KotlinCoreEnvironment.underApplicationLock {
                if (applicationArea.hasExtensionPoint(ClassTypePointerFactory.EP_NAME)) return@underApplicationLock
                CoreApplicationEnvironment.registerApplicationExtensionPoint(
                    ClassTypePointerFactory.EP_NAME,
                    ClassTypePointerFactory::class.java
                )

using ::tensorflow::shape_inference::InferenceContext;
using ::tensorflow::shape_inference::ShapeHandle;

TF_OpDefinitionBuilder* TF_NewOpDefinitionBuilder(const char* op_name) {
  auto* result = new OpDefBuilder(op_name);
  return reinterpret_cast<TF_OpDefinitionBuilder*>(result);
}

void TF_DeleteOpDefinitionBuilder(TF_OpDefinitionBuilder* builder) {
  delete reinterpret_cast<OpDefBuilder*>(builder);
}

func dumpRegexp(b *strings.Builder, re *Regexp) {
	if int(re.Op) >= len(opNames) || opNames[re.Op] == "" {
		fmt.Fprintf(b, "op%d", re.Op)
	} else {
		switch re.Op {
		default:
			b.WriteString(opNames[re.Op])
		case OpStar, OpPlus, OpQuest, OpRepeat:
			if re.Flags&NonGreedy != 0 {
				b.WriteByte('n')
			}
			b.WriteString(opNames[re.Op])
		case OpLiteral:
			if len(re.Rune) > 1 {
				b.WriteString("str")

      return failure();

    Operation *inner_op;

    StringRef op_name = op->getName().stripDialect();
    if (!func_symbols_.contains(op_name)) {
      std::string tf_op_name = llvm::formatv(
          "{0}.{1}", TF::TensorFlowDialect::getDialectNamespace(), op_name);
      OperationState state =
          OperationState(loc, tf_op_name, inner_op_operands, new_types, attrs,

  // `TF_AttrBuilderCheckCanRunOnDevice()` call(s) on the same `builder`.
  std::set<std::string> attr_names;
};

TF_AttrBuilder* TF_NewAttrBuilder(const char* op_name) {
  return new TF_AttrBuilder(op_name);
}

void TF_DeleteAttrBuilder(TF_AttrBuilder* builder) { delete builder; }

void TF_AttrBuilderSetType(TF_AttrBuilder* builder, const char* attr_name,

  return op;
}

TFE_Op* SendOp(TFE_Context* ctx, TFE_TensorHandle* in,
               const std::string& op_name, const std::string& send_device,
               const std::string& recv_device,
               tensorflow::uint64 send_device_incarnation) {
  TF_Status* status = TF_NewStatus();
  TFE_Op* op = TFE_NewOp(ctx, op_name.c_str(), status);
  CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TFE_OpAddInput(op, in, status);

    auto custom_op = llvm::isa<CustomOp>(op);
    if (custom_op) {
      auto q = llvm::cast<CustomOp>(op);
      std::string op_name = q.getCustomCode().str();
      if ((custom_op_map.find(op_name) == custom_op_map.end()) ||
          !custom_op_map.find(op_name)->second.no_side_effect)
        return failure();
    }
    rewriter.eraseOp(op);
    return success();
  }
  quant::CustomOpMap custom_op_map;

	return fmt.Sprintf("Prefix(%#x)", int(p))
}

// An Op is an x86 opcode.
type Op uint32

func (op Op) String() string {
	i := int(op)
	if i < 0 || i >= len(opNames) || opNames[i] == "" {
		return fmt.Sprintf("Op(%d)", i)
	}
	return opNames[i]
}

// An Args holds the instruction arguments.
// If an instruction has fewer than 4 arguments,
// the final elements in the array are nil.
type Args [4]Arg

        return element_type->isF32() || element_type->isInteger(32);
      }
    }
  }

  // Op is vacuously "supported" if it is not a tensorlist op.
  StringRef op_name = op->getName().getStringRef();
  if (!op_name.contains("TensorList")) return true;

  std::optional<mlir::Type> element_type = {};

  if (auto reserve = llvm::dyn_cast_or_null<TF::TensorListReserveOp>(op)) {

    return std::string("tf.If");
  }

  llvm::StringRef op_name(tflite::EnumNameBuiltinOperator(builtin_code));

  // If the Op name contains stablehlo
  if (IsStablehloOp(op_code)) {
    return llvm::Twine(
               llvm::Twine("vhlo.", op_name.drop_front(10).lower()).str(),
               "_v1")
        .str();
  }
  return llvm::Twine("tfl.", op_name.lower()).str();
}