions_serialize(opts, arena, size);
+  return serialized;
+}
+
+char* FieldDescriptor_serialized_options(const upb_FieldDef* fielddef,
+                                         size_t* size, upb_Arena* arena) {
+  const google_protobuf_FieldOptions* opts = upb_FieldDef_Options(fielddef);
+  char* serialized = google_protobuf_FieldOptions_serialize(opts, arena, size);
+  return serialized;
+}
+
+char* ServiceDescriptor_serialized_options(const upb_ServiceDef* servicedef,
+                                           size_t* size, upb_Arena* arena) {
+  const google_protobuf_ServiceOptions* opts =
+      upb_ServiceDef_Options(servicedef);
+  char* serialized =
+      google_protobuf_ServiceOptions_serialize(opts, arena, size);
+  return serialized;
+}
+
+char* MethodDescriptor_serialized_options(const upb_MethodDef* methoddef,
+                                          size_t* size, upb_Arena* arena) {
+  const google_protobuf_MethodOptions* opts = upb_MethodDef_Options(methoddef);
+  char* serialized = google_protobuf_MethodOptions_serialize(opts, arena, size);
+  return serialized;
+}

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/map.c

@@ -0,0 +1,731 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2014 Google Inc.  All rights reserved.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file or at
+// https://developers.google.com/open-source/licenses/bsd
+
+#include "convert.h"
+#include "defs.h"
+#include "message.h"
+#include "protobuf.h"
+
+// -----------------------------------------------------------------------------
+// Basic map operations on top of upb_Map.
+//
+// Note that we roll our own `Map` container here because, as for
+// `RepeatedField`, we want a strongly-typed container. This is so that any user
+// errors due to incorrect map key or value types are raised as close as
+// possible to the error site, rather than at some deferred point (e.g.,
+// serialization).
+// -----------------------------------------------------------------------------
+
+// -----------------------------------------------------------------------------
+// Map container type.
+// -----------------------------------------------------------------------------
+
+typedef struct {
+  const upb_Map* map;  // Can convert to mutable when non-frozen.
+  upb_CType key_type;
+  TypeInfo value_type_info;
+  VALUE value_type_class;
+  VALUE arena;
+} Map;
+
+static void Map_mark(void* _self) {
+  Map* self = _self;
+  rb_gc_mark(self->value_type_class);
+  rb_gc_mark(self->arena);
+}
+
+static size_t Map_memsize(const void* _self) { return sizeof(Map); }
+
+const rb_data_type_t Map_type = {
+    "Google::Protobuf::Map",
+    {Map_mark, RUBY_DEFAULT_FREE, Map_memsize},
+    .flags = RUBY_TYPED_FREE_IMMEDIATELY,
+};
+
+VALUE cMap;
+
+static Map* ruby_to_Map(VALUE _self) {
+  Map* self;
+  TypedData_Get_Struct(_self, Map, &Map_type, self);
+  return self;
+}
+
+static VALUE Map_alloc(VALUE klass) {
+  Map* self = ALLOC(Map);
+  self->map = NULL;
+  self->value_type_class = Qnil;
+  self->value_type_info.def.msgdef = NULL;
+  self->arena = Qnil;
+  return TypedData_Wrap_Struct(klass, &Map_type, self);
+}
+
+VALUE Map_GetRubyWrapper(const upb_Map* map, upb_CType key_type,
+                         TypeInfo value_type, VALUE arena) {
+  PBRUBY_ASSERT(map);
+  PBRUBY_ASSERT(arena != Qnil);
+
+  VALUE val = ObjectCache_Get(map);
+
+  if (val == Qnil) {
+    val = Map_alloc(cMap);
+    Map* self;
+    TypedData_Get_Struct(val, Map, &Map_type, self);
+    self->map = map;
+    self->arena = arena;
+    self->key_type = key_type;
+    self->value_type_info = value_type;
+    if (self->value_type_info.type == kUpb_CType_Message) {
+      const upb_MessageDef* val_m = self->value_type_info.def.msgdef;
+      self->value_type_class = Descriptor_DefToClass(val_m);
+    }
+    return ObjectCache_TryAdd(map, val);
+  }
+  return val;
+}
+
+static VALUE Map_new_this_type(Map* from) {
+  VALUE arena_rb = Arena_new();
+  upb_Map* map = upb_Map_New(Arena_get(arena_rb), from->key_type,
+                             from->value_type_info.type);
+  VALUE ret =
+      Map_GetRubyWrapper(map, from->key_type, from->value_type_info, arena_rb);
+  PBRUBY_ASSERT(ruby_to_Map(ret)->value_type_class == from->value_type_class);
+  return ret;
+}
+
+static TypeInfo Map_keyinfo(Map* self) {
+  TypeInfo ret;
+  ret.type = self->key_type;
+  ret.def.msgdef = NULL;
+  return ret;
+}
+
+static upb_Map* Map_GetMutable(VALUE _self) {
+  const upb_Map* map = ruby_to_Map(_self)->map;
+  Protobuf_CheckNotFrozen(_self, upb_Map_IsFrozen(map));
+  return (upb_Map*)map;
+}
+
+VALUE Map_CreateHash(const upb_Map* map, upb_CType key_type,
+                     TypeInfo val_info) {
+  VALUE hash = rb_hash_new();
+  TypeInfo key_info = TypeInfo_from_type(key_type);
+
+  if (!map) return hash;
+
+  size_t iter = kUpb_Map_Begin;
+  upb_MessageValue key, val;
+  while (upb_Map_Next(map, &key, &val, &iter)) {
+    VALUE key_val = Convert_UpbToRuby(key, key_info, Qnil);
+    VALUE val_val = Scalar_CreateHash(val, val_info);
+    rb_hash_aset(hash, key_val, val_val);
+  }
+
+  return hash;
+}
+
+VALUE Map_deep_copy(VALUE obj) {
+  Map* self = ruby_to_Map(obj);
+  VALUE new_arena_rb = Arena_new();
+  upb_Arena* arena = Arena_get(new_arena_rb);
+  upb_Map* new_map =
+      upb_Map_New(arena, self->key_type, self->value_type_info.type);
+  size_t iter = kUpb_Map_Begin;
+  upb_MessageValue key, val;
+  while (upb_Map_Next(self->map, &key, &val, &iter)) {
+    upb_MessageValue val_copy =
+        Msgval_DeepCopy(val, self->value_type_info, arena);
+    upb_Map_Set(new_map, key, val_copy, arena);
+  }
+
+  return Map_GetRubyWrapper(new_map, self->key_type, self->value_type_info,
+                            new_arena_rb);
+}
+
+const upb_Map* Map_GetUpbMap(VALUE val, const upb_FieldDef* field,
+                             upb_Arena* arena) {
+  const upb_FieldDef* key_field = map_field_key(field);
+  const upb_FieldDef* value_field = map_field_value(field);
+  TypeInfo value_type_info = TypeInfo_get(value_field);
+  Map* self;
+
+  if (!RB_TYPE_P(val, T_DATA) || !RTYPEDDATA_P(val) ||
+      RTYPEDDATA_TYPE(val) != &Map_type) {
+    rb_raise(cTypeError, "Expected Map instance");
+  }
+
+  self = ruby_to_Map(val);
+  if (self->key_type != upb_FieldDef_CType(key_field)) {
+    rb_raise(cTypeError, "Map key type does not match field's key type");
+  }
+  if (self->value_type_info.type != value_type_info.type) {
+    rb_raise(cTypeError, "Map value type does not match field's value type");
+  }
+  if (self->value_type_info.def.msgdef != value_type_info.def.msgdef) {
+    rb_raise(cTypeError, "Map value type has wrong message/enum class");
+  }
+
+  Arena_fuse(self->arena, arena);
+  return self->map;
+}
+
+void Map_Inspect(StringBuilder* b, const upb_Map* map, upb_CType key_type,
+                 TypeInfo val_type) {
+  bool first = true;
+  TypeInfo key_type_info = {key_type};
+  StringBuilder_Printf(b, "{");
+  if (map) {
+    size_t iter = kUpb_Map_Begin;
+    upb_MessageValue key, val;
+    while (upb_Map_Next(map, &key, &val, &iter)) {
+      if (first) {
+        first = false;
+      } else {
+        StringBuilder_Printf(b, ", ");
+      }
+      StringBuilder_PrintMsgval(b, key, key_type_info);
+      StringBuilder_Printf(b, "=>");
+      StringBuilder_PrintMsgval(b, val, val_type);
+    }
+  }
+  StringBuilder_Printf(b, "}");
+}
+
+static int merge_into_self_callback(VALUE key, VALUE val, VALUE _self) {
+  Map* self = ruby_to_Map(_self);
+  upb_Arena* arena = Arena_get(self->arena);
+  upb_MessageValue key_val =
+      Convert_RubyToUpb(key, "", Map_keyinfo(self), arena);
+  upb_MessageValue val_val =
+      Convert_RubyToUpb(val, "", self->value_type_info, arena);
+  upb_Map_Set(Map_GetMutable(_self), key_val, val_val, arena);
+  return ST_CONTINUE;
+}
+
+// Used only internally -- shared by #merge and #initialize.
+static VALUE Map_merge_into_self(VALUE _self, VALUE hashmap) {
+  if (TYPE(hashmap) == T_HASH) {
+    rb_hash_foreach(hashmap, merge_into_self_callback, _self);
+  } else if (RB_TYPE_P(hashmap, T_DATA) && RTYPEDDATA_P(hashmap) &&
+             RTYPEDDATA_TYPE(hashmap) == &Map_type) {
+    Map* self = ruby_to_Map(_self);
+    Map* other = ruby_to_Map(hashmap);
+    upb_Arena* arena = Arena_get(self->arena);
+    upb_Map* self_map = Map_GetMutable(_self);
+
+    Arena_fuse(other->arena, arena);
+
+    if (self->key_type != other->key_type ||
+        self->value_type_info.type != other->value_type_info.type ||
+        self->value_type_class != other->value_type_class) {
+      rb_raise(rb_eArgError, "Attempt to merge Map with mismatching types");
+    }
+
+    size_t iter = kUpb_Map_Begin;
+    upb_MessageValue key, val;
+    while (upb_Map_Next(other->map, &key, &val, &iter)) {
+      upb_Map_Set(self_map, key, val, arena);
+    }
+  } else {
+    rb_raise(rb_eArgError, "Unknown type merging into Map");
+  }
+  return _self;
+}
+
+/*
+ * call-seq:
+ *     Map.new(key_type, value_type, value_typeclass = nil, init_hashmap = {})
+ *     => new map
+ *
+ * Allocates a new Map container. This constructor may be called with 2, 3, or 4
+ * arguments. The first two arguments are always present and are symbols (taking
+ * on the same values as field-type symbols in message descriptors) that
+ * indicate the type of the map key and value fields.
+ *
+ * The supported key types are: :int32, :int64, :uint32, :uint64, :bool,
+ * :string, :bytes.
+ *
+ * The supported value types are: :int32, :int64, :uint32, :uint64, :bool,
+ * :string, :bytes, :enum, :message.
+ *
+ * The third argument, value_typeclass, must be present if value_type is :enum
+ * or :message. As in RepeatedField#new, this argument must be a message class
+ * (for :message) or enum module (for :enum).
+ *
+ * The last argument, if present, provides initial content for map. Note that
+ * this may be an ordinary Ruby hashmap or another Map instance with identical
+ * key and value types. Also note that this argument may be present whether or
+ * not value_typeclass is present (and it is unambiguously separate from
+ * value_typeclass because value_typeclass's presence is strictly determined by
+ * value_type). The contents of this initial hashmap or Map instance are
+ * shallow-copied into the new Map: the original map is unmodified, but
+ * references to underlying objects will be shared if the value type is a
+ * message type.
+ */
+static VALUE Map_init(int argc, VALUE* argv, VALUE _self) {
+  Map* self = ruby_to_Map(_self);
+  VALUE init_arg;
+
+  // We take either two args (:key_type, :value_type), three args (:key_type,
+  // :value_type, "ValueMessageType"), or four args (the above plus an initial
+  // hashmap).
+  if (argc < 2 || argc > 4) {
+    rb_raise(rb_eArgError, "Map constructor expects 2, 3 or 4 arguments.");
+  }
+
+  self->key_type = ruby_to_fieldtype(argv[0]);
+  self->value_type_info =
+      TypeInfo_FromClass(argc, argv, 1, &self->value_type_class, &init_arg);
+  self->arena = Arena_new();
+
+  // Check that the key type is an allowed type.
+  switch (self->key_type) {
+    case kUpb_CType_Int32:
+    case kUpb_CType_Int64:
+    case kUpb_CType_UInt32:
+    case kUpb_CType_UInt64:
+    case kUpb_CType_Bool:
+    case kUpb_CType_String:
+    case kUpb_CType_Bytes:
+      // These are OK.
+      break;
+    default:
+      rb_raise(rb_eArgError, "Invalid key type for map.");
+  }
+
+  self->map = upb_Map_New(Arena_get(self->arena), self->key_type,
+                          self->value_type_info.type);
+  VALUE stored = ObjectCache_TryAdd(self->map, _self);
+  (void)stored;
+  PBRUBY_ASSERT(stored == _self);
+
+  if (init_arg != Qnil) {
+    Map_merge_into_self(_self, init_arg);
+  }
+
+  return Qnil;
+}
+
+/*
+ * call-seq:
+ *     Map.each(&block)
+ *
+ * Invokes &block on each |key, value| pair in the map, in unspecified order.
+ * Note that Map also includes Enumerable; map thus acts like a normal Ruby
+ * sequence.
+ */
+static VALUE Map_each(VALUE _self) {
+  Map* self = ruby_to_Map(_self);
+  size_t iter = kUpb_Map_Begin;
+  upb_MessageValue key, val;
+
+  while (upb_Map_Next(self->map, &key, &val, &iter)) {
+    VALUE key_val = Convert_UpbToRuby(key, Map_keyinfo(self), self->arena);
+    VALUE val_val = Convert_UpbToRuby(val, self->value_type_info, self->arena);
+    rb_yield_values(2, key_val, val_val);
+  }
+
+  return Qnil;
+}
+
+/*
+ * call-seq:
+ *     Map.keys => [list_of_keys]
+ *
+ * Returns the list of keys contained in the map, in unspecified order.
+ */
+static VALUE Map_keys(VALUE _self) {
+  Map* self = ruby_to_Map(_self);
+  size_t iter = kUpb_Map_Begin;
+  VALUE ret = rb_ary_new();
+  upb_MessageValue key, val;
+
+  while (upb_Map_Next(self->map, &key, &val, &iter)) {
+    VALUE key_val = Convert_UpbToRuby(key, Map_keyinfo(self), self->arena);
+    rb_ary_push(ret, key_val);
+  }
+
+  return ret;
+}
+
+/*
+ * call-seq:
+ *     Map.values => [list_of_values]
+ *
+ * Returns the list of values contained in the map, in unspecified order.
+ */
+static VALUE Map_values(VALUE _self) {
+  Map* self = ruby_to_Map(_self);
+  size_t iter = kUpb_Map_Begin;
+  VALUE ret = rb_ary_new();
+  upb_MessageValue key, val;
+
+  while (upb_Map_Next(self->map, &key, &val, &iter)) {
+    VALUE val_val = Convert_UpbToRuby(val, self->value_type_info, self->arena);
+    rb_ary_push(ret, val_val);
+  }
+
+  return ret;
+}
+
+/*
+ * call-seq:
+ *     Map.[](key) => value
+ *
+ * Accesses the element at the given key. Throws an exception if the key type is
+ * incorrect. Returns nil when the key is not present in the map.
+ */
+static VALUE Map_index(VALUE _self, VALUE key) {
+  Map* self = ruby_to_Map(_self);
+  upb_MessageValue key_upb =
+      Convert_RubyToUpb(key, "", Map_keyinfo(self), NULL);
+  upb_MessageValue val;
+
+  if (upb_Map_Get(self->map, key_upb, &val)) {
+    return Convert_UpbToRuby(val, self->value_type_info, self->arena);
+  } else {
+    return Qnil;
+  }
+}
+
+/*
+ * call-seq:
+ *     Map.[]=(key, value) => value
+ *
+ * Inserts or overwrites the value at the given key with the given new value.
+ * Throws an exception if the key type is incorrect. Returns the new value that
+ * was just inserted.
+ */
+static VALUE Map_index_set(VALUE _self, VALUE key, VALUE val) {
+  Map* self = ruby_to_Map(_self);
+  upb_Arena* arena = Arena_get(self->arena);
+  upb_MessageValue key_upb =
+      Convert_RubyToUpb(key, "", Map_keyinfo(self), NULL);
+  upb_MessageValue val_upb =
+      Convert_RubyToUpb(val, "", self->value_type_info, arena);
+
+  upb_Map_Set(Map_GetMutable(_self), key_upb, val_upb, arena);
+
+  return val;
+}
+
+/*
+ * call-seq:
+ *     Map.has_key?(key) => bool
+ *
+ * Returns true if the given key is present in the map. Throws an exception if
+ * the key has the wrong type.
+ */
+static VALUE Map_has_key(VALUE _self, VALUE key) {
+  Map* self = ruby_to_Map(_self);
+  upb_MessageValue key_upb =
+      Convert_RubyToUpb(key, "", Map_keyinfo(self), NULL);
+
+  if (upb_Map_Get(self->map, key_upb, NULL)) {
+    return Qtrue;
+  } else {
+    return Qfalse;
+  }
+}
+
+/*
+ * call-seq:
+ *     Map.delete(key) => old_value
+ *
+ * Deletes the value at the given key, if any, returning either the old value or
+ * nil if none was present. Throws an exception if the key is of the wrong type.
+ */
+static VALUE Map_delete(VALUE _self, VALUE key) {
+  upb_Map* map = Map_GetMutable(_self);
+  Map* self = ruby_to_Map(_self);
+
+  upb_MessageValue key_upb =
+      Convert_RubyToUpb(key, "", Map_keyinfo(self), NULL);
+  upb_MessageValue val_upb;
+
+  if (upb_Map_Delete(map, key_upb, &val_upb)) {
+    return Convert_UpbToRuby(val_upb, self->value_type_info, self->arena);
+  } else {
+    return Qnil;
+  }
+}
+
+/*
+ * call-seq:
+ *     Map.clear
+ *
+ * Removes all entries from the map.
+ */
+static VALUE Map_clear(VALUE _self) {
+  upb_Map_Clear(Map_GetMutable(_self));
+  return Qnil;
+}
+
+/*
+ * call-seq:
+ *     Map.length
+ *
+ * Returns the number of entries (key-value pairs) in the map.
+ */
+static VALUE Map_length(VALUE _self) {
+  Map* self = ruby_to_Map(_self);
+  return ULL2NUM(upb_Map_Size(self->map));
+}
+
+/*
+ * call-seq:
+ *     Map.dup => new_map
+ *
+ * Duplicates this map with a shallow copy. References to all non-primitive
+ * element objects (e.g., submessages) are shared.
+ */
+static VALUE Map_dup(VALUE _self) {
+  Map* self = ruby_to_Map(_self);
+  VALUE new_map_rb = Map_new_this_type(self);
+  Map* new_self = ruby_to_Map(new_map_rb);
+  size_t iter = kUpb_Map_Begin;
+  upb_Arena* arena = Arena_get(new_self->arena);
+  upb_Map* new_map = Map_GetMutable(new_map_rb);
+
+  Arena_fuse(self->arena, arena);
+
+  upb_MessageValue key, val;
+  while (upb_Map_Next(self->map, &key, &val, &iter)) {
+    upb_Map_Set(new_map, key, val, arena);
+  }
+
+  return new_map_rb;
+}
+
+/*
+ * call-seq:
+ *     Map.==(other) => boolean
+ *
+ * Compares this map to another. Maps are equal if they have identical key sets,
+ * and for each key, the values in both maps compare equal. Elements are
+ * compared as per normal Ruby semantics, by calling their :== methods (or
+ * performing a more efficient comparison for primitive types).
+ *
+ * Maps with dissimilar key types or value types/typeclasses are never equal,
+ * even if value comparison (for example, between integers and floats) would
+ * have otherwise indicated that every element has equal value.
+ */
+VALUE Map_eq(VALUE _self, VALUE _other) {
+  Map* self = ruby_to_Map(_self);
+  Map* other;
+
+  // Allow comparisons to Ruby hashmaps by converting to a temporary Map
+  // instance. Slow, but workable.
+  if (TYPE(_other) == T_HASH) {
+    VALUE other_map = Map_new_this_type(self);
+    Map_merge_into_self(other_map, _other);
+    _other = other_map;
+  }
+
+  other = ruby_to_Map(_other);
+
+  if (self == other) {
+    return Qtrue;
+  }
+  if (self->key_type != other->key_type ||
+      self->value_type_info.type != other->value_type_info.type ||
+      self->value_type_class != other->value_type_class) {
+    return Qfalse;
+  }
+  if (upb_Map_Size(self->map) != upb_Map_Size(other->map)) {
+    return Qfalse;
+  }
+
+  // For each member of self, check that an equal member exists at the same key
+  // in other.
+  size_t iter = kUpb_Map_Begin;
+  upb_MessageValue key, val;
+  while (upb_Map_Next(self->map, &key, &val, &iter)) {
+    upb_MessageValue other_val;
+    if (!upb_Map_Get(other->map, key, &other_val)) {
+      // Not present in other map.
+      return Qfalse;
+    }
+    if (!Msgval_IsEqual(val, other_val, self->value_type_info)) {
+      // Present but different value.
+      return Qfalse;
+    }
+  }
+
+  return Qtrue;
+}
+
+/*
+ * call-seq:
+ *     Map.frozen? => bool
+ *
+ * Returns true if the map is frozen in either Ruby or the underlying
+ * representation. Freezes the Ruby map object if it is not already frozen in
+ * Ruby but it is frozen in the underlying representation.
+ */
+VALUE Map_frozen(VALUE _self) {
+  Map* self = ruby_to_Map(_self);
+  if (!upb_Map_IsFrozen(self->map)) {
+    PBRUBY_ASSERT(!RB_OBJ_FROZEN(_self));
+    return Qfalse;
+  }
+
+  // Lazily freeze the Ruby wrapper.
+  if (!RB_OBJ_FROZEN(_self)) RB_OBJ_FREEZE(_self);
+  return Qtrue;
+}
+
+/*
+ * call-seq:
+ *     Map.freeze => self
+ *
+ * Freezes the map object. We have to intercept this so we can freeze the
+ * underlying representation, not just the Ruby wrapper.
+ */
+VALUE Map_freeze(VALUE _self) {
+  Map* self = ruby_to_Map(_self);
+  if (RB_OBJ_FROZEN(_self)) {
+    PBRUBY_ASSERT(upb_Map_IsFrozen(self->map));
+    return _self;
+  }
+
+  if (!upb_Map_IsFrozen(self->map)) {
+    if (self->value_type_info.type == kUpb_CType_Message) {
+      upb_Map_Freeze(
+          Map_GetMutable(_self),
+          upb_MessageDef_MiniTable(self->value_type_info.def.msgdef));
+    } else {
+      upb_Map_Freeze(Map_GetMutable(_self), NULL);
+    }
+  }
+
+  RB_OBJ_FREEZE(_self);
+
+  return _self;
+}
+
+VALUE Map_EmptyFrozen(const upb_FieldDef* f) {
+  PBRUBY_ASSERT(upb_FieldDef_IsMap(f));
+  VALUE val = ObjectCache_Get(f);
+
+  if (val == Qnil) {
+    const upb_FieldDef* key_f = map_field_key(f);
+    const upb_FieldDef* val_f = map_field_value(f);
+    upb_CType key_type = upb_FieldDef_CType(key_f);
+    TypeInfo value_type_info = TypeInfo_get(val_f);
+    val = Map_alloc(cMap);
+    Map* self;
+    TypedData_Get_Struct(val, Map, &Map_type, self);
+    self->arena = Arena_new();
+    self->map =
+        upb_Map_New(Arena_get(self->arena), key_type, value_type_info.type);
+    self->key_type = key_type;
+    self->value_type_info = value_type_info;
+    if (self->value_type_info.type == kUpb_CType_Message) {
+      const upb_MessageDef* val_m = value_type_info.def.msgdef;
+      self->value_type_class = Descriptor_DefToClass(val_m);
+    }
+    return ObjectCache_TryAdd(f, Map_freeze(val));
+  }
+  PBRUBY_ASSERT(RB_OBJ_FROZEN(val));
+  PBRUBY_ASSERT(upb_Map_IsFrozen(ruby_to_Map(val)->map));
+  return val;
+}
+
+/*
+ * call-seq:
+ *     Map.hash => hash_value
+ *
+ * Returns a hash value based on this map's contents.
+ */
+VALUE Map_hash(VALUE _self) {
+  Map* self = ruby_to_Map(_self);
+  uint64_t hash = 0;
+
+  size_t iter = kUpb_Map_Begin;
+  TypeInfo key_info = {self->key_type};
+  upb_MessageValue key, val;
+  while (upb_Map_Next(self->map, &key, &val, &iter)) {
+    hash = Msgval_GetHash(key, key_info, hash);
+    hash = Msgval_GetHash(val, self->value_type_info, hash);
+  }
+
+  return LL2NUM(hash);
+}
+
+/*
+ * call-seq:
+ *     Map.to_h => {}
+ *
+ * Returns a Ruby Hash object containing all the values within the map
+ */
+VALUE Map_to_h(VALUE _self) {
+  Map* self = ruby_to_Map(_self);
+  return Map_CreateHash(self->map, self->key_type, self->value_type_info);
+}
+
+/*
+ * call-seq:
+ *     Map.inspect => string
+ *
+ * Returns a string representing this map's elements. It will be formatted as
+ * "{key => value, key => value, ...}", with each key and value string
+ * representation computed by its own #inspect method.
+ */
+VALUE Map_inspect(VALUE _self) {
+  Map* self = ruby_to_Map(_self);
+
+  StringBuilder* builder = StringBuilder_New();
+  Map_Inspect(builder, self->map, self->key_type, self->value_type_info);
+  VALUE ret = StringBuilder_ToRubyString(builder);
+  StringBuilder_Free(builder);
+  return ret;
+}
+
+/*
+ * call-seq:
+ *     Map.merge(other_map) => map
+ *
+ * Copies key/value pairs from other_map into a copy of this map. If a key is
+ * set in other_map and this map, the value from other_map overwrites the value
+ * in the new copy of this map. Returns the new copy of this map with merged
+ * contents.
+ */
+static VALUE Map_merge(VALUE _self, VALUE hashmap) {
+  VALUE dupped = Map_dup(_self);
+  return Map_merge_into_self(dupped, hashmap);
+}
+
+void Map_register(VALUE module) {
+  VALUE klass = rb_define_class_under(module, "Map", rb_cObject);
+  rb_define_alloc_func(klass, Map_alloc);
+  rb_gc_register_address(&cMap);
+  cMap = klass;
+
+  rb_define_method(klass, "initialize", Map_init, -1);
+  rb_define_method(klass, "each", Map_each, 0);
+  rb_define_method(klass, "keys", Map_keys, 0);
+  rb_define_method(klass, "values", Map_values, 0);
+  rb_define_method(klass, "[]", Map_index, 1);
+  rb_define_method(klass, "[]=", Map_index_set, 2);
+  rb_define_method(klass, "has_key?", Map_has_key, 1);
+  rb_define_method(klass, "delete", Map_delete, 1);
+  rb_define_method(klass, "clear", Map_clear, 0);
+  rb_define_method(klass, "length", Map_length, 0);
+  rb_define_method(klass, "size", Map_length, 0);
+  rb_define_method(klass, "dup", Map_dup, 0);
+  // Also define #clone so that we don't inherit Object#clone.
+  rb_define_method(klass, "clone", Map_dup, 0);
+  rb_define_method(klass, "==", Map_eq, 1);
+  rb_define_method(klass, "freeze", Map_freeze, 0);
+  rb_define_method(klass, "frozen?", Map_frozen, 0);
+  rb_define_method(klass, "hash", Map_hash, 0);
+  rb_define_method(klass, "to_h", Map_to_h, 0);
+  rb_define_method(klass, "inspect", Map_inspect, 0);
+  rb_define_method(klass, "merge", Map_merge, 1);
+  rb_include_module(klass, rb_mEnumerable);
+}

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/map.h

@@ -0,0 +1,48 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.  All rights reserved.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file or at
+// https://developers.google.com/open-source/licenses/bsd
+
+#ifndef RUBY_PROTOBUF_MAP_H_
+#define RUBY_PROTOBUF_MAP_H_
+
+#include "protobuf.h"
+#include "ruby-upb.h"
+
+// Returns a frozen sentinel Ruby wrapper object for an empty upb_Map with the
+// key and value types specified by the field. Creates one if it doesn't exist.
+VALUE Map_EmptyFrozen(const upb_FieldDef* f);
+
+// Returns a Ruby wrapper object for the given map, which will be created if
+// one does not exist already.
+VALUE Map_GetRubyWrapper(const upb_Map *map, upb_CType key_type,
+                         TypeInfo value_type, VALUE arena);
+
+// Gets the underlying upb_Map for this Ruby map object, which must have
+// key/value type that match |field|. If this is not a map or the type doesn't
+// match, raises an exception.
+const upb_Map *Map_GetUpbMap(VALUE val, const upb_FieldDef *field,
+                             upb_Arena *arena);
+
+// Implements #inspect for this map by appending its contents to |b|.
+void Map_Inspect(StringBuilder *b, const upb_Map *map, upb_CType key_type,
+                 TypeInfo val_type);
+
+// Returns a new Hash object containing the contents of this Map.
+VALUE Map_CreateHash(const upb_Map *map, upb_CType key_type, TypeInfo val_info);
+
+// Returns a deep copy of this Map object.
+VALUE Map_deep_copy(VALUE obj);
+
+// Ruby class of Google::Protobuf::Map.
+extern VALUE cMap;
+
+// Call at startup to register all types in this module.
+void Map_register(VALUE module);
+
+// Recursively freeze map
+VALUE Map_freeze(VALUE _self);
+
+#endif  // RUBY_PROTOBUF_MAP_H_

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/message.h

@@ -0,0 +1,82 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.  All rights reserved.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file or at
+// https://developers.google.com/open-source/licenses/bsd
+
+#ifndef RUBY_PROTOBUF_MESSAGE_H_
+#define RUBY_PROTOBUF_MESSAGE_H_
+
+#include "protobuf.h"
+#include "ruby-upb.h"
+
+// Gets the underlying upb_Message* and upb_MessageDef for the given Ruby
+// message wrapper. Requires that |value| is indeed a message object.
+const upb_Message* Message_Get(VALUE value, const upb_MessageDef** m);
+
+// Like Message_Get(), but checks that the object is not frozen and returns a
+// mutable pointer.
+upb_Message* Message_GetMutable(VALUE value, const upb_MessageDef** m);
+
+// Returns the Arena object for this message.
+VALUE Message_GetArena(VALUE value);
+
+// Converts |value| into a upb_Message value of the expected upb_MessageDef
+// type, raising an error if this is not possible. Used when assigning |value|
+// to a field of another message, which means the message must be of a
+// particular type.
+//
+// This will perform automatic conversions in some cases (for example, Time ->
+// Google::Protobuf::Timestamp). If any new message is created, it will be
+// created on |arena|, and any existing message will have its arena fused with
+// |arena|.
+const upb_Message* Message_GetUpbMessage(VALUE value, const upb_MessageDef* m,
+                                         const char* name, upb_Arena* arena);
+
+// Gets or constructs a Ruby wrapper object for the given message. The wrapper
+// object will reference |arena| and ensure that it outlives this object.
+VALUE Message_GetRubyWrapper(const upb_Message* msg, const upb_MessageDef* m,
+                             VALUE arena);
+
+// Gets the given field from this message.
+VALUE Message_getfield(VALUE _self, const upb_FieldDef* f);
+
+// Implements #inspect for this message, printing the text to |b|.
+void Message_PrintMessage(StringBuilder* b, const upb_Message* msg,
+                          const upb_MessageDef* m);
+
+// Returns a hash value for the given message.
+uint64_t Message_Hash(const upb_Message* msg, const upb_MessageDef* m,
+                      uint64_t seed);
+
+// Returns a deep copy of the given message.
+upb_Message* Message_deep_copy(const upb_Message* msg, const upb_MessageDef* m,
+                               upb_Arena* arena);
+
+// Checks that this Ruby object is a message, and raises an exception if not.
+void Message_CheckClass(VALUE klass);
+
+// Returns a new Hash object containing the contents of this message.
+VALUE Scalar_CreateHash(upb_MessageValue val, TypeInfo type_info);
+
+// Creates a message class or enum module for this descriptor, respectively.
+VALUE build_class_from_descriptor(VALUE descriptor);
+VALUE build_module_from_enumdesc(VALUE _enumdesc);
+
+// Returns the Descriptor/EnumDescriptor for the given message class or enum
+// module, respectively. Returns nil if this is not a message class or enum
+// module.
+VALUE MessageOrEnum_GetDescriptor(VALUE klass);
+
+// Decodes a Message from a byte sequence.
+VALUE Message_decode_bytes(int size, const char* bytes, int options,
+                           VALUE klass, bool freeze);
+
+// Recursively freeze message
+VALUE Message_freeze(VALUE _self);
+
+// Call at startup to register all types in this module.
+void Message_register(VALUE protobuf);
+
+#endif  // RUBY_PROTOBUF_MESSAGE_H_

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/protobuf.c

@@ -0,0 +1,356 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2014 Google Inc.  All rights reserved.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file or at
+// https://developers.google.com/open-source/licenses/bsd
+
+#include "protobuf.h"
+
+#include "defs.h"
+#include "map.h"
+#include "message.h"
+#include "repeated_field.h"
+
+VALUE cParseError;
+VALUE cTypeError;
+
+const upb_FieldDef *map_field_key(const upb_FieldDef *field) {
+  const upb_MessageDef *entry = upb_FieldDef_MessageSubDef(field);
+  return upb_MessageDef_FindFieldByNumber(entry, 1);
+}
+
+const upb_FieldDef *map_field_value(const upb_FieldDef *field) {
+  const upb_MessageDef *entry = upb_FieldDef_MessageSubDef(field);
+  return upb_MessageDef_FindFieldByNumber(entry, 2);
+}
+
+// -----------------------------------------------------------------------------
+// StringBuilder, for inspect
+// -----------------------------------------------------------------------------
+
+struct StringBuilder {
+  size_t size;
+  size_t cap;
+  char *data;
+};
+
+typedef struct StringBuilder StringBuilder;
+
+static size_t StringBuilder_SizeOf(size_t cap) {
+  return sizeof(StringBuilder) + cap;
+}
+
+StringBuilder *StringBuilder_New() {
+  const size_t cap = 128;
+  StringBuilder *builder = malloc(sizeof(*builder));
+  builder->size = 0;
+  builder->cap = cap;
+  builder->data = malloc(builder->cap);
+  return builder;
+}
+
+void StringBuilder_Free(StringBuilder *b) {
+  free(b->data);
+  free(b);
+}
+
+void StringBuilder_Printf(StringBuilder *b, const char *fmt, ...) {
+  size_t have = b->cap - b->size;
+  size_t n;
+  va_list args;
+
+  va_start(args, fmt);
+  n = vsnprintf(&b->data[b->size], have, fmt, args);
+  va_end(args);
+
+  if (have <= n) {
+    while (have <= n) {
+      b->cap *= 2;
+      have = b->cap - b->size;
+    }
+    b->data = realloc(b->data, StringBuilder_SizeOf(b->cap));
+    va_start(args, fmt);
+    n = vsnprintf(&b->data[b->size], have, fmt, args);
+    va_end(args);
+    PBRUBY_ASSERT(n < have);
+  }
+
+  b->size += n;
+}
+
+VALUE StringBuilder_ToRubyString(StringBuilder *b) {
+  VALUE ret = rb_str_new(b->data, b->size);
+  rb_enc_associate(ret, rb_utf8_encoding());
+  return ret;
+}
+
+static void StringBuilder_PrintEnum(StringBuilder *b, int32_t val,
+                                    const upb_EnumDef *e) {
+  const upb_EnumValueDef *ev = upb_EnumDef_FindValueByNumber(e, val);
+  if (ev) {
+    StringBuilder_Printf(b, ":%s", upb_EnumValueDef_Name(ev));
+  } else {
+    StringBuilder_Printf(b, "%" PRId32, val);
+  }
+}
+
+void StringBuilder_PrintMsgval(StringBuilder *b, upb_MessageValue val,
+                               TypeInfo info) {
+  switch (info.type) {
+    case kUpb_CType_Bool:
+      StringBuilder_Printf(b, "%s", val.bool_val ? "true" : "false");
+      break;
+    case kUpb_CType_Float: {
+      VALUE str = rb_inspect(DBL2NUM(val.float_val));
+      StringBuilder_Printf(b, "%s", RSTRING_PTR(str));
+      break;
+    }
+    case kUpb_CType_Double: {
+      VALUE str = rb_inspect(DBL2NUM(val.double_val));
+      StringBuilder_Printf(b, "%s", RSTRING_PTR(str));
+      break;
+    }
+    case kUpb_CType_Int32:
+      StringBuilder_Printf(b, "%" PRId32, val.int32_val);
+      break;
+    case kUpb_CType_UInt32:
+      StringBuilder_Printf(b, "%" PRIu32, val.uint32_val);
+      break;
+    case kUpb_CType_Int64:
+      StringBuilder_Printf(b, "%" PRId64, val.int64_val);
+      break;
+    case kUpb_CType_UInt64:
+      StringBuilder_Printf(b, "%" PRIu64, val.uint64_val);
+      break;
+    case kUpb_CType_String:
+      StringBuilder_Printf(b, "\"%.*s\"", (int)val.str_val.size,
+                           val.str_val.data);
+      break;
+    case kUpb_CType_Bytes:
+      StringBuilder_Printf(b, "\"%.*s\"", (int)val.str_val.size,
+                           val.str_val.data);
+      break;
+    case kUpb_CType_Enum:
+      StringBuilder_PrintEnum(b, val.int32_val, info.def.enumdef);
+      break;
+    case kUpb_CType_Message:
+      Message_PrintMessage(b, val.msg_val, info.def.msgdef);
+      break;
+  }
+}
+
+// -----------------------------------------------------------------------------
+// Arena
+// -----------------------------------------------------------------------------
+
+typedef struct {
+  upb_Arena *arena;
+  // IMPORTANT: WB_PROTECTED objects must only use the RB_OBJ_WRITE()
+  // macro to update VALUE references, as to trigger write barriers.
+  VALUE pinned_objs;
+} Arena;
+
+static void Arena_mark(void *data) {
+  Arena *arena = data;
+  rb_gc_mark(arena->pinned_objs);
+}
+
+static void Arena_free(void *data) {
+  Arena *arena = data;
+  upb_Arena_Free(arena->arena);
+  xfree(arena);
+}
+
+static size_t Arena_memsize(const void *data) {
+  const Arena *arena = data;
+  size_t fused_count;
+  size_t memsize = upb_Arena_SpaceAllocated(arena->arena, &fused_count);
+  if (fused_count > 1) {
+    // If other arena were fused we attribute an equal
+    // share of memory usage to each one.
+    memsize /= fused_count;
+  }
+  return memsize + sizeof(Arena);
+}
+
+static VALUE cArena;
+
+const rb_data_type_t Arena_type = {
+    "Google::Protobuf::Internal::Arena",
+    {Arena_mark, Arena_free, Arena_memsize},
+    .flags = RUBY_TYPED_FREE_IMMEDIATELY | RUBY_TYPED_WB_PROTECTED,
+};
+
+static void *ruby_upb_allocfunc(upb_alloc *alloc, void *ptr, size_t oldsize,
+                                size_t size) {
+  if (size == 0) {
+    xfree(ptr);
+    return NULL;
+  } else {
+    return xrealloc(ptr, size);
+  }
+}
+
+upb_alloc ruby_upb_alloc = {&ruby_upb_allocfunc};
+
+static VALUE Arena_alloc(VALUE klass) {
+  Arena *arena = ALLOC(Arena);
+  arena->arena = upb_Arena_Init(NULL, 0, &ruby_upb_alloc);
+  arena->pinned_objs = Qnil;
+  return TypedData_Wrap_Struct(klass, &Arena_type, arena);
+}
+
+upb_Arena *Arena_get(VALUE _arena) {
+  Arena *arena;
+  TypedData_Get_Struct(_arena, Arena, &Arena_type, arena);
+  return arena->arena;
+}
+
+void Arena_fuse(VALUE _arena, upb_Arena *other) {
+  Arena *arena;
+  TypedData_Get_Struct(_arena, Arena, &Arena_type, arena);
+  if (!upb_Arena_Fuse(arena->arena, other)) {
+    rb_raise(rb_eRuntimeError,
+             "Unable to fuse arenas. This should never happen since Ruby does "
+             "not use initial blocks");
+  }
+}
+
+VALUE Arena_new() { return Arena_alloc(cArena); }
+
+void Arena_register(VALUE module) {
+  VALUE internal = rb_define_module_under(module, "Internal");
+  VALUE klass = rb_define_class_under(internal, "Arena", rb_cObject);
+  rb_define_alloc_func(klass, Arena_alloc);
+  rb_gc_register_address(&cArena);
+  cArena = klass;
+}
+
+// -----------------------------------------------------------------------------
+// Object Cache
+// -----------------------------------------------------------------------------
+
+// Public ObjectCache API.
+
+VALUE weak_obj_cache = Qnil;
+ID item_get;
+ID item_try_add;
+
+static void ObjectCache_Init(VALUE protobuf) {
+  item_get = rb_intern("get");
+  item_try_add = rb_intern("try_add");
+
+  rb_gc_register_address(&weak_obj_cache);
+  VALUE internal = rb_const_get(protobuf, rb_intern("Internal"));
+#if SIZEOF_LONG >= SIZEOF_VALUE
+  VALUE cache_class = rb_const_get(internal, rb_intern("ObjectCache"));
+#else
+  VALUE cache_class = rb_const_get(internal, rb_intern("LegacyObjectCache"));
+#endif
+
+  weak_obj_cache = rb_class_new_instance(0, NULL, cache_class);
+  rb_const_set(internal, rb_intern("OBJECT_CACHE"), weak_obj_cache);
+  rb_const_set(internal, rb_intern("SIZEOF_LONG"), INT2NUM(SIZEOF_LONG));
+  rb_const_set(internal, rb_intern("SIZEOF_VALUE"), INT2NUM(SIZEOF_VALUE));
+}
+
+static VALUE ObjectCache_GetKey(const void *key) {
+  VALUE key_val = (VALUE)key;
+  PBRUBY_ASSERT((key_val & 3) == 0);
+  // Ensure the key can be stored as a Fixnum since 1 bit is needed for
+  // FIXNUM_FLAG and 1 bit is needed for the sign bit.
+  VALUE new_key = LL2NUM(key_val >> 2);
+  PBRUBY_ASSERT(FIXNUM_P(new_key));
+  return new_key;
+}
+
+VALUE ObjectCache_TryAdd(const void *key, VALUE val) {
+  VALUE key_val = ObjectCache_GetKey(key);
+  return rb_funcall(weak_obj_cache, item_try_add, 2, key_val, val);
+}
+
+// Returns the cached object for this key, if any. Otherwise returns Qnil.
+VALUE ObjectCache_Get(const void *key) {
+  VALUE key_val = ObjectCache_GetKey(key);
+  return rb_funcall(weak_obj_cache, item_get, 1, key_val);
+}
+
+/*
+ * call-seq:
+ *     Google::Protobuf.discard_unknown(msg)
+ *
+ * Discard unknown fields in the given message object and recursively discard
+ * unknown fields in submessages.
+ */
+static VALUE Google_Protobuf_discard_unknown(VALUE self, VALUE msg_rb) {
+  const upb_MessageDef *m;
+  upb_Message *msg = Message_GetMutable(msg_rb, &m);
+  if (!upb_Message_DiscardUnknown(msg, m, 128)) {
+    rb_raise(rb_eRuntimeError, "Messages nested too deeply.");
+  }
+
+  return Qnil;
+}
+
+/*
+ * call-seq:
+ *     Google::Protobuf.deep_copy(obj) => copy_of_obj
+ *
+ * Performs a deep copy of a RepeatedField instance, a Map instance, or a
+ * message object, recursively copying its members.
+ */
+VALUE Google_Protobuf_deep_copy(VALUE self, VALUE obj) {
+  VALUE klass = CLASS_OF(obj);
+  if (klass == cRepeatedField) {
+    return RepeatedField_deep_copy(obj);
+  } else if (klass == cMap) {
+    return Map_deep_copy(obj);
+  } else {
+    VALUE new_arena_rb = Arena_new();
+    upb_Arena *new_arena = Arena_get(new_arena_rb);
+    const upb_MessageDef *m;
+    const upb_Message *msg = Message_Get(obj, &m);
+    upb_Message *new_msg = Message_deep_copy(msg, m, new_arena);
+    return Message_GetRubyWrapper(new_msg, m, new_arena_rb);
+  }
+}
+
+// -----------------------------------------------------------------------------
+// Initialization/entry point.
+// -----------------------------------------------------------------------------
+
+// This must be named "Init_protobuf_c" because the Ruby module is named
+// "protobuf_c" -- the VM looks for this symbol in our .so.
+__attribute__((visibility("default"))) void Init_protobuf_c() {
+  VALUE google = rb_define_module("Google");
+  VALUE protobuf = rb_define_module_under(google, "Protobuf");
+
+  ObjectCache_Init(protobuf);
+  Arena_register(protobuf);
+  Defs_register(protobuf);
+  RepeatedField_register(protobuf);
+  Map_register(protobuf);
+  Message_register(protobuf);
+
+  cParseError = rb_const_get(protobuf, rb_intern("ParseError"));
+  rb_gc_register_mark_object(cParseError);
+  cTypeError = rb_const_get(protobuf, rb_intern("TypeError"));
+  rb_gc_register_mark_object(cTypeError);
+
+  rb_define_singleton_method(protobuf, "discard_unknown",
+                             Google_Protobuf_discard_unknown, 1);
+  rb_define_singleton_method(protobuf, "deep_copy", Google_Protobuf_deep_copy,
+                             1);
+}
+
+// -----------------------------------------------------------------------------
+// Utilities
+// -----------------------------------------------------------------------------
+
+// Raises a Ruby error if val is frozen in Ruby or UPB.
+void Protobuf_CheckNotFrozen(VALUE val, bool upb_frozen) {
+  if (RB_UNLIKELY(rb_obj_frozen_p(val)||upb_frozen)) {
+    rb_error_frozen_object(val);
+  }
+}

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/protobuf.h

@@ -0,0 +1,108 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2014 Google Inc.  All rights reserved.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file or at
+// https://developers.google.com/open-source/licenses/bsd
+
+#ifndef __GOOGLE_PROTOBUF_RUBY_PROTOBUF_H__
+#define __GOOGLE_PROTOBUF_RUBY_PROTOBUF_H__
+
+// Ruby 3+ defines NDEBUG itself, see: https://bugs.ruby-lang.org/issues/18777
+#ifdef NDEBUG
+#include <ruby.h>
+#else
+#include <ruby.h>
+#undef NDEBUG
+#endif
+
+#include <ruby/version.h>
+
+#if RUBY_API_VERSION_CODE < 20700
+#error Protobuf requires Ruby >= 2.7
+#endif
+
+#include <assert.h>  // Must be included after the NDEBUG logic above.
+#include <ruby/encoding.h>
+#include <ruby/vm.h>
+
+#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).
+const upb_FieldDef* map_field_key(const upb_FieldDef* field);
+const upb_FieldDef* map_field_value(const upb_FieldDef* field);
+
+// -----------------------------------------------------------------------------
+// Arena
+// -----------------------------------------------------------------------------
+
+// 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 Arena_new();
+upb_Arena* Arena_get(VALUE arena);
+
+// Fuses this arena to another, throwing a Ruby exception if this is not
+// possible.
+void Arena_fuse(VALUE arena, upb_Arena* other);
+
+// -----------------------------------------------------------------------------
+// ObjectCache
+// -----------------------------------------------------------------------------
+
+// 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).
+
+// Tries to add a new entry to the cache, returning the newly installed value or
+// the pre-existing entry.
+VALUE ObjectCache_TryAdd(const void* key, VALUE val);
+
+// Returns the cached object for this key, if any. Otherwise returns Qnil.
+VALUE ObjectCache_Get(const void* key);
+
+// -----------------------------------------------------------------------------
+// StringBuilder, for inspect
+// -----------------------------------------------------------------------------
+
+struct StringBuilder;
+typedef struct StringBuilder StringBuilder;
+
+StringBuilder* StringBuilder_New();
+void StringBuilder_Free(StringBuilder* b);
+void StringBuilder_Printf(StringBuilder* b, const char* fmt, ...);
+VALUE StringBuilder_ToRubyString(StringBuilder* b);
+
+void StringBuilder_PrintMsgval(StringBuilder* b, upb_MessageValue val,
+                               TypeInfo info);
+
+// -----------------------------------------------------------------------------
+// Utilities.
+// -----------------------------------------------------------------------------
+
+extern VALUE cTypeError;
+
+#ifdef NDEBUG
+#define PBRUBY_ASSERT(expr) \
+  do {                      \
+  } while (false && (expr))
+#else
+#define PBRUBY_ASSERT(expr) \
+  if (!(expr))              \
+  rb_bug("Assertion failed at %s:%d, expr: %s", __FILE__, __LINE__, #expr)
+#endif
+
+// Raises a Ruby error if val is frozen in Ruby or upb_frozen is true.
+void Protobuf_CheckNotFrozen(VALUE val, bool upb_frozen);
+
+#define PBRUBY_MAX(x, y) (((x) > (y)) ? (x) : (y))
+
+#define UPB_UNUSED(var) (void)var
+
+#endif  // __GOOGLE_PROTOBUF_RUBY_PROTOBUF_H__

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/repeated_field.c

@@ -0,0 +1,691 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2014 Google Inc.  All rights reserved.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file or at
+// https://developers.google.com/open-source/licenses/bsd
+
+#include "repeated_field.h"
+
+#include "convert.h"
+#include "defs.h"
+#include "message.h"
+#include "protobuf.h"
+
+// -----------------------------------------------------------------------------
+// Repeated field container type.
+// -----------------------------------------------------------------------------
+
+typedef struct {
+  const upb_Array* array;  // Can get as mutable when non-frozen.
+  TypeInfo type_info;
+  VALUE type_class;  // To GC-root the msgdef/enumdef in type_info.
+  VALUE arena;       // To GC-root the upb_Array.
+} RepeatedField;
+
+VALUE cRepeatedField;
+
+static void RepeatedField_mark(void* _self) {
+  RepeatedField* self = (RepeatedField*)_self;
+  rb_gc_mark(self->type_class);
+  rb_gc_mark(self->arena);
+}
+
+const rb_data_type_t RepeatedField_type = {
+    "Google::Protobuf::RepeatedField",
+    {RepeatedField_mark, RUBY_DEFAULT_FREE, NULL},
+    .flags = RUBY_TYPED_FREE_IMMEDIATELY,
+};
+
+static RepeatedField* ruby_to_RepeatedField(VALUE _self) {
+  RepeatedField* self;
+  TypedData_Get_Struct(_self, RepeatedField, &RepeatedField_type, self);
+  return self;
+}
+
+static upb_Array* RepeatedField_GetMutable(VALUE _self) {
+  const upb_Array* array = ruby_to_RepeatedField(_self)->array;
+  Protobuf_CheckNotFrozen(_self, upb_Array_IsFrozen(array));
+  return (upb_Array*)array;
+}
+
+VALUE RepeatedField_alloc(VALUE klass) {
+  RepeatedField* self = ALLOC(RepeatedField);
+  self->arena = Qnil;
+  self->type_class = Qnil;
+  self->array = NULL;
+  return TypedData_Wrap_Struct(klass, &RepeatedField_type, self);
+}
+
+VALUE RepeatedField_EmptyFrozen(const upb_FieldDef* f) {
+  PBRUBY_ASSERT(upb_FieldDef_IsRepeated(f));
+  VALUE val = ObjectCache_Get(f);
+
+  if (val == Qnil) {
+    val = RepeatedField_alloc(cRepeatedField);
+    RepeatedField* self;
+    TypedData_Get_Struct(val, RepeatedField, &RepeatedField_type, self);
+    self->arena = Arena_new();
+    TypeInfo type_info = TypeInfo_get(f);
+    self->array = upb_Array_New(Arena_get(self->arena), type_info.type);
+    self->type_info = type_info;
+    if (self->type_info.type == kUpb_CType_Message) {
+      self->type_class = Descriptor_DefToClass(type_info.def.msgdef);
+    }
+    val = ObjectCache_TryAdd(f, RepeatedField_freeze(val));
+  }
+  PBRUBY_ASSERT(RB_OBJ_FROZEN(val));
+  PBRUBY_ASSERT(upb_Array_IsFrozen(ruby_to_RepeatedField(val)->array));
+  return val;
+}
+
+VALUE RepeatedField_GetRubyWrapper(const upb_Array* array, TypeInfo type_info,
+                                   VALUE arena) {
+  PBRUBY_ASSERT(array);
+  PBRUBY_ASSERT(arena != Qnil);
+  VALUE val = ObjectCache_Get(array);
+
+  if (val == Qnil) {
+    val = RepeatedField_alloc(cRepeatedField);
+    RepeatedField* self;
+    TypedData_Get_Struct(val, RepeatedField, &RepeatedField_type, self);
+    self->array = array;
+    self->arena = arena;
+    self->type_info = type_info;
+    if (self->type_info.type == kUpb_CType_Message) {
+      self->type_class = Descriptor_DefToClass(type_info.def.msgdef);
+    }
+    val = ObjectCache_TryAdd(array, val);
+  }
+
+  PBRUBY_ASSERT(ruby_to_RepeatedField(val)->type_info.type == type_info.type);
+  PBRUBY_ASSERT(ruby_to_RepeatedField(val)->type_info.def.msgdef ==
+                type_info.def.msgdef);
+  PBRUBY_ASSERT(ruby_to_RepeatedField(val)->array == array);
+  return val;
+}
+
+static VALUE RepeatedField_new_this_type(RepeatedField* from) {
+  VALUE arena_rb = Arena_new();
+  upb_Array* array = upb_Array_New(Arena_get(arena_rb), from->type_info.type);
+  VALUE ret = RepeatedField_GetRubyWrapper(array, from->type_info, arena_rb);
+  PBRUBY_ASSERT(ruby_to_RepeatedField(ret)->type_class == from->type_class);
+  return ret;
+}
+
+void RepeatedField_Inspect(StringBuilder* b, const upb_Array* array,
+                           TypeInfo info) {
+  bool first = true;
+  StringBuilder_Printf(b, "[");
+  size_t n = array ? upb_Array_Size(array) : 0;
+  for (size_t i = 0; i < n; i++) {
+    if (first) {
+      first = false;
+    } else {
+      StringBuilder_Printf(b, ", ");
+    }
+    StringBuilder_PrintMsgval(b, upb_Array_Get(array, i), info);
+  }
+  StringBuilder_Printf(b, "]");
+}
+
+VALUE RepeatedField_deep_copy(VALUE _self) {
+  RepeatedField* self = ruby_to_RepeatedField(_self);
+  VALUE new_rptfield = RepeatedField_new_this_type(self);
+  RepeatedField* new_self = ruby_to_RepeatedField(new_rptfield);
+  VALUE arena_rb = new_self->arena;
+  upb_Array* new_array = RepeatedField_GetMutable(new_rptfield);
+  upb_Arena* arena = Arena_get(arena_rb);
+  size_t elements = upb_Array_Size(self->array);
+
+  upb_Array_Resize(new_array, elements, arena);
+
+  size_t size = upb_Array_Size(self->array);
+  for (size_t i = 0; i < size; i++) {
+    upb_MessageValue msgval = upb_Array_Get(self->array, i);
+    upb_MessageValue copy = Msgval_DeepCopy(msgval, self->type_info, arena);
+    upb_Array_Set(new_array, i, copy);
+  }
+
+  return new_rptfield;
+}
+
+const upb_Array* RepeatedField_GetUpbArray(VALUE val, const upb_FieldDef* field,
+                                           upb_Arena* arena) {
+  RepeatedField* self;
+  TypeInfo type_info = TypeInfo_get(field);
+
+  if (!RB_TYPE_P(val, T_DATA) || !RTYPEDDATA_P(val) ||
+      RTYPEDDATA_TYPE(val) != &RepeatedField_type) {
+    rb_raise(cTypeError, "Expected repeated field array");
+  }
+
+  self = ruby_to_RepeatedField(val);
+  if (self->type_info.type != type_info.type) {
+    rb_raise(cTypeError, "Repeated field array has wrong element type");
+  }
+
+  if (self->type_info.def.msgdef != type_info.def.msgdef) {
+    rb_raise(cTypeError, "Repeated field array has wrong message/enum class");
+  }
+
+  Arena_fuse(self->arena, arena);
+  return self->array;
+}
+
+static int index_position(VALUE _index, RepeatedField* repeated_field) {
+  int index = NUM2INT(_index);
+  if (index < 0) index += upb_Array_Size(repeated_field->array);
+  return index;
+}
+
+static VALUE RepeatedField_subarray(RepeatedField* self, long beg, long len) {
+  size_t size = upb_Array_Size(self->array);
+  VALUE ary = rb_ary_new2(size);
+  long i;
+
+  for (i = beg; i < beg + len; i++) {
+    upb_MessageValue msgval = upb_Array_Get(self->array, i);
+    VALUE elem = Convert_UpbToRuby(msgval, self->type_info, self->arena);
+    rb_ary_push(ary, elem);
+  }
+  return ary;
+}
+
+/*
+ * call-seq:
+ *     RepeatedField.each(&block)
+ *
+ * Invokes the block once for each element of the repeated field. RepeatedField
+ * also includes Enumerable; combined with this method, the repeated field thus
+ * acts like an ordinary Ruby sequence.
+ */
+static VALUE RepeatedField_each(VALUE _self) {
+  RepeatedField* self = ruby_to_RepeatedField(_self);
+  int size = upb_Array_Size(self->array);
+  int i;
+
+  for (i = 0; i < size; i++) {
+    upb_MessageValue msgval = upb_Array_Get(self->array, i);
+    VALUE val = Convert_UpbToRuby(msgval, self->type_info, self->arena);
+    rb_yield(val);
+  }
+  return _self;
+}
+
+/*
+ * call-seq:
+ *     RepeatedField.[](index) => value
+ *
+ * Accesses the element at the given index. Returns nil on out-of-bounds
+ */
+static VALUE RepeatedField_index(int argc, VALUE* argv, VALUE _self) {
+  RepeatedField* self = ruby_to_RepeatedField(_self);
+  long size = upb_Array_Size(self->array);
+
+  VALUE arg = argv[0];
+  long beg, len;
+
+  if (argc == 1) {
+    if (FIXNUM_P(arg)) {
+      /* standard case */
+      upb_MessageValue msgval;
+      int index = index_position(argv[0], self);
+      if (index < 0 || (size_t)index >= upb_Array_Size(self->array)) {
+        return Qnil;
+      }
+      msgval = upb_Array_Get(self->array, index);
+      return Convert_UpbToRuby(msgval, self->type_info, self->arena);
+    } else {
+      /* check if idx is Range */
+      switch (rb_range_beg_len(arg, &beg, &len, size, 0)) {
+        case Qfalse:
+          break;
+        case Qnil:
+          return Qnil;
+        default:
+          return RepeatedField_subarray(self, beg, len);
+      }
+    }
+  }
+
+  /* assume 2 arguments */
+  beg = NUM2LONG(argv[0]);
+  len = NUM2LONG(argv[1]);
+  if (beg < 0) {
+    beg += size;
+  }
+  if (beg >= size) {
+    return Qnil;
+  }
+  return RepeatedField_subarray(self, beg, len);
+}
+
+/*
+ * call-seq:
+ *     RepeatedField.[]=(index, value)
+ *
+ * Sets the element at the given index. On out-of-bounds assignments, extends
+ * the array and fills the hole (if any) with default values.
+ */
+static VALUE RepeatedField_index_set(VALUE _self, VALUE _index, VALUE val) {
+  RepeatedField* self = ruby_to_RepeatedField(_self);
+  int size = upb_Array_Size(self->array);
+  upb_Array* array = RepeatedField_GetMutable(_self);
+  upb_Arena* arena = Arena_get(self->arena);
+  upb_MessageValue msgval = Convert_RubyToUpb(val, "", self->type_info, arena);
+
+  int index = index_position(_index, self);
+  if (index < 0 || index >= (INT_MAX - 1)) {
+    return Qnil;
+  }
+
+  if (index >= size) {
+    upb_Array_Resize(array, index + 1, arena);
+    upb_MessageValue fill;
+    memset(&fill, 0, sizeof(fill));
+    for (int i = size; i < index; i++) {
+      // Fill default values.
+      // TODO: should this happen at the upb level?
+      upb_Array_Set(array, i, fill);
+    }
+  }
+
+  upb_Array_Set(array, index, msgval);
+  return Qnil;
+}
+
+/*
+ * call-seq:
+ *     RepeatedField.push(value, ...)
+ *
+ * Adds a new element to the repeated field.
+ */
+static VALUE RepeatedField_push_vararg(int argc, VALUE* argv, VALUE _self) {
+  RepeatedField* self = ruby_to_RepeatedField(_self);
+  upb_Arena* arena = Arena_get(self->arena);
+  upb_Array* array = RepeatedField_GetMutable(_self);
+  int i;
+
+  for (i = 0; i < argc; i++) {
+    upb_MessageValue msgval =
+        Convert_RubyToUpb(argv[i], "", self->type_info, arena);
+    upb_Array_Append(array, msgval, arena);
+  }
+
+  return _self;
+}
+
+/*
+ * call-seq:
+ *     RepeatedField.<<(value)
+ *
+ * Adds a new element to the repeated field.
+ */
+static VALUE RepeatedField_push(VALUE _self, VALUE val) {
+  RepeatedField* self = ruby_to_RepeatedField(_self);
+  upb_Arena* arena = Arena_get(self->arena);
+  upb_Array* array = RepeatedField_GetMutable(_self);
+
+  upb_MessageValue msgval = Convert_RubyToUpb(val, "", self->type_info, arena);
+  upb_Array_Append(array, msgval, arena);
+
+  return _self;
+}
+
+/*
+ * Private ruby method, used by RepeatedField.pop
+ */
+static VALUE RepeatedField_pop_one(VALUE _self) {
+  RepeatedField* self = ruby_to_RepeatedField(_self);
+  size_t size = upb_Array_Size(self->array);
+  upb_Array* array = RepeatedField_GetMutable(_self);
+  upb_MessageValue last;
+  VALUE ret;
+
+  if (size == 0) {
+    return Qnil;
+  }
+
+  last = upb_Array_Get(self->array, size - 1);
+  ret = Convert_UpbToRuby(last, self->type_info, self->arena);
+
+  upb_Array_Resize(array, size - 1, Arena_get(self->arena));
+  return ret;
+}
+
+/*
+ * call-seq:
+ *     RepeatedField.replace(list)
+ *
+ * Replaces the contents of the repeated field with the given list of elements.
+ */
+static VALUE RepeatedField_replace(VALUE _self, VALUE list) {
+  RepeatedField* self = ruby_to_RepeatedField(_self);
+  upb_Array* array = RepeatedField_GetMutable(_self);
+  int i;
+
+  Check_Type(list, T_ARRAY);
+  upb_Array_Resize(array, 0, Arena_get(self->arena));
+
+  for (i = 0; i < RARRAY_LEN(list); i++) {
+    RepeatedField_push(_self, rb_ary_entry(list, i));
+  }
+
+  return list;
+}
+
+/*
+ * call-seq:
+ *     RepeatedField.clear
+ *
+ * Clears (removes all elements from) this repeated field.
+ */
+static VALUE RepeatedField_clear(VALUE _self) {
+  RepeatedField* self = ruby_to_RepeatedField(_self);
+  upb_Array* array = RepeatedField_GetMutable(_self);
+  upb_Array_Resize(array, 0, Arena_get(self->arena));
+  return _self;
+}
+
+/*
+ * call-seq:
+ *     RepeatedField.length
+ *
+ * Returns the length of this repeated field.
+ */
+static VALUE RepeatedField_length(VALUE _self) {
+  RepeatedField* self = ruby_to_RepeatedField(_self);
+  return INT2NUM(upb_Array_Size(self->array));
+}
+
+/*
+ * call-seq:
+ *     RepeatedField.dup => repeated_field
+ *
+ * Duplicates this repeated field with a shallow copy. References to all
+ * non-primitive element objects (e.g., submessages) are shared.
+ */
+static VALUE RepeatedField_dup(VALUE _self) {
+  RepeatedField* self = ruby_to_RepeatedField(_self);
+  VALUE new_rptfield = RepeatedField_new_this_type(self);
+  RepeatedField* new_rptfield_self = ruby_to_RepeatedField(new_rptfield);
+  upb_Array* new_array = RepeatedField_GetMutable(new_rptfield);
+  upb_Arena* arena = Arena_get(new_rptfield_self->arena);
+  int size = upb_Array_Size(self->array);
+  int i;
+
+  Arena_fuse(self->arena, arena);
+
+  for (i = 0; i < size; i++) {
+    upb_MessageValue msgval = upb_Array_Get(self->array, i);
+    upb_Array_Append(new_array, msgval, arena);
+  }
+
+  return new_rptfield;
+}
+
+/*
+ * call-seq:
+ *     RepeatedField.to_ary => array
+ *
+ * Used when converted implicitly into array, e.g. compared to an Array.
+ * Also called as a fallback of Object#to_a
+ */
+VALUE RepeatedField_to_ary(VALUE _self) {
+  RepeatedField* self = ruby_to_RepeatedField(_self);
+  int size = upb_Array_Size(self->array);
+  VALUE ary = rb_ary_new2(size);
+  int i;
+
+  for (i = 0; i < size; i++) {
+    upb_MessageValue msgval = upb_Array_Get(self->array, i);
+    VALUE val = Convert_UpbToRuby(msgval, self->type_info, self->arena);
+    rb_ary_push(ary, val);
+  }
+
+  return ary;
+}
+
+/*
+ * call-seq:
+ *     RepeatedField.==(other) => boolean
+ *
+ * Compares this repeated field to another. Repeated fields are equal if their
+ * element types are equal, their lengths are equal, and each element is equal.
+ * Elements are compared as per normal Ruby semantics, by calling their :==
+ * methods (or performing a more efficient comparison for primitive types).
+ *
+ * Repeated fields with dissimilar element types are never equal, even if value
+ * comparison (for example, between integers and floats) would have otherwise
+ * indicated that every element has equal value.
+ */
+VALUE RepeatedField_eq(VALUE _self, VALUE _other) {
+  RepeatedField* self;
+  RepeatedField* other;
+
+  if (_self == _other) {
+    return Qtrue;
+  }
+
+  if (TYPE(_other) == T_ARRAY) {
+    VALUE self_ary = RepeatedField_to_ary(_self);
+    return rb_equal(self_ary, _other);
+  }
+
+  self = ruby_to_RepeatedField(_self);
+  other = ruby_to_RepeatedField(_other);
+  size_t n = upb_Array_Size(self->array);
+
+  if (self->type_info.type != other->type_info.type ||
+      self->type_class != other->type_class ||
+      upb_Array_Size(other->array) != n) {
+    return Qfalse;
+  }
+
+  for (size_t i = 0; i < n; i++) {
+    upb_MessageValue val1 = upb_Array_Get(self->array, i);
+    upb_MessageValue val2 = upb_Array_Get(other->array, i);
+    if (!Msgval_IsEqual(val1, val2, self->type_info)) {
+      return Qfalse;
+    }
+  }
+
+  return Qtrue;
+}
+
+/*
+ * call-seq:
+ *     RepeatedField.frozen? => bool
+ *
+ * Returns true if the repeated field is frozen in either Ruby or the underlying
+ * representation. Freezes the Ruby repeated field object if it is not already
+ * frozen in Ruby but it is frozen in the underlying representation.
+ */
+VALUE RepeatedField_frozen(VALUE _self) {
+  RepeatedField* self = ruby_to_RepeatedField(_self);
+  if (!upb_Array_IsFrozen(self->array)) {
+    PBRUBY_ASSERT(!RB_OBJ_FROZEN(_self));
+    return Qfalse;
+  }
+
+  // Lazily freeze the Ruby wrapper.
+  if (!RB_OBJ_FROZEN(_self)) RB_OBJ_FREEZE(_self);
+  return Qtrue;
+}
+
+/*
+ * call-seq:
+ *     RepeatedField.freeze => self
+ *
+ * Freezes the repeated field object. We have to intercept this so we can freeze
+ * the underlying representation, not just the Ruby wrapper.
+ */
+VALUE RepeatedField_freeze(VALUE _self) {
+  RepeatedField* self = ruby_to_RepeatedField(_self);
+  if (RB_OBJ_FROZEN(_self)) {
+    PBRUBY_ASSERT(upb_Array_IsFrozen(self->array));
+    return _self;
+  }
+
+  if (!upb_Array_IsFrozen(self->array)) {
+    if (self->type_info.type == kUpb_CType_Message) {
+      upb_Array_Freeze(RepeatedField_GetMutable(_self),
+                       upb_MessageDef_MiniTable(self->type_info.def.msgdef));
+    } else {
+      upb_Array_Freeze(RepeatedField_GetMutable(_self), NULL);
+    }
+  }
+  RB_OBJ_FREEZE(_self);
+  return _self;
+}
+
+/*
+ * call-seq:
+ *     RepeatedField.hash => hash_value
+ *
+ * Returns a hash value computed from this repeated field's elements.
+ */
+VALUE RepeatedField_hash(VALUE _self) {
+  RepeatedField* self = ruby_to_RepeatedField(_self);
+  uint64_t hash = 0;
+  size_t n = upb_Array_Size(self->array);
+
+  for (size_t i = 0; i < n; i++) {
+    upb_MessageValue val = upb_Array_Get(self->array, i);
+    hash = Msgval_GetHash(val, self->type_info, hash);
+  }
+
+  return LL2NUM(hash);
+}
+
+/*
+ * call-seq:
+ *     RepeatedField.+(other) => repeated field
+ *
+ * Returns a new repeated field that contains the concatenated list of this
+ * repeated field's elements and other's elements. The other (second) list may
+ * be either another repeated field or a Ruby array.
+ */
+VALUE RepeatedField_plus(VALUE _self, VALUE list) {
+  VALUE dupped_ = RepeatedField_dup(_self);
+
+  if (TYPE(list) == T_ARRAY) {
+    int i;
+    for (i = 0; i < RARRAY_LEN(list); i++) {
+      VALUE elem = rb_ary_entry(list, i);
+      RepeatedField_push(dupped_, elem);
+    }
+  } else if (RB_TYPE_P(list, T_DATA) && RTYPEDDATA_P(list) &&
+             RTYPEDDATA_TYPE(list) == &RepeatedField_type) {
+    RepeatedField* self = ruby_to_RepeatedField(_self);
+    RepeatedField* list_rptfield = ruby_to_RepeatedField(list);
+    RepeatedField* dupped = ruby_to_RepeatedField(dupped_);
+    upb_Array* dupped_array = RepeatedField_GetMutable(dupped_);
+    upb_Arena* arena = Arena_get(dupped->arena);
+    Arena_fuse(list_rptfield->arena, arena);
+    int size = upb_Array_Size(list_rptfield->array);
+    int i;
+
+    if (self->type_info.type != list_rptfield->type_info.type ||
+        self->type_class != list_rptfield->type_class) {
+      rb_raise(rb_eArgError,
+               "Attempt to append RepeatedField with different element type.");
+    }
+
+    for (i = 0; i < size; i++) {
+      upb_MessageValue msgval = upb_Array_Get(list_rptfield->array, i);
+      upb_Array_Append(dupped_array, msgval, arena);
+    }
+  } else {
+    rb_raise(rb_eArgError, "Unknown type appending to RepeatedField");
+  }
+
+  return dupped_;
+}
+
+/*
+ * call-seq:
+ *     RepeatedField.concat(other) => self
+ *
+ * concats the passed in array to self.  Returns a Ruby array.
+ */
+VALUE RepeatedField_concat(VALUE _self, VALUE list) {
+  int i;
+
+  Check_Type(list, T_ARRAY);
+  for (i = 0; i < RARRAY_LEN(list); i++) {
+    RepeatedField_push(_self, rb_ary_entry(list, i));
+  }
+  return _self;
+}
+
+/*
+ * call-seq:
+ *     RepeatedField.new(type, type_class = nil, initial_elems = [])
+ *
+ * Creates a new repeated field. The provided type must be a Ruby symbol, and
+ * can take on the same values as those accepted by FieldDescriptor#type=. If
+ * the type is :message or :enum, type_class must be non-nil, and must be the
+ * Ruby class or module returned by Descriptor#msgclass or
+ * EnumDescriptor#enummodule, respectively. An initial list of elements may also
+ * be provided.
+ */
+VALUE RepeatedField_init(int argc, VALUE* argv, VALUE _self) {
+  RepeatedField* self = ruby_to_RepeatedField(_self);
+  upb_Arena* arena;
+  VALUE ary = Qnil;
+
+  self->arena = Arena_new();
+  arena = Arena_get(self->arena);
+
+  if (argc < 1) {
+    rb_raise(rb_eArgError, "Expected at least 1 argument.");
+  }
+
+  self->type_info = TypeInfo_FromClass(argc, argv, 0, &self->type_class, &ary);
+  self->array = upb_Array_New(arena, self->type_info.type);
+  VALUE stored_val = ObjectCache_TryAdd(self->array, _self);
+  PBRUBY_ASSERT(stored_val == _self);
+
+  if (ary != Qnil) {
+    if (!RB_TYPE_P(ary, T_ARRAY)) {
+      rb_raise(rb_eArgError, "Expected array as initialize argument");
+    }
+    for (int i = 0; i < RARRAY_LEN(ary); i++) {
+      RepeatedField_push(_self, rb_ary_entry(ary, i));
+    }
+  }
+  return Qnil;
+}
+
+void RepeatedField_register(VALUE module) {
+  VALUE klass = rb_define_class_under(module, "RepeatedField", rb_cObject);
+  rb_define_alloc_func(klass, RepeatedField_alloc);
+  rb_gc_register_address(&cRepeatedField);
+  cRepeatedField = klass;
+
+  rb_define_method(klass, "initialize", RepeatedField_init, -1);
+  rb_define_method(klass, "each", RepeatedField_each, 0);
+  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_vararg, -1);
+  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);
+  rb_define_method(klass, "clear", RepeatedField_clear, 0);
+  rb_define_method(klass, "length", RepeatedField_length, 0);
+  rb_define_method(klass, "size", RepeatedField_length, 0);
+  rb_define_method(klass, "dup", RepeatedField_dup, 0);
+  // Also define #clone so that we don't inherit Object#clone.
+  rb_define_method(klass, "clone", RepeatedField_dup, 0);
+  rb_define_method(klass, "==", RepeatedField_eq, 1);
+  rb_define_method(klass, "to_ary", RepeatedField_to_ary, 0);
+  rb_define_method(klass, "freeze", RepeatedField_freeze, 0);
+  rb_define_method(klass, "frozen?", RepeatedField_frozen, 0);
+  rb_define_method(klass, "hash", RepeatedField_hash, 0);
+  rb_define_method(klass, "+", RepeatedField_plus, 1);
+  rb_define_method(klass, "concat", RepeatedField_concat, 1);
+  rb_include_module(klass, rb_mEnumerable);
+}

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/repeated_field.h

@@ -0,0 +1,45 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.  All rights reserved.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file or at
+// https://developers.google.com/open-source/licenses/bsd
+
+#ifndef RUBY_PROTOBUF_REPEATED_FIELD_H_
+#define RUBY_PROTOBUF_REPEATED_FIELD_H_
+
+#include "protobuf.h"
+#include "ruby-upb.h"
+
+// Returns a frozen sentinel Ruby wrapper object for an empty upb_Array of the
+// type specified by the field. Creates one if it doesn't exist.
+VALUE RepeatedField_EmptyFrozen(const upb_FieldDef* f);
+
+// Returns a Ruby wrapper object for the given upb_Array, which will be created
+// if one does not exist already.
+VALUE RepeatedField_GetRubyWrapper(const upb_Array* msg, TypeInfo type_info,
+                                   VALUE arena);
+
+// Gets the underlying upb_Array for this Ruby RepeatedField object, which must
+// have a type that matches |f|. If this is not a repeated field or the type
+// doesn't match, raises an exception.
+const upb_Array* RepeatedField_GetUpbArray(VALUE value, const upb_FieldDef* f,
+                                           upb_Arena* arena);
+
+// Implements #inspect for this repeated field by appending its contents to |b|.
+void RepeatedField_Inspect(StringBuilder* b, const upb_Array* array,
+                           TypeInfo info);
+
+// Returns a deep copy of this RepeatedField object.
+VALUE RepeatedField_deep_copy(VALUE obj);
+
+// Ruby class of Google::Protobuf::RepeatedField.
+extern VALUE cRepeatedField;
+
+// Call at startup to register all types in this module.
+void RepeatedField_register(VALUE module);
+
+// Recursively freeze RepeatedField.
+VALUE RepeatedField_freeze(VALUE _self);
+
+#endif  // RUBY_PROTOBUF_REPEATED_FIELD_H_

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/shared_convert.c

@@ -0,0 +1,69 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2023 Google Inc.  All rights reserved.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file or at
+// https://developers.google.com/open-source/licenses/bsd
+
+// -----------------------------------------------------------------------------
+// Ruby <-> upb data conversion functions. Strictly free of dependencies on
+// Ruby interpreter internals.
+
+#include "shared_convert.h"
+
+bool shared_Msgval_IsEqual(upb_MessageValue val1, upb_MessageValue val2,
+                           upb_CType type, const upb_MessageDef* msgdef,
+                           upb_Status* status) {
+  switch (type) {
+    case kUpb_CType_Bool:
+      return memcmp(&val1, &val2, 1) == 0;
+    case kUpb_CType_Float:
+    case kUpb_CType_Int32:
+    case kUpb_CType_UInt32:
+    case kUpb_CType_Enum:
+      return memcmp(&val1, &val2, 4) == 0;
+    case kUpb_CType_Double:
+    case kUpb_CType_Int64:
+    case kUpb_CType_UInt64:
+      return memcmp(&val1, &val2, 8) == 0;
+    case kUpb_CType_String:
+    case kUpb_CType_Bytes:
+      return val1.str_val.size == val2.str_val.size &&
+             memcmp(val1.str_val.data, val2.str_val.data, val1.str_val.size) ==
+                 0;
+    case kUpb_CType_Message: {
+      const upb_MiniTable* m = upb_MessageDef_MiniTable(msgdef);
+      const int options = 0;
+      return upb_Message_IsEqual(val1.msg_val, val2.msg_val, m, options);
+    }
+    default:
+      upb_Status_SetErrorMessage(status, "Internal error, unexpected type");
+      return false;
+  }
+}
+
+uint64_t shared_Msgval_GetHash(upb_MessageValue val, upb_CType type,
+                               const upb_MessageDef* msgdef, uint64_t seed,
+                               upb_Status* status) {
+  switch (type) {
+    case kUpb_CType_Bool:
+      return _upb_Hash(&val, 1, seed);
+    case kUpb_CType_Float:
+    case kUpb_CType_Int32:
+    case kUpb_CType_UInt32:
+    case kUpb_CType_Enum:
+      return _upb_Hash(&val, 4, seed);
+    case kUpb_CType_Double:
+    case kUpb_CType_Int64:
+    case kUpb_CType_UInt64:
+      return _upb_Hash(&val, 8, seed);
+    case kUpb_CType_String:
+    case kUpb_CType_Bytes:
+      return _upb_Hash(val.str_val.data, val.str_val.size, seed);
+    case kUpb_CType_Message:
+      return shared_Message_Hash(val.msg_val, msgdef, seed, status);
+    default:
+      upb_Status_SetErrorMessage(status, "Internal error, unexpected type");
+      return 0;
+  }
+}

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/shared_convert.h

@@ -0,0 +1,26 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2023 Google Inc.  All rights reserved.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file or at
+// https://developers.google.com/open-source/licenses/bsd
+
+// -----------------------------------------------------------------------------
+// Ruby <-> upb data conversion functions. Strictly free of dependencies on
+// Ruby interpreter internals.
+
+#ifndef RUBY_PROTOBUF_SHARED_CONVERT_H_
+#define RUBY_PROTOBUF_SHARED_CONVERT_H_
+
+#include "ruby-upb.h"
+#include "shared_message.h"
+
+bool shared_Msgval_IsEqual(upb_MessageValue val1, upb_MessageValue val2,
+                           upb_CType type, const upb_MessageDef* msgdef,
+                           upb_Status* status);
+
+uint64_t shared_Msgval_GetHash(upb_MessageValue val, upb_CType type,
+                               const upb_MessageDef* msgdef, uint64_t seed,
+                               upb_Status* status);
+
+#endif  // RUBY_PROTOBUF_SHARED_CONVERT_H_

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/shared_message.c

@@ -0,0 +1,37 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2023 Google Inc.  All rights reserved.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file or at
+// https://developers.google.com/open-source/licenses/bsd
+
+// -----------------------------------------------------------------------------
+// Ruby Message functions. Strictly free of dependencies on
+// Ruby interpreter internals.
+
+#include "shared_message.h"
+
+// Support function for Message_Hash. Returns a hash value for the given
+// message.
+uint64_t shared_Message_Hash(const upb_Message* msg, const upb_MessageDef* m,
+                             uint64_t seed, upb_Status* status) {
+  upb_Arena* arena = upb_Arena_New();
+  char* data;
+  size_t size;
+
+  // Hash a deterministically serialized payloads with no unknown fields.
+  upb_EncodeStatus encode_status = upb_Encode(
+      msg, upb_MessageDef_MiniTable(m),
+      kUpb_EncodeOption_SkipUnknown | kUpb_EncodeOption_Deterministic, arena,
+      &data, &size);
+
+  if (encode_status == kUpb_EncodeStatus_Ok) {
+    uint64_t ret = _upb_Hash(data, size, seed);
+    upb_Arena_Free(arena);
+    return ret;
+  }
+
+  upb_Arena_Free(arena);
+  upb_Status_SetErrorMessage(status, "Error calculating hash");
+  return 0;
+}

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/shared_message.h

@@ -0,0 +1,21 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2023 Google Inc.  All rights reserved.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file or at
+// https://developers.google.com/open-source/licenses/bsd
+
+// -----------------------------------------------------------------------------
+// Ruby Message functions. Strictly free of dependencies on
+// Ruby interpreter internals.
+
+#ifndef RUBY_PROTOBUF_SHARED_MESSAGE_H_
+#define RUBY_PROTOBUF_SHARED_MESSAGE_H_
+
+#include "ruby-upb.h"
+
+// Returns a hash value for the given message.
+uint64_t shared_Message_Hash(const upb_Message* msg, const upb_MessageDef* m,
+                             uint64_t seed, upb_Status* status);
+
+#endif  // RUBY_PROTOBUF_SHARED_MESSAGE_H_

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/third_party/utf8_range/LICENSE

@@ -0,0 +1,22 @@
+MIT License
+
+Copyright (c) 2019 Yibo Cai
+Copyright 2022 Google LLC
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/third_party/utf8_range/utf8_range.c

@@ -0,0 +1,467 @@
+// Copyright 2023 Google LLC
+//
+// Use of this source code is governed by an MIT-style
+// license that can be found in the LICENSE file or at
+// https://opensource.org/licenses/MIT.
+
+/* This is a wrapper for the Google range-sse.cc algorithm which checks whether
+ * a sequence of bytes is a valid UTF-8 sequence and finds the longest valid
+ * prefix of the UTF-8 sequence.
+ *
+ * The key difference is that it checks for as much ASCII symbols as possible
+ * and then falls back to the range-sse.cc algorithm. The changes to the
+ * algorithm are cosmetic, mostly to trick the clang compiler to produce optimal
+ * code.
+ *
+ * For API see the utf8_validity.h header.
+ */
+#include "utf8_range.h"
+
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+
+#ifdef __SSE4_1__
+#include <emmintrin.h>
+#include <smmintrin.h>
+#include <tmmintrin.h>
+#endif
+
+#if defined(__GNUC__)
+#define FORCE_INLINE_ATTR __attribute__((always_inline))
+#elif defined(_MSC_VER)
+#define FORCE_INLINE_ATTR __forceinline
+#else
+#define FORCE_INLINE_ATTR
+#endif
+
+static FORCE_INLINE_ATTR inline uint64_t utf8_range_UnalignedLoad64(
+    const void* p) {
+  uint64_t t;
+  memcpy(&t, p, sizeof t);
+  return t;
+}
+
+static FORCE_INLINE_ATTR inline int utf8_range_AsciiIsAscii(unsigned char c) {
+  return c < 128;
+}
+
+static FORCE_INLINE_ATTR inline int utf8_range_IsTrailByteOk(const char c) {
+  return (int8_t)(c) <= (int8_t)(0xBF);
+}
+
+/* If return_position is false then it returns 1 if |data| is a valid utf8
+ * sequence, otherwise returns 0.
+ * If return_position is set to true, returns the length in bytes of the prefix
+   of |data| that is all structurally valid UTF-8.
+ */
+static size_t utf8_range_ValidateUTF8Naive(const char* data, const char* end,
+                                           int return_position) {
+  /* We return err_pos in the loop which is always 0 if !return_position */
+  size_t err_pos = 0;
+  size_t codepoint_bytes = 0;
+  /* The early check is done because of early continue's on codepoints of all
+   * sizes, i.e. we first check for ascii and if it is, we call continue, then
+   * for 2 byte codepoints, etc. This is done in order to reduce indentation and
+   * improve readability of the codepoint validity check.
+   */
+  while (data + codepoint_bytes < end) {
+    if (return_position) {
+      err_pos += codepoint_bytes;
+    }
+    data += codepoint_bytes;
+    const size_t len = end - data;
+    const unsigned char byte1 = data[0];
+
+    /* We do not skip many ascii bytes at the same time as this function is
+       used for tail checking (< 16 bytes) and for non x86 platforms. We also
+       don't think that cases where non-ASCII codepoints are followed by ascii
+       happen often. For small strings it also introduces some penalty. For
+       purely ascii UTF8 strings (which is the overwhelming case) we call
+       SkipAscii function which is multiplatform and extremely fast.
+     */
+    /* [00..7F] ASCII -> 1 byte */
+    if (utf8_range_AsciiIsAscii(byte1)) {
+      codepoint_bytes = 1;
+      continue;
+    }
+    /* [C2..DF], [80..BF] -> 2 bytes */
+    if (len >= 2 && byte1 >= 0xC2 && byte1 <= 0xDF &&
+        utf8_range_IsTrailByteOk(data[1])) {
+      codepoint_bytes = 2;
+      continue;
+    }
+    if (len >= 3) {
+      const unsigned char byte2 = data[1];
+      const unsigned char byte3 = data[2];
+
+      /* Is byte2, byte3 between [0x80, 0xBF]
+       * Check for 0x80 was done above.
+       */
+      if (!utf8_range_IsTrailByteOk(byte2) ||
+          !utf8_range_IsTrailByteOk(byte3)) {
+        return err_pos;
+      }
+
+      if (/* E0, A0..BF, 80..BF */
+          ((byte1 == 0xE0 && byte2 >= 0xA0) ||
+           /* E1..EC, 80..BF, 80..BF */
+           (byte1 >= 0xE1 && byte1 <= 0xEC) ||
+           /* ED, 80..9F, 80..BF */
+           (byte1 == 0xED && byte2 <= 0x9F) ||
+           /* EE..EF, 80..BF, 80..BF */
+           (byte1 >= 0xEE && byte1 <= 0xEF))) {
+        codepoint_bytes = 3;
+        continue;
+      }
+      if (len >= 4) {
+        const unsigned char byte4 = data[3];
+        /* Is byte4 between 0x80 ~ 0xBF */
+        if (!utf8_range_IsTrailByteOk(byte4)) {
+          return err_pos;
+        }
+
+        if (/* F0, 90..BF, 80..BF, 80..BF */
+            ((byte1 == 0xF0 && byte2 >= 0x90) ||
+             /* F1..F3, 80..BF, 80..BF, 80..BF */
+             (byte1 >= 0xF1 && byte1 <= 0xF3) ||
+             /* F4, 80..8F, 80..BF, 80..BF */
+             (byte1 == 0xF4 && byte2 <= 0x8F))) {
+          codepoint_bytes = 4;
+          continue;
+        }
+      }
+    }
+    return err_pos;
+  }
+  if (return_position) {
+    err_pos += codepoint_bytes;
+  }
+  /* if return_position is false, this returns 1.
+   * if return_position is true, this returns err_pos.
+   */
+  return err_pos + (1 - return_position);
+}
+
+#ifdef __SSE4_1__
+/* Returns the number of bytes needed to skip backwards to get to the first
+   byte of codepoint.
+ */
+static inline int utf8_range_CodepointSkipBackwards(int32_t codepoint_word) {
+  const int8_t* const codepoint = (const int8_t*)(&codepoint_word);
+  if (!utf8_range_IsTrailByteOk(codepoint[3])) {
+    return 1;
+  } else if (!utf8_range_IsTrailByteOk(codepoint[2])) {
+    return 2;
+  } else if (!utf8_range_IsTrailByteOk(codepoint[1])) {
+    return 3;
+  }
+  return 0;
+}
+#endif  // __SSE4_1__
+
+/* Skipping over ASCII as much as possible, per 8 bytes. It is intentional
+   as most strings to check for validity consist only of 1 byte codepoints.
+ */
+static inline const char* utf8_range_SkipAscii(const char* data,
+                                               const char* end) {
+  while (8 <= end - data &&
+         (utf8_range_UnalignedLoad64(data) & 0x8080808080808080) == 0) {
+    data += 8;
+  }
+  while (data < end && utf8_range_AsciiIsAscii(*data)) {
+    ++data;
+  }
+  return data;
+}
+
+static FORCE_INLINE_ATTR inline size_t utf8_range_Validate(
+    const char* data, size_t len, int return_position) {
+  if (len == 0) return 1 - return_position;
+  const char* const end = data + len;
+  data = utf8_range_SkipAscii(data, end);
+  /* SIMD algorithm always outperforms the naive version for any data of
+     length >=16.
+   */
+  if (end - data < 16) {
+    return (return_position ? (data - (end - len)) : 0) +
+           utf8_range_ValidateUTF8Naive(data, end, return_position);
+  }
+#ifndef __SSE4_1__
+  return (return_position ? (data - (end - len)) : 0) +
+         utf8_range_ValidateUTF8Naive(data, end, return_position);
+#else
+  /* This code checks that utf-8 ranges are structurally valid 16 bytes at once
+   * using superscalar instructions.
+   * The mapping between ranges of codepoint and their corresponding utf-8
+   * sequences is below.
+   */
+
+  /*
+   * U+0000...U+007F     00...7F
+   * U+0080...U+07FF     C2...DF 80...BF
+   * U+0800...U+0FFF     E0      A0...BF 80...BF
+   * U+1000...U+CFFF     E1...EC 80...BF 80...BF
+   * U+D000...U+D7FF     ED      80...9F 80...BF
+   * U+E000...U+FFFF     EE...EF 80...BF 80...BF
+   * U+10000...U+3FFFF   F0      90...BF 80...BF 80...BF
+   * U+40000...U+FFFFF   F1...F3 80...BF 80...BF 80...BF
+   * U+100000...U+10FFFF F4      80...8F 80...BF 80...BF
+   */
+
+  /* First we compute the type for each byte, as given by the table below.
+   * This type will be used as an index later on.
+   */
+
+  /*
+   * Index  Min Max Byte Type
+   *  0     00  7F  Single byte sequence
+   *  1,2,3 80  BF  Second, third and fourth byte for many of the sequences.
+   *  4     A0  BF  Second byte after E0
+   *  5     80  9F  Second byte after ED
+   *  6     90  BF  Second byte after F0
+   *  7     80  8F  Second byte after F4
+   *  8     C2  F4  First non ASCII byte
+   *  9..15 7F  80  Invalid byte
+   */
+
+  /* After the first step we compute the index for all bytes, then we permute
+     the bytes according to their indices to check the ranges from the range
+     table.
+   * The range for a given type can be found in the range_min_table and
+     range_max_table, the range for type/index X is in range_min_table[X] ...
+     range_max_table[X].
+   */
+
+  /* Algorithm:
+   * Put index zero to all bytes.
+   * Find all non ASCII characters, give them index 8.
+   * For each tail byte in a codepoint sequence, give it an index corresponding
+     to the 1 based index from the end.
+   * If the first byte of the codepoint is in the [C0...DF] range, we write
+     index 1 in the following byte.
+   * If the first byte of the codepoint is in the range [E0...EF], we write
+     indices 2 and 1 in the next two bytes.
+   * If the first byte of the codepoint is in the range [F0...FF] we write
+     indices 3,2,1 into the next three bytes.
+   * For finding the number of bytes we need to look at high nibbles (4 bits)
+     and do the lookup from the table, it can be done with shift by 4 + shuffle
+     instructions. We call it `first_len`.
+   * Then we shift first_len by 8 bits to get the indices of the 2nd bytes.
+   * Saturating sub 1 and shift by 8 bits to get the indices of the 3rd bytes.
+   * Again to get the indices of the 4th bytes.
+   * Take OR of all that 4 values and check within range.
+   */
+  /* For example:
+   * input       C3 80 68 E2 80 20 A6 F0 A0 80 AC 20 F0 93 80 80
+   * first_len   1  0  0  2  0  0  0  3  0  0  0  0  3  0  0  0
+   * 1st byte    8  0  0  8  0  0  0  8  0  0  0  0  8  0  0  0
+   * 2nd byte    0  1  0  0  2  0  0  0  3  0  0  0  0  3  0  0 // Shift + sub
+   * 3rd byte    0  0  0  0  0  1  0  0  0  2  0  0  0  0  2  0 // Shift + sub
+   * 4th byte    0  0  0  0  0  0  0  0  0  0  1  0  0  0  0  1 // Shift + sub
+   * Index       8  1  0  8  2  1  0  8  3  2  1  0  8  3  2  1 // OR of results
+   */
+
+  /* Checking for errors:
+   * Error checking is done by looking up the high nibble (4 bits) of each byte
+     against an error checking table.
+   * Because the lookup value for the second byte depends of the value of the
+     first byte in codepoint, we use saturated operations to adjust the index.
+   * Specifically we need to add 2 for E0, 3 for ED, 3 for F0 and 4 for F4 to
+     match the correct index.
+       * If we subtract from all bytes EF then EO -> 241, ED -> 254, F0 -> 1,
+         F4 -> 5
+       * Do saturating sub 240, then E0 -> 1, ED -> 14 and we can do lookup to
+         match the adjustment
+       * Add saturating 112, then F0 -> 113, F4 -> 117, all that were > 16 will
+         be more 128 and lookup in ef_fe_table will return 0 but for F0
+         and F4 it will be 4 and 5 accordingly
+   */
+  /*
+   * Then just check the appropriate ranges with greater/smaller equal
+     instructions. Check tail with a naive algorithm.
+   * To save from previous 16 byte checks we just align previous_first_len to
+     get correct continuations of the codepoints.
+   */
+
+  /*
+   * Map high nibble of "First Byte" to legal character length minus 1
+   * 0x00 ~ 0xBF --> 0
+   * 0xC0 ~ 0xDF --> 1
+   * 0xE0 ~ 0xEF --> 2
+   * 0xF0 ~ 0xFF --> 3
+   */
+  const __m128i first_len_table =
+      _mm_setr_epi8(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 2, 3);
+
+  /* Map "First Byte" to 8-th item of range table (0xC2 ~ 0xF4) */
+  const __m128i first_range_table =
+      _mm_setr_epi8(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 8, 8, 8, 8);
+
+  /*
+   * Range table, map range index to min and max values
+   */
+  const __m128i range_min_table =
+      _mm_setr_epi8(0x00, 0x80, 0x80, 0x80, 0xA0, 0x80, 0x90, 0x80, 0xC2, 0x7F,
+                    0x7F, 0x7F, 0x7F, 0x7F, 0x7F, 0x7F);
+
+  const __m128i range_max_table =
+      _mm_setr_epi8(0x7F, 0xBF, 0xBF, 0xBF, 0xBF, 0x9F, 0xBF, 0x8F, 0xF4, 0x80,
+                    0x80, 0x80, 0x80, 0x80, 0x80, 0x80);
+
+  /*
+   * Tables for fast handling of four special First Bytes(E0,ED,F0,F4), after
+   * which the Second Byte are not 80~BF. It contains "range index adjustment".
+   * +------------+---------------+------------------+----------------+
+   * | First Byte | original range| range adjustment | adjusted range |
+   * +------------+---------------+------------------+----------------+
+   * | E0         | 2             | 2                | 4              |
+   * +------------+---------------+------------------+----------------+
+   * | ED         | 2             | 3                | 5              |
+   * +------------+---------------+------------------+----------------+
+   * | F0         | 3             | 3                | 6              |
+   * +------------+---------------+------------------+----------------+
+   * | F4         | 4             | 4                | 8              |
+   * +------------+---------------+------------------+----------------+
+   */
+
+  /* df_ee_table[1] -> E0, df_ee_table[14] -> ED as ED - E0 = 13 */
+  // The values represent the adjustment in the Range Index table for a correct
+  // index.
+  const __m128i df_ee_table =
+      _mm_setr_epi8(0, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3, 0);
+
+  /* ef_fe_table[1] -> F0, ef_fe_table[5] -> F4, F4 - F0 = 4 */
+  // The values represent the adjustment in the Range Index table for a correct
+  // index.
+  const __m128i ef_fe_table =
+      _mm_setr_epi8(0, 3, 0, 0, 0, 4, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0);
+
+  __m128i prev_input = _mm_set1_epi8(0);
+  __m128i prev_first_len = _mm_set1_epi8(0);
+  __m128i error = _mm_set1_epi8(0);
+  while (end - data >= 16) {
+    const __m128i input =
+        _mm_loadu_si128((const __m128i*)(data));
+
+    /* high_nibbles = input >> 4 */
+    const __m128i high_nibbles =
+        _mm_and_si128(_mm_srli_epi16(input, 4), _mm_set1_epi8(0x0F));
+
+    /* first_len = legal character length minus 1 */
+    /* 0 for 00~7F, 1 for C0~DF, 2 for E0~EF, 3 for F0~FF */
+    /* first_len = first_len_table[high_nibbles] */
+    __m128i first_len = _mm_shuffle_epi8(first_len_table, high_nibbles);
+
+    /* First Byte: set range index to 8 for bytes within 0xC0 ~ 0xFF */
+    /* range = first_range_table[high_nibbles] */
+    __m128i range = _mm_shuffle_epi8(first_range_table, high_nibbles);
+
+    /* Second Byte: set range index to first_len */
+    /* 0 for 00~7F, 1 for C0~DF, 2 for E0~EF, 3 for F0~FF */
+    /* range |= (first_len, prev_first_len) << 1 byte */
+    range = _mm_or_si128(range, _mm_alignr_epi8(first_len, prev_first_len, 15));
+
+    /* Third Byte: set range index to saturate_sub(first_len, 1) */
+    /* 0 for 00~7F, 0 for C0~DF, 1 for E0~EF, 2 for F0~FF */
+    __m128i tmp1;
+    __m128i tmp2;
+    /* tmp1 = saturate_sub(first_len, 1) */
+    tmp1 = _mm_subs_epu8(first_len, _mm_set1_epi8(1));
+    /* tmp2 = saturate_sub(prev_first_len, 1) */
+    tmp2 = _mm_subs_epu8(prev_first_len, _mm_set1_epi8(1));
+    /* range |= (tmp1, tmp2) << 2 bytes */
+    range = _mm_or_si128(range, _mm_alignr_epi8(tmp1, tmp2, 14));
+
+    /* Fourth Byte: set range index to saturate_sub(first_len, 2) */
+    /* 0 for 00~7F, 0 for C0~DF, 0 for E0~EF, 1 for F0~FF */
+    /* tmp1 = saturate_sub(first_len, 2) */
+    tmp1 = _mm_subs_epu8(first_len, _mm_set1_epi8(2));
+    /* tmp2 = saturate_sub(prev_first_len, 2) */
+    tmp2 = _mm_subs_epu8(prev_first_len, _mm_set1_epi8(2));
+    /* range |= (tmp1, tmp2) << 3 bytes */
+    range = _mm_or_si128(range, _mm_alignr_epi8(tmp1, tmp2, 13));
+
+    /*
+     * Now we have below range indices calculated
+     * Correct cases:
+     * - 8 for C0~FF
+     * - 3 for 1st byte after F0~FF
+     * - 2 for 1st byte after E0~EF or 2nd byte after F0~FF
+     * - 1 for 1st byte after C0~DF or 2nd byte after E0~EF or
+     *         3rd byte after F0~FF
+     * - 0 for others
+     * Error cases:
+     *   >9 for non ascii First Byte overlapping
+     *   E.g., F1 80 C2 90 --> 8 3 10 2, where 10 indicates error
+     */
+
+    /* Adjust Second Byte range for special First Bytes(E0,ED,F0,F4) */
+    /* Overlaps lead to index 9~15, which are illegal in range table */
+    __m128i shift1;
+    __m128i pos;
+    __m128i range2;
+    /* shift1 = (input, prev_input) << 1 byte */
+    shift1 = _mm_alignr_epi8(input, prev_input, 15);
+    pos = _mm_sub_epi8(shift1, _mm_set1_epi8(0xEF));
+    /*
+     * shift1:  | EF  F0 ... FE | FF  00  ... ...  DE | DF  E0 ... EE |
+     * pos:     | 0   1      15 | 16  17           239| 240 241    255|
+     * pos-240: | 0   0      0  | 0   0            0  | 0   1      15 |
+     * pos+112: | 112 113    127|       >= 128        |     >= 128    |
+     */
+    tmp1 = _mm_subs_epu8(pos, _mm_set1_epi8(-16));
+    range2 = _mm_shuffle_epi8(df_ee_table, tmp1);
+    tmp2 = _mm_adds_epu8(pos, _mm_set1_epi8(112));
+    range2 = _mm_add_epi8(range2, _mm_shuffle_epi8(ef_fe_table, tmp2));
+
+    range = _mm_add_epi8(range, range2);
+
+    /* Load min and max values per calculated range index */
+    __m128i min_range = _mm_shuffle_epi8(range_min_table, range);
+    __m128i max_range = _mm_shuffle_epi8(range_max_table, range);
+
+    /* Check value range */
+    if (return_position) {
+      error = _mm_cmplt_epi8(input, min_range);
+      error = _mm_or_si128(error, _mm_cmpgt_epi8(input, max_range));
+      /* 5% performance drop from this conditional branch */
+      if (!_mm_testz_si128(error, error)) {
+        break;
+      }
+    } else {
+      error = _mm_or_si128(error, _mm_cmplt_epi8(input, min_range));
+      error = _mm_or_si128(error, _mm_cmpgt_epi8(input, max_range));
+    }
+
+    prev_input = input;
+    prev_first_len = first_len;
+
+    data += 16;
+  }
+  /* If we got to the end, we don't need to skip any bytes backwards */
+  if (return_position && (data - (end - len)) == 0) {
+    return utf8_range_ValidateUTF8Naive(data, end, return_position);
+  }
+  /* Find previous codepoint (not 80~BF) */
+  data -= utf8_range_CodepointSkipBackwards(_mm_extract_epi32(prev_input, 3));
+  if (return_position) {
+    return (data - (end - len)) +
+           utf8_range_ValidateUTF8Naive(data, end, return_position);
+  }
+  /* Test if there was any error */
+  if (!_mm_testz_si128(error, error)) {
+    return 0;
+  }
+  /* Check the tail */
+  return utf8_range_ValidateUTF8Naive(data, end, return_position);
+#endif
+}
+
+int utf8_range_IsValid(const char* data, size_t len) {
+  return utf8_range_Validate(data, len, /*return_position=*/0) != 0;
+}
+
+size_t utf8_range_ValidPrefix(const char* data, size_t len) {
+  return utf8_range_Validate(data, len, /*return_position=*/1);
+}

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/third_party/utf8_range/utf8_range.h

@@ -0,0 +1,22 @@
+#ifndef THIRD_PARTY_UTF8_RANGE_UTF8_RANGE_H_
+#define THIRD_PARTY_UTF8_RANGE_UTF8_RANGE_H_
+
+#include <stddef.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// Returns 1 if the sequence of characters is a valid UTF-8 sequence, otherwise
+// 0.
+int utf8_range_IsValid(const char* data, size_t len);
+
+// Returns the length in bytes of the prefix of str that is all
+// structurally valid UTF-8.
+size_t utf8_range_ValidPrefix(const char* data, size_t len);
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#endif  // THIRD_PARTY_UTF8_RANGE_UTF8_RANGE_H_

bin/gems/google-protobuf-4.29.3-x86_64-linux/ext/google/protobuf_c/wrap_memcpy.c

@@ -0,0 +1,29 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2017 Google Inc.  All rights reserved.
+//
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file or at
+// https://developers.google.com/open-source/licenses/bsd
+
+#include <string.h>
+
+// On x86-64 Linux with glibc, we link against the 2.2.5 version of memcpy so
+// that we avoid depending on the 2.14 version of the symbol. This way,
+// distributions that are using pre-2.14 versions of glibc can successfully use
+// the gem we distribute
+// (https://github.com/protocolbuffers/protobuf/issues/2783).
+//
+// This wrapper is enabled by passing the linker flags -Wl,-wrap,memcpy in
+// extconf.rb.
+#ifdef __linux__
+#if defined(__x86_64__) && defined(__GNU_LIBRARY__)
+__asm__(".symver memcpy,memcpy@GLIBC_2.2.5");
+void *__wrap_memcpy(void *dest, const void *src, size_t n) {
+  return memcpy(dest, src, n);
+}
+#else
+void *__wrap_memcpy(void *dest, const void *src, size_t n) {
+  return memmove(dest, src, n);
+}
+#endif
+#endif