Q音直播Flutter包裁剪方案(iOS)

原创 allentywang 腾讯音乐技术团队

1、背景

Flutter作为一款优秀的跨平台方案，我们Q音团队一致保持高度关注，团队内部也一直在努力促进Flutter的应用框架建设。在Q音直播接入Flutter的过程中，需要解决的首要问题便是”Flutter包体积变大”。本文将一步步剖析Flutter的包体积问题，带领大家探寻每一个可能的包体积优化点，结合实际项目和引擎源码，最终给出详细的包体积优化实现方案。欢迎大家相互交流Flutter相关技术。

1.1 Flutter混合开发模式

一般的如果我们想在现有原生App中加入Flutter，需要通过以下两种方式对Flutter进行引入：

将原生工程作为 Flutter 工程的子工程，由 Flutter 统一管理。这种模式，就是统一管理模式，即讲Flutter作为子工程集成到项目中。
将 Flutter 工程作为原生工程共用的子模块，维持原有的原生工程管理方式不变。这种模式，就是三端分离模式，即Flutter单独作为一个端进行开发。

统一管理模式搭建简单，但是缺点明显，不仅三端（Android、iOS、Flutter）代码耦合严重，相关工具链耗时也随之大幅增长，导致开发效率降低。所以这里使用三端代码分离的模式来进行依赖治理，实现了 Flutter 工程的轻量级接入。即 Android 侧使用 aar集成Flutter产物、iOS 使用 pod集成Flutter产物。

1.2 Flutter瘦身需求

当App引入Flutter带来一个明显问题，包体积增大。对于三端分离模式，包体积增量在Android上即为Flutter的aar产物，在iOS上表现为Flutter的framwork产物。因此，要解决包体积问题，需要对aar和framework的体积进行优化。

1.3 本文内容涉及的开发环境

Flutter 1.17.1 • channel stable
Mac OS X 10.15.4
Xcode - develop for iOS and macOS (Xcode 11.4)
Python 2.7.16

2、iOS framework产物分析

我们在实际工程中使用的是产物集成方式，Flutter代码会被编译打包成一个framework，即Flutter业务代码将会以framework的形式带入iOS宿主App。那么如何去对这个framework进行体积优化呢？下面我们首先对framework的内容进行详细分析。

2.1 framework结构

以Release模式下Futter产物为例，使用tree命令查看Release目录结构，我们可以看到iOS产物为两个framework，其中App.framework是Dart业务代码产物，Flutter.framework是从Flutter SDK中拷贝过来的引擎产物。下图中，我给体积占比大的文件添加了说明，其它Headers、和plist文件大小可以忽略不计。

Release
    ├── App.framework
    │   ├── App            //AOT Snapshot数据，由我们的Dart业务代码编译而成，Mach-O格式的动态链接库
    │   ├── Info.plist     
    │   └── flutter_assets //资源文件存放
    └── Flutter.framework
        ├── Flutter        //Flutter引擎，Mach-O格式的动态链接库
        ├── Headers
        ├── Info.plist
        ├── Modules
        ├── _CodeSignature
        └── icudtl.dat     //国际化支持相关文件

2.2 Framework体积分析（Release模式下）

我们将Release目录下的大文件以表格形式列出，这些文件即是我们去做体积优化的方向。

名称	大小	说明
App	7.3M	Dart业务代码AOT编译产物
flutter_assets	2M	图片、字体等资源文件
Flutter	11M	引擎
icudtl.dat	884k	国际化支持相关文件
其他第三方插件	800K	Flutter_boost等第三方插件

3、iOS减包思路

上一节我们分析了framework产物的组成占比，并且列出了体积最大的4个文件：App、flutter_assets、Flutter、icudtl.dat。下面我们继续对这4个文件进行深入分析，看看是否有优化空间。

优化思路分为3个方向

缩。即自我数据压缩，这种方法能够减少framework的体积，但是对最终app打包出来的体积影响较小，因为打包也是进行了数据压缩。
删。删除无用部分，或者不需要用到的部分。
挪。如果不能删除，考虑是否能分离出来，通过下载的方式动态去加载分离的部分。

3.1 App.framework/flutter_assets

flutter_assets目录存放的资源文件，如果不想flutter_assets带入App，我们可以将其移出，在运行需要时动态下载。移除方法具体有以下两个方法。

修改flutter_tools编译打包脚本。在生成framework过程中，就将flutter_asserts移除保存到别处，即生成的framework天生与flutter_asserts就是分开的。优点是得到即可用，隐藏中间过程。
生成framework后进行二次处理。在生成framework后，通过脚本将framework中的flutter_asserts移除。优点是不需要修改打包工具flutter_tools源码，缺点是增加了脚本对framework的处理，拉长了工作流程。

当前使用的是第二种方法，直接对产物进行二次处理，只为一个flutter_assets修改打包源码有点得不偿失。

移除flutter_assets后对引擎启动是否有影响？查看源码发现flutter_assets在FlutterDartProject.mm的DefaultSettingsForProcess函数中被使用，初始化过程中会在app目录检查是否有flutter_assets，并且将路径保存在Settings的assets_path变量中，供引擎后续使用。因此我们得出结论：flutter_assets是放在Framework内部，还是动态下载下来的，对程序运行没有影响，只要将flutter_assets的正确位置的告知引擎即可。

//FlutterDartProject.mm// Checks to see if the flutter assets directory is already present.
  if (settings.assets_path.size() == 0) {
    NSString* assetsName = [FlutterDartProject flutterAssetsName:bundle];
    NSString* assetsPath = [bundle pathForResource:assetsName ofType:@""];    if (assetsPath.length == 0) {
      assetsPath = [mainBundle pathForResource:assetsName ofType:@""];//主目录检查assetsPath
    }    if (assetsPath.length == 0) {
      NSLog(@"Failed to find assets path for \"%@\"", assetsName);
    } else {
      settings.assets_path = assetsPath.UTF8String;      // Check if there is an application kernel snapshot in the assets directory we could
      // potentially use.  Looking for the snapshot makes sense only if we have a VM that can use
      // it.
      if (!flutter::DartVM::IsRunningPrecompiledCode()) {
        NSURL* applicationKernelSnapshotURL =
            [NSURL URLWithString:@(kApplicationKernelSnapshotFileName)
                   relativeToURL:[NSURL fileURLWithPath:assetsPath]];        if ([[NSFileManager defaultManager] fileExistsAtPath:applicationKernelSnapshotURL.path]) {
          settings.application_kernel_asset = applicationKernelSnapshotURL.path.UTF8String;
        } else {
          NSLog(@"Failed to find snapshot: %@", applicationKernelSnapshotURL.path);
        }
      }
    }
  }

//这里列出settings.h中的相关Flutter配置代码struct Settings {
  Settings();

  Settings(const Settings& other);

  ~Settings();  //add by allentywang.  dart data path.
  std::string ios_vm_snapshot_data_path;  //自定义 vm data 路径
  std::string ios_isolate_snapshot_data_path; //自定义 isolate data 路径
  //end

  ...

  std::string icu_data_path; //这个是Flutter.framework中icudtl.dat的路径
  // Assets settings
  std::string assets_path; //这个是App.framework中flutter.assets的路径};

3.2 Flutter.framework/icudtl.dat

Flutter.framework中的icudtl.dat保存了引擎的国际化支持信息。通过查看源码发现，icudtl.dat在引擎初始化时被加载，依赖的只有Setting中的配置路径icu_data_path。因此这个也可以挪走。icudtl.dat大小固定为800多k，还是比较可观的。同样和flutter_assets一样，如果移走了icudtl.dat，我们需要在引擎初始化时指定外部的icudtl.dat路径。

  //settings.h中

  // The icu_initialization_required setting does not have a corresponding
  // switch because it is intended to be decided during build time, not runtime.
  // Some companies apply source modification here because their build system
  // brings its own ICU data files.
  bool icu_initialization_required = true;
  std::string icu_data_path; //icu路径
  MappingCallback icu_mapper;

//观察shell.cc源码，得出结论，可以将icudtl移除，只需要在初始化时重新指定settings中icu_data_path的路径即可if (settings.icu_initialization_required) {  if (settings.icu_data_path.size() != 0) {
    fml::icu::InitializeICU(settings.icu_data_path);//我们可以修改settings.icu_data_path路径，达到加载保存在外部的icu目的
  } else if (settings.icu_mapper) {
    fml::icu::InitializeICUFromMapping(settings.icu_mapper());
  } else {
    FML_DLOG(WARNING) << "Skipping ICU initialization in the shell.";
  }
}

3.3 App.Framework/App

我们App.Framework产物下有两个比较大的文件，一个是前面提到的flutter_assets，另外一个就是App。flutter_assets的优化前面已经讲过了，可以通过移除，然后动态下载的方式进行优化。那么App这个文件能否采用同样的方式呢？下面我们进行逐步分析。

3.3.1 App是什么？

App是dart代码编译出来的可执行文件，App体积还是比较大的，我们先对App内容进行分析，下面是使用nm命令显示的App内容

nm App
...
000000000038c440 t Precompiled_int_init__int64OverflowLimits_0150898_11508
000000000038a2b0 t Precompiled_int_parse_11499
000000000038a178 t Precompiled_int_tryParse_11498
0000000000392ae8 t Precompiled_num_parse_11570
0000000000392a74 t Precompiled_num_tryParse_11569
0000000000018c54 t Precompiled_pragma_get_name_131
000000000065c008 b _kDartIsolateSnapshotBss
00000000003a1270 S _kDartIsolateSnapshotData
0000000000009000 T _kDartIsolateSnapshotInstructions
000000000065c000 b _kDartVmSnapshotBss
0000000000399400 S _kDartVmSnapshotData
0000000000004000 T _kDartVmSnapshotInstructions
                 U dyld_stub_binder

注意这里我省略了相当多的符号，如果只想看Dart VM Data相关产物，可以使用xcrun strip命令进行过滤。自己研究的过程中发现Flutter1.9的版本没有这些Precompiled符号内容，原因是在打包脚本$FLUTTER_ROOT/packages/flutter_tools/bin/xcode_backend.sh中做了如下处理，而高版本的Flutter去掉了这些处理，保留了符号信息。

# 生成 dSYM 文件
RunCommand xcrun dsymutil -o "${build_dir}/dSYMs.noindex/App.framework.dSYM" "${app_framework}/App"

StreamOutput " ├─Stripping debug symbols..."
# 剥离调试符号表
RunCommand xcrun strip -x -S "${derived_dir}/App.framework/App"
StreamOutput "done"

过滤掉符号信息后，App中只剩下下面4个东西

kDartIsolateSnapshotData
kDarVmSnapshotData
kDartIsolateSnapshotInstructions
kDartVmSnapshotInstructions

这4个东西是什么？探究过程我就省略不细说了，这里直接说结论。kDartIsolateSnapshotData、kDarVmSnapshotData为可执行文件App的数据段，kDartIsolateSnapshotInstructions、kDartVmSnapshotInstructions为代码段，由于iOS系统的限制，整个App可执行文件不可以动态下发。但是kDartIsolateSnapshotData、kDartVmSnapshotData为数据段，它们在加载时不存在限制，可以动态加载。因此得出结论kDartIsolateSnapshotData和kDarVmSnapshotData可以挪到App外部，进行动态加载。

3.3.2 App内的kDartIsolateSnapshotData、kDarVmSnapshotData是如何生成的？

前面我们讲到kDartIsolateSnapshotData、kDarVmSnapshotData可以从App中移除，改为动态加载。那么它们是怎么被生成到App中的？我们又如何把它们从App中分离呢？

首先介绍一下Flutter虚拟机的运行模式。以iOS为例，Debug模式下Flutter的Dart虚拟机是JIT运行模式，JIT直接运行源码或者app.dill ，这也是Flutter热重载的原理。而在Release下，Flutter的Dart虚拟机是AOT运行模式，直接运行编译期编译好的机器码App。我们只能对Release模式下的App做文章，因为Debug模式下App包含很少的东西，里面没有可运行代码（这也是Debug的App.framework/App非常小的原因，使用nm查看App，发现里面什么都没有）。
AOT产物是如何生成的？查阅了Flutter源码和相关资料，我们发现Dart代码会使用gen_snapshot工具来编译成.S文件，然后通过xcrun工具来进行汇编和链接最终生成App。整个编译过程关键分为两步，一个是gen_sanpshot 编译，另外一个是xcrun编译。其中gen_snapshot由Flutter engine提供，属于DartVm代码部分，我们完全可以定制gen_sanpshot来改变App组成，进而达到减包的目的。

因此想要移除kDartIsolateSnapshotData、kDarVmSnapshotData，我们必须从gen_sanpshot入手

3.3.3 Dart源码的编译与加载

数据段编译过程：这里借用网上Dart源码编译成App.framework的流程图

数据段运行时加载过程：App运行前，Dart 虚拟机需要加载保存在App中的数据和代码，为了得到可供虚拟机运行的DartVMData，引擎初始化时按照下面步骤依次调用相关代码来完成DartVMData的创建。下面我从引擎源码上追踪了数据段的完整加载过程。

第一步，DartVMData::Create 根据Setting中的配置，调用VMSnapshotFromSettings、IsolateSnapshotFromSettings

std::shared_ptr<const DartVMData> DartVMData::Create(
    Settings settings,
    fml::RefPtr<DartSnapshot> vm_snapshot,
    fml::RefPtr<DartSnapshot> isolate_snapshot) {  if (!vm_snapshot || !vm_snapshot->IsValid()) {    // Caller did not provide a valid VM snapshot. Attempt to infer one
    // from the settings.
    vm_snapshot = DartSnapshot::VMSnapshotFromSettings(settings);//创建VM Snapshot
    if (!vm_snapshot) {
      FML_LOG(ERROR)
          << "VM snapshot invalid and could not be inferred from settings.";      return {};
    }
  }  if (!isolate_snapshot || !isolate_snapshot->IsValid()) {    // Caller did not provide a valid isolate snapshot. Attempt to infer one
    // from the settings.
    isolate_snapshot = DartSnapshot::IsolateSnapshotFromSettings(settings);//创建Isolate Snapshot
    if (!isolate_snapshot) {
      FML_LOG(ERROR) << "Isolate snapshot invalid and could not be inferred "
                        "from settings.";      return {};
    }
  }  return std::shared_ptr<const DartVMData>(new DartVMData(
      std::move(settings),         //
      std::move(vm_snapshot),      //
      std::move(isolate_snapshot)  //
      ));//生成DartVMData}

第二步，VMSnapshotFromSettings、IsolateSnapshotFromSettings 这里的作用是完成数据段和代码段的重建

fml::RefPtr<DartSnapshot> DartSnapshot::VMSnapshotFromSettings(    const Settings& settings) {
  TRACE_EVENT0("flutter", "DartSnapshot::VMSnapshotFromSettings");  auto snapshot =
      fml::MakeRefCounted<DartSnapshot>(ResolveVMData(settings),         //数据段重建
                                        ResolveVMInstructions(settings)  //代码段重建
      );  if (snapshot->IsValid()) {    return snapshot;
  }  return nullptr;
}

第三步，ResolveVMData 调用SearchMapping创建符号Mappping

static std::shared_ptr<const fml::Mapping> ResolveVMData(    const Settings& settings) {#if DART_SNAPSHOT_STATIC_LINK
  return std::make_unique<fml::NonOwnedMapping>(kDartVmSnapshotData, 0);#else   // DART_SNAPSHOT_STATIC_LINK
  return SearchMapping(
    settings.vm_snapshot_data,          // embedder_mapping_callback
    settings.vm_snapshot_data_path,     // file_path
    settings.application_library_path,  // native_library_path
    DartSnapshot::kVMDataSymbol,        // native_library_symbol_name
    false                               // is_executable
  );#endif  // DART_SNAPSHOT_STATIC_LINK}

第四步，SearchMapping 调用fml::NativeLibrary::Create读取文件

static std::shared_ptr<const fml::Mapping> SearchMapping(
    MappingCallback embedder_mapping_callback,    const std::string& file_path,    const std::vector<std::string>& native_library_path,    const char* native_library_symbol_name,    bool is_executable) {  // Ask the embedder. There is no fallback as we expect the embedders (via
  // their embedding APIs) to just specify the mappings directly.
  if (embedder_mapping_callback) {    return embedder_mapping_callback();
  }  // Attempt to open file at path specified.
  if (file_path.size() > 0) {    if (auto file_mapping = GetFileMapping(file_path, is_executable)) {      return file_mapping;
    }
  }  // Look in application specified native library if specified.
  for (const std::string& path : native_library_path) {    auto native_library = fml::NativeLibrary::Create(path.c_str());//通过NativeLibrary加载文件
    auto symbol_mapping = std::make_unique<const fml::SymbolMapping>(
        native_library, native_library_symbol_name);    if (symbol_mapping->GetMapping() != nullptr) {      return symbol_mapping;
    }
  }  // Look inside the currently loaded process.
  {    auto loaded_process = fml::NativeLibrary::CreateForCurrentProcess();    auto symbol_mapping = std::make_unique<const fml::SymbolMapping>(
        loaded_process, native_library_symbol_name);    if (symbol_mapping->GetMapping() != nullptr) {      return symbol_mapping;
    }
  }  return nullptr;
}

第五步，NativeLibrary 最终调用dlopen去获取符号内容

NativeLibrary::NativeLibrary(const char* path) {
  ::dlerror();
  handle_ = ::dlopen(path, RTLD_NOW);//实际上调用dlopen
  if (handle_ == nullptr) {
    FML_DLOG(ERROR) << "Could not open library '" << path << "' due to error '"
                    << ::dlerror() << "'.";
  }
}

最终结论：我们可以移除kDartIsolateSnapshotData、kDarVmSnapshotData放到App外部，然后通过修改ResolveVMData，在引擎初始化时，从外部文件中重建Dart虚拟机所需要的数据结构，达到缩减App体积的目的。

3.4 FLutter.framework/Flutter

Flutter.framework/Flutter为引擎产物，其大小是固定的，但是初始占比比较大。这部分能优化的空间很小，主要是通过裁剪引擎不需要的功能，减少体积。编译引擎时可以选择性编译skia和boringssl，收益大概只有几百K。

除此之外可以对Flutter的符号进行分离。

#将flutter的符号，存为.dSYM文件
xcrun dsymutil -o $frameworkpath/Flutter.framework.dSYM  $frameworkpath/Release/Flutter.framework/flutter
#然后strip掉符号
xcrun strip -x -S $frameworkpath/Release/Flutter.framework/flutter

4、实践

前面都是在分析如何去iOS产物进行优化，接下来我们真正开始动手减包之路。

4.1 引擎编译配置

这部分涉及到的引擎编译部分知识，可自行参考相关资料，这里不再进行细节描述。以下列出相关编译脚本，和需要注意的事项。

在引擎源码目录下创建以下脚本

debug模式

  #!/bin/bash
  #ios_debug.sh
  #需要在环境变量中添加export FLUTTER_SDK=“your flutter install path”
  #引擎编译很耗内存，电脑性能吃紧的把10调低一点

  #构建iOS debug 引擎
  Echo “当前目录在：”
  pwd

  # 构建模拟器
  Echo "构建iOS debug 模拟器"
  ./flutter/tools/gn --unoptimized --runtime-mode debug --simulator
  ./flutter/tools/gn --unoptimized --ios --runtime-mode debug --simulator
  ninja -C out/host_debug_sim_unopt -j 10
  ninja -C out/ios_debug_sim_unopt -j 10
  Echo  ""

  # 构建armv7
  Echo "构建iOS debug arm"
  ./flutter/tools/gn --unoptimized --runtime-mode debug --ios-cpu arm
  ./flutter/tools/gn --unoptimized --ios --runtime-mode debug --ios-cpu arm
  ninja -C out/host_debug_unopt_arm -j 10
  ninja -C out/ios_debug_unopt_arm -j 10
  Echo  ""


  # 构建arm64
  Echo "构建iOS debug arm64"
  ./flutter/tools/gn --unoptimized --runtime-mode debug --ios-cpu arm64
  ./flutter/tools/gn --unoptimized --ios --runtime-mode debug --ios-cpu arm64
  ninja -C out/host_debug_unopt -j 10
  ninja -C out/ios_debug_unopt -j 10
  Echo  ""


  #归档
  Echo  "开始归档"
  rm -rf tmp/*
  flutter_lipo=./arch_file
  buidmode=ios


  cp -rf out/ios_debug_unopt/Flutter.framework tmp/

  lipo -create -output tmp/Flutter.framework/Flutter \
  out/ios_debug_sim_unopt/Flutter.framework/Flutter \
  out/ios_debug_unopt/Flutter.framework/Flutter \
  out/ios_debug_unopt_arm/Flutter.framework/Flutter

  #cd tmp
  #zip -r Flutter.framework.zip Flutter.framework
  #cd ..
  #mkdir -p "${FLUTTER_SDK}"/bin/cache/artifacts/engine/"${buidmode}"
  #cp -f tmp/Flutter.framework.zip "${FLUTTER_SDK}"/bin/cache/artifacts/engine/"${buidmode}"/

  cp -rf tmp/Flutter.framework "${FLUTTER_SDK}"/bin/cache/artifacts/engine/"${buidmode}"/

  #copy gen_snapshot
  cp -f out/ios_debug_unopt/clang_x64/gen_snapshot "${FLUTTER_SDK}"/bin/cache/artifacts/engine/"${buidmode}"/gen_snapshot_arm64
  cp -f out/ios_debug_unopt_arm/clang_x64/gen_snapshot  "${FLUTTER_SDK}"/bin/cache/artifacts/engine/"${buidmode}"/gen_snapshot_armv7

  Echo  "归档完毕"

Profile模式

  #!/bin/bash
  #ios_profile.sh
  #需要在环境变量中添加export FLUTTER_SDK=“your flutter install path”


  #构建iOS debug 引擎

  Echo “当前目录在：”
  pwd


  # 构建arm
  Echo "构建iOS profile arm"
  #./flutter/tools/gn --runtime-mode profile --ios-cpu arm
  #ninja -C out/host_profile_arm -j 6

  ./flutter/tools/gn --ios --runtime-mode profile --ios-cpu arm
  ninja -C out/ios_profile_arm -j 6
  echo ""

  # 构建arm64
  Echo "构建iOS profile arm64"
  #./flutter/tools/gn --runtime-mode profile --ios-cpu arm64
  #ninja -C out/host_profile -j 6

  ./flutter/tools/gn --ios --runtime-mode profile --ios-cpu arm64
  ninja -C out/ios_profile -j 6
  echo ""

  Echo  "开始归档"
  rm -rf tmp/*
  flutter_lipo=./arch_file
  buidmode=ios-profile

  cp -rf out/ios_profile/Flutter.framework tmp/
  lipo -create -output tmp/Flutter.framework/Flutter \
  out/ios_profile/Flutter.framework/Flutter \
  out/ios_profile_arm/Flutter.framework/Flutter \
  "${flutter_lipo}"/flutter-profile-x86_64

  #cd tmp
  #zip -r Flutter.framework.zip Flutter.framework
  #cd ..
  #mkdir -p "${FLUTTER_SDK}"/bin/cache/artifacts/engine/"${buidmode}"
  #cp -f tmp/Flutter.framework.zip "${FLUTTER_SDK}"/bin/cache/artifacts/engine/"${buidmode}"/

  cp -rf tmp/Flutter.framework "${FLUTTER_SDK}"/bin/cache/artifacts/engine/"${buidmode}"/

  #copy gen_snapshot
  cp -f out/ios_profile/clang_x64/gen_snapshot "${FLUTTER_SDK}"/bin/cache/artifacts/engine/"${buidmode}"/gen_snapshot_arm64
  cp -f out/ios_profile_arm/clang_x64/gen_snapshot  "${FLUTTER_SDK}"/bin/cache/artifacts/engine/"${buidmode}"/gen_snapshot_armv7

Release模式

  #!/bin/bash
  #ios_release.sh
  #需要在环境变量中添加export FLUTTER_SDK=“your flutter install path”
  #构建iOS Release 引擎
  Echo “当前目录在：”
  pwd

  # 构建arm
  Echo "构建iOS release armv7"
  #./flutter/tools/gn --runtime-mode release --ios-cpu arm
  #ninja -C out/host_release_arm -j 6
  ./flutter/tools/gn --ios --runtime-mode release --ios-cpu arm
  ninja -C out/ios_release_arm -j 6
  echo ""

  # 构建arm64
  Echo "构建iOS release arm64"
  #./flutter/tools/gn --runtime-mode release --ios-cpu arm64
  #ninja -C out/host_release -j 6
  ./flutter/tools/gn --ios --runtime-mode release --ios-cpu arm64
  ninja -C out/ios_release -j 6
  echo ""

  Echo "合并归档"
  rm -rf tmp/*
  flutter_lipo=./arch_file
  buidmode=ios-release

  cp -rf out/ios_release/Flutter.framework tmp/

  lipo -create -output tmp/Flutter.framework/Flutter \
  out/ios_release/Flutter.framework/Flutter \
  out/ios_release_arm/Flutter.framework/Flutter \
  "${flutter_lipo}"/flutter-release-x86_64

  #cd tmp
  #zip -r Flutter.framework.zip Flutter.framework
  #cd ..
  #mkdir -p "${FLUTTER_SDK}"/bin/cache/artifacts/engine/"${buidmode}"
  #cp -f tmp/Flutter.framework.zip "${FLUTTER_SDK}"/bin/cache/artifacts/engine/"${buidmode}"/

  cp -rf tmp/Flutter.framework "${FLUTTER_SDK}"/bin/cache/artifacts/engine/"${buidmode}"/

  #copy gen_snapshot
  cp -f out/ios_release/clang_x64/gen_snapshot "${FLUTTER_SDK}"/bin/cache/artifacts/engine/"${buidmode}"/gen_snapshot_arm64
  cp -f out/ios_release_arm/clang_x64/gen_snapshot  "${FLUTTER_SDK}"/bin/cache/artifacts/engine/"${buidmode}"/gen_snapshot_armv7
  Echo  "归档完毕"

其他详细编译参数，使用./flutter/tools/gn —help命令查看。

另外flutter引擎默认不支持bitcode，如果需要支持，需要在编译脚本后添加—bitcode

#例如，添加--bitcode，最终的引擎产物将包含bitcode
./flutter/tools/gn --ios --runtime-mode release --ios-cpu arm --bitcode

4.2 xcode配置引擎调试环境

以Debug模式为例，找到引擎源码目录src/out/ios_debug_unopt下的products.xcodeproj。该工程为引擎源码对应的iOS工程
打开flutter工程下的./iOS/Runner.xcworkspace。该目录由Flutter SDK 中的flutter_tools自动生成，保存了运行flutter所需要的iOS宿主模板Runner工程
把products.xcodeproj拖入到Runner项目中，并在Generated.xcconfig中添加相关环境变量

#以使用ios_debug_unopt引擎为例，添加如下代码，即可调试引擎
FLUTTER_FRAMEWORK_DIR=/Users/wangtengyu/allenwork/engine/src/out/ios_debug_unopt
LOCAL_ENGINE=ios_debug_unopt
FLUTTER_ENGINE=/Users/wangtengyu/allenwork/engine/src

4.3 减包引擎源码修改

4.3.1 添加自定义配置，Settings

Settings是一个重要的配置结构体，位于src/flutter/common/settings.h中，这个公共头文件被大量使用。我们可以在settings.h添加我们自己的代码，来实现想要的功能。

为了自定义DartVMData加载路径，我们在settings结构体中添加了2个string成员用来保存vm和isolate数据文件路径。

struct Settings {
  Settings();

  Settings(const Settings& other);

  ~Settings();  //add by allentywang.  dart data path.
  std::string ios_vm_snapshot_data_path;  // vm data path
  std::string ios_isolate_snapshot_data_path; // isolate data path
  //end
  ...
  ...
}

4.3.2 修改Dart编译工具，gen_snapshot

Dart业务代码使用gen_snapshot工具编译到App中，在程序运行时通过引擎内部的虚拟机加载App中的Dart编译代码。下面从写入和读取这两个方面介绍，如何分离Dart编译产物的数据段。

重定向App数据段的写入

//文件：image_snapshot.cc//添加头文件#include "bin/file.h"#include <iostream>//添加写入函数//add by allen. 把stream写到本地文件void WriteTextToLocalFile(WriteStream* clustered_stream, bool vm){#if defined(TARGET_OS_MACOS_IOS)  //add by allentywang
    auto OpenFile = [](const char* filename){
        Syslog::Print("open file : %s\n", filename);
        bin::File* file = bin::File::Open(NULL, filename, bin::File::kWriteTruncate);        if (file == NULL) {
          Syslog::PrintErr("Error: Unable to write file: %s\n", filename);
          Dart_ExitScope();
          Dart_ShutdownIsolate();          exit(255);
        }        return file;
    };    auto StreamingWriteCallback = [](void* callback_data,                                     const uint8_t* buffer,
                                     intptr_t size) {
        bin::File* file = reinterpret_cast < bin::File* >(callback_data);        if (!file->WriteFully(buffer, size)) {
          Syslog::PrintErr("Error: Unable to write snapshot file\n");
          Dart_ExitScope();
          Dart_ShutdownIsolate();          exit(255);
    }

    };#if defined(TARGET_ARCH_ARM64)
    printf("this is arm64\n");
    bin::File *file = OpenFile(vm ? "./SnapshotData/arm64/VmSnapshotData.S" : "./SnapshotData/arm64/IsolateSnapshotData.S");#else//#if defined(TARGET_ARCH_ARM)
    printf("this is armv7\n");
    bin::File *file = OpenFile(vm ? "./SnapshotData/armv7/VmSnapshotData.S" : "./SnapshotData/armv7/IsolateSnapshotData.S");#endif //end of TARGET_ARCH_ARM64
  bin::RefCntReleaseScope rs(file);
  StreamingWriteStream stream = StreamingWriteStream(512 * KB, StreamingWriteCallback, file);

  uword buffer = reinterpret_cast<uword>(clustered_stream->buffer());
  intptr_t length = clustered_stream->bytes_written();
  uword start = buffer;
  uword end = buffer + length;  auto const end_of_words =
      Utils::RoundDown(end, sizeof(compiler::target::uword));  for (auto cursor = reinterpret_cast< compiler::target::uword* >(start);
       cursor < reinterpret_cast< compiler::target::uword* >(end_of_words);
       cursor++) {    #if defined(TARGET_ARCH_IS_64_BIT)
    stream.Print(".quad 0x%0.16" Px "\n", *cursor);    #else
    stream.Print(".long 0x%0.8" Px "\n", *cursor);    #endif
  }  if (end != end_of_words) {    auto start_of_rest = reinterpret_cast< const uint8_t* >(end_of_words);
    stream.Print(".byte ");    for (auto cursor = start_of_rest;
         cursor < reinterpret_cast< const uint8_t* >(end); cursor++) {      if (cursor != start_of_rest) stream.Print(", ");
      stream.Print("0x%0.2" Px "", *cursor);
    }
    stream.Print("\n");
  }#endif //end of TARGET_OS_MACOS_IOS}

...//修改数据段写入
 void AssemblyImageWriter::WriteText(WriteStream* clustered_stream, bool vm) {
  ...#if defined(TARGET_OS_MACOS_IOS)
  WriteTextToLocalFile(clustered_stream, vm);//ios下写到外部文件中,WriteTextToLocalFile为上面我们自己的写入函数#else
  const char* data_symbol =
      vm ? "_kDartVmSnapshotData" : "_kDartIsolateSnapshotData";
  assembly_stream_.Print(".globl %s\n", data_symbol);
  Align(kMaxObjectAlignment);
  assembly_stream_.Print("%s:\n", data_symbol);
  uword buffer = reinterpret_cast<uword>(clustered_stream->buffer());
  intptr_t length = clustered_stream->bytes_written();
  WriteByteSequence(buffer, buffer + length);
  Syslog::Print("write file : %s\n", data_symbol);#endif //end of TARGET_OS_MACOS_IOS}

注意：要根据不同架构，分别保存VmSnapshotData.S和IsolateSnapshotData.S。架构区分宏定义为TARGET_ARCH_ARM

上面我们分离出了VmSnapshotData.S和IsolateSnapshotData.S，接下来需要将其编译成机器代码IsolateData.dat和VMData.dat

#以armv7的VmSnapshotData.S和IsolateSnapshotData.S为例，arm64类同
armv7=./SnapshotData/armv7
echo "编译数据段"
xcrun cc -arch armv7  -c $armv7/IsolateSnapshotData.S -o $armv7/HeadIsolateData.dat
xcrun cc -arch armv7  -c $armv7/VmSnapshotData.S -o $armv7/HeadVMData.dat
# 去掉多余头部
tail -c +313 $armv7/HeadIsolateData.dat > $armv7/IsolateData.dat
tail -c +313 $armv7/HeadVMData.dat > $armv7/VMData.dat

IsolateData.dat和VMData.dat为我们最终想要分离的数据段

自定义App数据段的重建

根据前面3.3.3节的分析，重建数据段，我们可以从ResolveVMData入手，下面是修改的代码。

//文件：dart_snapshot.cc//新增自定义重建函数//add by allentywang. create data Mappingstd::shared_ptr<const fml::Mapping> SetupMapping(const std::string &path) {  // Check if the path exists and it readable directly.

  auto fd = fml::OpenFile(path.c_str(), false, fml::FilePermission::kRead);  // Check the path relative to the current executable.
  if (!fd.is_valid()) {    auto directory = fml::paths::GetExecutableDirectoryPath();    if (!directory.first) {      return nullptr;
    }

    std::string path_relative_to_executable = fml::paths::JoinPaths({directory.second, path});

    fd = fml::OpenFile(path_relative_to_executable.c_str(),                       false,
                       fml::FilePermission::kRead);
  }  if (!fd.is_valid()) {    return nullptr;
  }

  std::initializer_list<fml::FileMapping::Protection> protection = {fml::FileMapping::Protection::kRead};  auto file_mapping = std::make_unique<fml::FileMapping>(fd, std::move(protection));  if (file_mapping->GetSize() != 0) {    return file_mapping;
  }  return nullptr;
}//end//修改原来的ResolveVMData和ResolveIsolateData函数。static std::shared_ptr<const fml::Mapping> ResolveVMData(    const Settings& settings) {#if DART_SNAPSHOT_STATIC_LINK
  return std::make_unique<fml::NonOwnedMapping>(kDartVmSnapshotData, 0);#else   // DART_SNAPSHOT_STATIC_LINK#if OS_IOS //add by allentywang
    if (settings.ios_vm_snapshot_data_path.empty()) {//ios
        printf("ResolveVMData from local\n");          return SearchMapping(
              settings.vm_snapshot_data,          // embedder_mapping_callback
              settings.vm_snapshot_data_path,     // file_path
              settings.application_library_path,  // native_library_path
              DartSnapshot::kVMDataSymbol,        // native_library_symbol_name
              false                               // is_executable
              );
    } else {        printf("ResolveVMData from settings.ios_vm_snapshot_data_path\n");      return SetupMapping(settings.ios_vm_snapshot_data_path);
    }#else
    return SearchMapping(
        settings.vm_snapshot_data,          // embedder_mapping_callback
        settings.vm_snapshot_data_path,     // file_path
        settings.application_library_path,  // native_library_path
        DartSnapshot::kVMDataSymbol,        // native_library_symbol_name
        false                               // is_executable
    );#endif#endif  // DART_SNAPSHOT_STATIC_LINK}static std::shared_ptr<const fml::Mapping> ResolveIsolateData(    const Settings& settings) {#if DART_SNAPSHOT_STATIC_LINK
  return std::make_unique<fml::NonOwnedMapping>(kDartIsolateSnapshotData, 0);#else   // DART_SNAPSHOT_STATIC_LINK#if OS_IOSif (settings.ios_isolate_snapshot_data_path.empty()) {    printf("ResolveVMData from local\n");  return SearchMapping(
    settings.isolate_snapshot_data,       // embedder_mapping_callback
    settings.isolate_snapshot_data_path,  // file_path
    settings.application_library_path,    // native_library_path
    DartSnapshot::kIsolateDataSymbol,     // native_library_symbol_name
    false                                 // is_executable
  );
} else {    printf("ResolveVMData from settings.ios_isolate_snapshot_data_path\n");  return SetupMapping(settings.ios_isolate_snapshot_data_path);
}#elsereturn SearchMapping(
    settings.isolate_snapshot_data,       // embedder_mapping_callback
    settings.isolate_snapshot_data_path,  // file_path
    settings.application_library_path,    // native_library_path
    DartSnapshot::kIsolateDataSymbol,     // native_library_symbol_name
    false                                 // is_executable);#endif#endif  // DART_SNAPSHOT_STATIC_LINK}

要使代码生效，我们还需要指定正确的settings.ios_isolate_snapshot_data_path路径，这个路径变量就是之前在settings结构体中新增的配置。接下来路径指定需要在DefaultSettingsForProcess函数中完成。

4.3.4 修改引擎配置的加载，DefaultSettingsForProcess

引擎初始化时，会调用DefaultSettingsForProcess函数，来初始化Settings配置。由于我们删除了framework内部的flutter_assets和icudtl.dat，分离了App数据段IsolateData.dat和VMData.dat。接下来我们需要对这4个产物路径进行重新设置，让引擎能够顺利初始化。代码如下

// 文件：FlutterDartProject.mm//新增函数initSettingvoid initSettings(flutter::Settings &settings){    printf("%s %s:%d\n", __FUNCTION__, __FILE__, __LINE__);    // 检查资源目录，如果找到对应资源，则保存路径到Settings#if (FLUTTER_RUNTIME_MODE != FLUTTER_RUNTIME_MODE_DEBUG)//debug不做任何处理，只针对profile和release
    printf("settings.icu_data_path = %s\n", settings.icu_data_path.c_str());    // flutter_assets
    printf("settings.assets_path = %s\n", settings.assets_path.c_str());#if 1 //defined(TARGET_ARCH_ARM64) 架构区分在中有相关宏定义
    // VMData.dat
    NSString *vmDataPath = [[NSBundle mainBundle] pathForResource:@"VMData" ofType:@"dat"];    if (vmDataPath == nil)
    {
        NSLog(@"can not found VMData.dat from resource, download from server?");
    }else{
        NSLog(@"check vmDataPath from resource: %@", vmDataPath);
        settings.ios_vm_snapshot_data_path = vmDataPath.UTF8String;
    }    // IsolateData.dat
    NSString *isolateDataPath = [[NSBundle mainBundle] pathForResource:@"IsolateData" ofType:@"dat"];    if (vmDataPath == nil)
    {
        NSLog(@"can not found IsolateData.dat from resource, download from server?");
    }else{
        NSLog(@"check isolateDataPath from resource: %@", isolateDataPath);
        settings.ios_isolate_snapshot_data_path = isolateDataPath.UTF8String;
    }#else
    printf("arch is armv7\n");#endif

    // icudtl.dat
    NSString *icudtlPath = [[NSBundle mainBundle] pathForResource:@"icudtl" ofType:@"dat"];    if (icudtlPath == nil)
    {
        NSLog(@"can not found icudtl.dat from resource, download from server?");
    }else{
        NSLog(@"check icudtl from resource: %@", icudtlPath);
        settings.icu_data_path = icudtlPath.UTF8String;
    }    // flutter_assets//    settings.assets_path = assetsPath.UTF8String;


    // end#endif}

... static flutter::Settings DefaultSettingsForProcess(NSBundle* bundle = nil) {  auto command_line = flutter::CommandLineFromNSProcessInfo();  // Precedence:
  // 1. Settings from the specified NSBundle.
  // 2. Settings passed explicitly via command-line arguments.
  // 3. Settings from the NSBundle with the default bundle ID.
  // 4. Settings from the main NSBundle and default values.

  NSBundle* mainBundle = [NSBundle mainBundle];
  NSBundle* engineBundle = [NSBundle bundleForClass:[FlutterViewController class]];  bool hasExplicitBundle = bundle != nil;  if (bundle == nil) {
    bundle = [NSBundle bundleWithIdentifier:[FlutterDartProject defaultBundleIdentifier]];
  }  if (bundle == nil) {
    bundle = mainBundle;
  }  auto settings = flutter::SettingsFromCommandLine(command_line);
  initSettings(settings);//获取到settings后，调用initSettings函数对setting进行修改，将4个路径改到实际对应的位置

  ...
}

完成后，重新编译引擎，替换引擎即可。

4.4 Flutter侧产物打包脚本

注意脚本中使用CloseBitcode函数关闭了bitcode，让flutter build ios-framework命令生成的framework不带bitcode。因为xcode工程默认是开启bitcode的，而前面我们的引擎产物没有加--bitcode参数，不关闭xcode的bitcode选项是无法编译成功的。

脚本过程说明就省略了，这里直接贴上代码

#ios_build_reduce.sh
#检查路径是否存在，不存在就退出脚本
AssertExists() {
  if [[ ! -e "$1" ]]; then
    if [[ -h "$1" ]]; then
      echo "The path $1 is a symlink to a path that does not exist"
    else
      echo "The path $1 does not exist"
    fi
    exit -1
  fi
  return 0
}

#要关闭bitcode功能，必须修改.ios下的pods配置
CloseBitcode () {
  str1="#this is generate from ios_build_reduce.sh to set 'ENABLE_BITCODE' = 'NO' for all targets of pods build setting \n\
  post_install do |installer|\n\
    installer.pods_project.targets.each do |target|\n\
      target.build_configurations.each do |config|\n\
        config.build_settings['ENABLE_BITCODE'] = 'NO'\n\
      end\n\
    end\n\
  end"
  echo "$str1" >> $1
}

#运行脚本，开启VERBOSE_SCRIPT_LOGGING时提供命令记录
RunCommand() {
  if [[ -n "$VERBOSE_SCRIPT_LOGGING" ]]; then
    echo "♦ $*" #show script
  fi
  "$@"  #run
  return $? #return last result
}

#VERBOSE_SCRIPT_LOGGING="开启脚本记录"
echo "iOS Flutter 产物集成 start！！！！！！"
cd ..

#bitcode需要手动配置.ios下的Pods工程，不能clean
flutter clean
echo "flutter clean done !!!"
flutter pub get
echo "flutter pub get done !!!"

#初始化目录
armv7=./SnapshotData/armv7
arm64=./SnapshotData/arm64
mkdir -p $armv7
mkdir -p $arm64
frameworkpath='build/ios/framework'
#rm       $frameworkpath/FlutterPackage.zip
#rm -rf   $frameworkpath/flutter_reduce
mkdir -p $frameworkpath/flutter_reduce/arm64
mkdir -p $frameworkpath/flutter_reduce/armv7

#修改podfile
podfile='./.ios/Podfile'
AssertExists "$podfile"
CloseBitcode "$podfile"

#编译framework
flutter build ios-framework
echo "flutter-framework 打包成功 !!!"

debugpath='build/ios/framework/Debug'
releasepath='build/ios/framework/Release'
libpath='../../ios/LocalLib/Flutter'

rm -rf "$libpath/Debug"
rm -rf "$libpath/Release"

echo "编译数据段"
xcrun cc -arch armv7  -c $armv7/IsolateSnapshotData.S -o $armv7/HeadIsolateData.dat
xcrun cc -arch armv7  -c $armv7/VmSnapshotData.S -o $armv7/HeadVMData.dat
# 去掉多余头部
tail -c +313 $armv7/HeadIsolateData.dat > $armv7/IsolateData.dat
tail -c +313 $armv7/HeadVMData.dat > $armv7/VMData.dat

xcrun cc -arch arm64  -c $arm64/IsolateSnapshotData.S -o $arm64/HeadIsolateData.dat
xcrun cc -arch arm64  -c $arm64/VmSnapshotData.S -o $arm64/HeadVMData.dat
# 去掉多余头部
tail -c +313 $arm64/HeadIsolateData.dat > $arm64/IsolateData.dat
tail -c +313 $arm64/HeadVMData.dat > $arm64/VMData.dat

cp -rf $armv7/IsolateData.dat  $frameworkpath/flutter_reduce/armv7
cp -rf $armv7/VMData.dat  $frameworkpath/flutter_reduce/armv7

cp -rf $arm64/IsolateData.dat  $frameworkpath/flutter_reduce/arm64
cp -rf $arm64/VMData.dat  $frameworkpath/flutter_reduce/arm64

echo "移除flutter_asserts"
mv $releasepath/App.framework/flutter_assets $frameworkpath/flutter_reduce/
echo "移除icudtl.dat"
mv $releasepath/Flutter.framework/icudtl.dat $frameworkpath/flutter_reduce/

echo "压缩分离产物FlutterPackage.zip"
zip -q -r $frameworkpath/FlutterPackage.zip $frameworkpath/flutter_reduce

echo  "framework优化体积："
printf "未压缩总量：%s----%s\n" `du  -sh $frameworkpath/flutter_reduce`
printf "         \t├─%s----%s\n" `du  -sh $frameworkpath/flutter_reduce/armv7`
printf "         \t├─%s----%s\n" `du  -sh $frameworkpath/flutter_reduce/arm64`
printf "         \t├─%s----%s\n" `du  -sh $frameworkpath/flutter_reduce/icudtl.dat`
printf "         \t├─%s----%s\n" `du  -sh $frameworkpath/flutter_reduce/flutter_assets`
RunCommand printf "压缩后总量：%s----%s\n" `du  -sh $frameworkpath/FlutterPackage.zip`
printf "\n"
printf "flutter(armv7、arm64) strip之前：%s----%s\n" `du  -sh $frameworkpath/Release/Flutter.framework/flutter`
AssertExists "${frameworkpath}"
xcrun dsymutil -o $frameworkpath/Flutter.framework.dSYM  $frameworkpath/Release/Flutter.framework/flutter
echo "分离符号：${frameworkpath}/Flutter.framework.dSYM"
xcrun strip -x -S $frameworkpath/Release/Flutter.framework/flutter
printf "flutter(armv7、arm64) strip之后：%s----%s\n" `du  -sh $frameworkpath/Release/Flutter.framework/flutter`

4.5 iOS侧Pod集成

iOS宿主工程Podfile添加如下代码。使用FlutterDebug.podspec和FlutterRelease.podspec来引入flutter的framework产物

# 是否源码集成, 0:否 / 1:是
IsFlutterSourceCode = 0

# 集成 FlutterModule
def FlutterModuleIntegration
    if IsFlutterSourceCode == 1
        # 源码集成（官方方案）
        flutter_application_path = '../native_modules/mlive/'
        load File.join(flutter_application_path, '.ios', 'Flutter', 'podhelper.rb')
        install_all_flutter_pods(flutter_application_path)
    else
        # 区分debug/release场景使用不同的调试包
        pod 'FlutterDebug' ,:configurations => ['Debug'] ,:path => 'LocalLib/Flutter'
        pod 'FlutterRelease' ,:configurations => ['Release', 'AppStore', 'iAP'] ,:path => 'LocalLib/Flutter'
    end
end

4.6 最终效果

名称	优化方式	减包收益
icudtl.dat	下载	884k
flutter_assets	下载	2.1M
App数据段	下载	单架构平均2.8M
flutter	strip调试符号	单架构平均6M
framework总计收益	—-	11.7M
size Report最终收益	—-	—-

总结：我们通过删除不必要的文件、移走部分文件改为下发、去掉Flutter的符号文件、引擎大小优化等措施，使iOS接入Flutter的体积成本降到10M。

5、参考文章

字节跳动- 【如何缩减接近 50% 的 Flutter 包体积】：https://juejin.im/post/5de8a32c51882512664affa4

QQ音乐招聘Android/ios客户端开发，点击左下方“查看原文”投递简历～

也可将简历发送至邮箱：tmezp@tencent.com

继续滑动看下一个