Flutter导出tflite_flutter至iOS原生项目遇符号查找失败求助

问题场景 #

环境：Flutter 3.7.0、Xcode 14；依赖：tflite_flutter 0.10.4
将Flutter模块导出为Framework集成到iOS原生项目后，运行时报错：

Unhandled Exception: Invalid argument(s): Failed to lookup symbol
'TfLiteInterpreterOptionsCreate': dlsym(RTLD_DEFAULT, TfLiteInterpreterOptionsCreate): symbol not found
#0 DynamicLibrary.lookup (dart:ffi-patch/ffi_dynamic_library_patch.dart:33)
#1 TensorFlowLiteBindings._TfLiteInterpreterOptionsCreatePtr (package:tflite_flutter/src/bindings/tensorflow_lite_bindings_generated.dart:336)
#2 TensorFlowLiteBindings._TfLiteInterpreterOptionsCreate (package:tflite_flutter/src/bindings/tensorflow_lite_bindings_generated.dart)
#3 TensorFlowLiteBindings.TfLiteInterpreterOptionsCreate (package:tflite_flutter/src/bindings/tensorflow_lite_bindings_generated.dart)
#4 new InterpreterOptions (package:tflite_flutter/src/interpreter_options.dart:36)
#5 Detector._loadModel (package:tf_module/service/detector_service.dart:109)
#6 Detector.start (package:tf_module/service/detector_service.dart:99)
<asynchronous suspension>

已尝试修改Xcode中Target Runner>Build Settings>Strip Style为Non-Global Symbols，未生效。

解决方案 #

• 确认TensorFlow Lite库的嵌入状态
  导出Flutter Framework后，检查原生iOS项目的Frameworks, Libraries, and Embedded Content中是否包含libtensorflowlite_c.framework。如果没有，从Flutter项目的ios/Pods/tflite_flutter路径下找到该框架，添加到原生项目中，并设置Embed选项为Embed & Sign。

• 调整原生主Target的符号剥离配置
  不要只修改Runner Target，需要针对原生项目的主Target修改以下Build Settings：

  • 将Strip Debug Symbols During Copy设置为NO
  • 将Strip Style设置为Non-Global Symbols
  • 将Dead Code Stripping设置为NO

• 更新tflite_flutter依赖版本
  tflite_flutter 0.10.4与Flutter 3.7.0、Xcode 14的兼容性存在问题，执行flutter pub upgrade tflite_flutter升级到最新稳定版（如0.14.0+），之后重新导出Flutter Framework并集成。

• 手动指定动态库加载路径
  在Flutter代码初始化Interpreter时，手动指定TensorFlow Lite库的路径，避免动态库查找失败：

  import 'dart:ffi';
  import 'dart:io';
  import 'package:path/path.dart' as path;
  import 'package:tflite_flutter/tflite_flutter.dart';
  
  Future<void> loadModel() async {
    final libraryPath = path.join(
      Directory.current.path,
      'Frameworks',
      'libtensorflowlite_c.framework',
      'libtensorflowlite_c',
    );
    final dylib = DynamicLibrary.open(libraryPath);
    final interpreter = await Interpreter.fromAsset(
      'your_model.tflite',
      dylib: dylib,
      options: InterpreterOptions()..useGpuDelegate = false,
    );
    // 后续模型操作
  }
  

内容的提问来源于stack exchange，提问作者Ray Gu