Flutter Principle
Flutter vs React Native 性能对比
Flutter (written in Dart) -> Skia (written in C/C++) -> OpenGL/Metal -> GPU -> Frame Buffer
Flutter 利用 Skia 绘图引擎，直接通过 CPU、GPU 进行绘制，不需要依赖任何原生的控件。Skia 引擎会将 Dart 视图结构数据加工成 GPU 数据，交由 OpenGL 最终提供给 GPU 渲染。
React Native(JSX) -> JavaScript (JS Bundle) -> JavaScriptCore -> Bridge -> iOS (OC/Swift) -> CoreAnimation -> OpenGL/Metal -> GPU -> Frame Buffer
通过 JavaScript 虚拟机扩展调用系统组件，由 Android 和 iOS 进行组件的渲染。
Flutter 是重写了一整套包括底层渲染逻辑和上层开发语言的完整解决方案。Dart 是少数同时支持 JIT（Just In Time，即时编译）和 AOT（Ahead of Time，运行前编译）的语言。可以保证视图渲染在 Android 和 iOS 上的高度一致性；在代码执行效率和渲染性能上也可以媲美原生 App。
在开发期使用 JIT，支持有状态的热重载，开发效率极高；而发布期使用 AOT，代码执行更高效，代码性能和用户体验也更卓越。
Flutter apps run in a VM that offers stateful hot reload of changes without needing a full recompile. For release, Flutter apps are compiled directly to machine code (Intel x64 or ARM instructions).
Dart 避免了抢占式调度和共享内存，可以在没有锁的情况下进行对象分配和垃圾回收，在性能方面表现相当不错。
Declarative UI
Frameworks from Win32 to web to Android and iOS typically use an imperative 命令式 style of UI programming. You manually construct a full-functioned UI entity, such as a UIView or equivalent, and later mutate it using methods and setters when the UI changes. （控制器持有 UIView 的实例，并调用实例方法）
In order to lighten the burden on developers from having to program how to transition between various UI states, Flutter, by contrast, lets the developer describe the current UI state and leaves the transitioning to the framework.
Here, rather than mutating an UIView instance when the UI changes, Flutter constructs new Widget instances. The framework manages many of the responsibilities of a traditional UI object (such as maintaining the state of the layout) behind the scenes with RenderObjects. RenderObjects persist between frames and Flutter’s lightweight Widgets tell the framework to mutate the RenderObjects between states. The Flutter framework handles the rest.
Flutter architectural overview
Most developers will interact with Flutter via the Flutter Framework. The Flutter Engine is a portable runtime for hosting Flutter applications. It implements Flutter's core libraries, including animation and graphics, file and network I/O, accessibility support, plugin architecture, and a Dart runtime and compile toolchain.
三棵树
Widget 树：开发者写的 dart 代码，对图形化界面的描述/蓝图。
Element 树：Element 与 Widget 一一对应，Flutter 引擎维护管理，并在 setState 时对比、刷新。
Render 树：已布局完毕，所有元素的位置、宽高，图片、纹理等信息都已确定，可以交给 Skia 引擎绘图的树。
When your app creates and displays a scene, the UI thread creates a layer tree, a lightweight object containing device-agnostic painting commands, and sends the layer tree to the raster thread to be rendered on the device.
PlatformView 到了 iOS 平台上是个 UIView，如果 Widget 树里有 PlatformView，那么它会以 UIView 的 subview 形式存在。
四个线程
- UI thread: All of your Dart code runs on the UI thread. The UI thread executes Dart code in the Dart VM.
- Platform thread: The platform’s main thread. Plugin code runs here.
- Raster thread: Skia, the graphics library, runs on this thread. 接受 UI thread 创建的 layer tree，输出光栅化图像，然后将光栅化的数据交给 GPU 绘制。
- I/O thread: Performs expensive tasks (mostly I/O) that would otherwise block either the UI or raster threads.
120 Hz
Flutter 120hz 高刷新率在 Android 和 iOS 上的调研总结 - 掘金
Support Variable Refresh Rate Displays (PUBLICLY SHARED) - Google Docs
Garbage Collector
Using the Memory view | Flutter
官方文档里引用的文章：Flutter: Don’t Fear the Garbage Collector | by Matt Sullivan | Flutter | Medium
Dart 垃圾回收：分代的，两步：新生代 Young Space Scavenger、老生代 Parallel Marking and Concurrent Sweeping
新生代（Cheney 算法）：
- 针对生命周期很短的对象，例如
StatelessWidget。阻塞的，但比第二代算法快得多，结合空闲时调度，消除可感知的卡顿。 - 新对象被分配在连续的内存块（与 malloc 不同）
- 新对象被分配的空间分为两个“半空间”，新对象被分配在活跃的半空间，当它满了，就把仍然存活的对象拷贝到不活跃的半空间
Bump pointer works by allocating memory from a contiguous block of memory and keeping track of the next available address to allocate. When a new object needs to be allocated, the garbage collector simply moves the pointer to the next available address and returns it. This is a simple and efficient method of allocation, but it can lead to fragmentation of the memory block.
malloc does not require a contiguous block of memory. Instead, it searches for a free block of memory of the requested size and returns a pointer to that block. This can lead to more efficient use of memory and less fragmentation, but it also requires more bookkeeping overhead.
老生代：
- 那些在上一阶段没有被回收的，会成为老生代
- 两步：遍历对象引用图，标记存活的对象；扫描全部内存，清楚未被标记的对象
- 比较慢，但是发生频率很低