React16源码: React中的renderRoot的源码实现

renderRoot

1 ）概述

• renderRoot 是一个非常复杂的方法
• 这个方法里处理很多各种各样的逻辑, 它主要的工作内容是什么？
• A. 它调用 workLoop 进行循环单元更新
  • 遍历整个 Fiber Tree，把每一个组件或者 dom 节点对应的
  • Fiber 节点拿出来单一的进行更新，这是一个循环的操作
  • 把整棵 Fiber Tree 都遍历一遍，这就是 workLoop
• B. 捕获错误并进行处理
  • 在进行每一个单元更新的时候，这个遍历逻辑的时候，有可能会出现一些错误
  • 有些是可预期的，比如说是 Suspense 的功能, throw 一个 Promise 对象
  • 这个时候，是我们要特殊的对它进行处理的
  • 有一些是不可预期的，比如说, render 里面报了一个错误
  • 我们也要对它进行一些处理，要告诉我们的react这个地方
  • 我们出现了一个错误，在react16之后，有了 Error Boundary
  • 可以在组件内捕获渲染的错误
• C. 在走完流程之后要进行善后
  • 因为流程走完之后会有各种不同的情况
  • 比如说有错误的情况，比如说有任务被挂起的情况，也就是Suspense的情况
  • 这些任务，都要按照特定的逻辑给它进行一些处理
• 这就是 renderRoot 这个方法，它的主要的核心工作内容

2 ）流程图

• 进入 renderRoot, 它里面有一个很核心的循环 do while workLoop
• 这个while循环就是调用 workLoop 对整棵树，它每个节点进行一个遍历，并且拿出来单独进行更新
• 因为每个 Fiber节点上，如果有更新的话，它会记入 updateQueen
• 我们通过 updateQueen 上是否有内容来判断它是否要进行更新
• 以及可以计算出它的新的 state，得到最新的 children，拿到所有最新的节点
• 在 workLoop 的过程当中，它在做什么呢？
• nextUnitOfWork 就是每一个节点在遍历的过程当中，它自己更新完之后，它会返回它的第一个child
• 它的第一个child 就是作为 nextUnitOfWork，因为我们执行了一个节点的更新之后，我们需要返回
• 返回之后，我们要判断一些逻辑
• 比如，对于异步的操作，每个节点更新完之后都要判断 !shouldYield()
• 判断我们现在的时间片是否还有？如果还有的话，再继续，如果没有的话，就要跳出了
• 接下去就会执行 performUnitOfWork
• 之后，执行 beginWork, completeUnitOfWork 这些，当然中间会判断是否有 next
• next就是我在更新完一个结点之后，它是否还有下一个节点需要更新
• 如果有next的情况，我们就返回，然后去判断这个逻辑是否还有
• 这就是整个对整个 fiber tree 每个节点的遍历更新
• 在这个更新过程当中，如果有任何catch，就是捕获到异常
• 那么首先会进行一系列的判断，然后对它执行 throwException 或者是 onUncaughtError
• 它们对应的逻辑会不一样, 如果这个节点它是可处理的错误
• 我会直接对它进行 completeUnitOfWork() 因为更新到这个节点之后，它抛出错误了
• 说明我们这个节点下面的所有子节点都不需要再更新了
• 执行完成之后，我们就会调用 continue，对于这个 do while 循环，它又继续调用 workLoop
• 也就是说我们把一棵子数的错误处理完之后，它还可以继续对别的子树进行更新
• 整体更新完之后，就会 break，之后会有各种不同的情况，比如说有致命的错误等
• 它们都会调用不同的逻辑进行一个处理，这里 nextRenderDidError 是一个可处理的错误
• 比如说, 是可以被捕获的错误，有组件能捕获它，而 nextLatestAbsoluteTimeoutMs 是抛出promise的错误
• 它是一个被挂起的一个任务，对应要执行一个被被挂起的操作, 最后如果上面的情况都没有出现
• 直接 onComplete，之后就可以在root节点上 set finishedwork，这样, 就可以对它整体进行一个更新
• 就可以执行 completeRoot，就可以把 fiber树 变成我们真正的dom 树去更新整个页面
• 这就是整个 renderRoot 它的一个逻辑，简单总结
  • 先正常的执行每个单元的更新
  • 然后捕获到任何错误进行一定的处理
  • 最终把整个树遍历完之后根据不同的情况再进行一个处理

3 ）源码

定位到 packages/react-reconciler/src/ReactFiberScheduler.js

先看 renderRoot 的代码

function renderRoot(
  root: FiberRoot,
  isYieldy: boolean,
  isExpired: boolean,
): void {
  invariant(
    !isWorking,
    'renderRoot was called recursively. This error is likely caused ' +
      'by a bug in React. Please file an issue.',
  );
  isWorking = true;
  ReactCurrentOwner.currentDispatcher = Dispatcher;

  const expirationTime = root.nextExpirationTimeToWorkOn;

  // Check if we're starting from a fresh stack, or if we're resuming from
  // previously yielded work.
  // 刚进来的时候，进行这个判断
  // nextRoot 和 nextRenderExpirationTime 对应着接下来要渲染的节点和对应的过期时间 这是两个公共变量
  // 在这种情况下，说明调用这个方法的时候，接收到的参数和之前的不一样，可能就是之前的异步任务被新进来的高优先级的任务给打断了
  if (
    expirationTime !== nextRenderExpirationTime ||
    root !== nextRoot ||
    nextUnitOfWork === null
  ) {
    // Reset the stack and start working from the root.
    resetStack();
    nextRoot = root;
    nextRenderExpirationTime = expirationTime;
    // nextUnitOfWork 来自于 createWorkInProgress
    // 就是把当前的应用的状态对应的Fiber节点，拷贝了一份叫做 workInProgress 的对象
    // 因为我们不能直接在当前对象的Fiber节点上操作，它会影响我们目前的dom节点展示的样子
    // 所以要复制一份拷贝，对拷贝进行操作，workInProgress 和 current 之间会有一个转换的关系
    // 在renderRoot开始之后，我们真正操作的节点都是 workInProgress，没有直接在 current 上操作
    nextUnitOfWork = createWorkInProgress(
      nextRoot.current,
      null,
      nextRenderExpirationTime,
    );
    // 这种和不同expirationTime会有关系
    root.pendingCommitExpirationTime = NoWork;

    if (enableSchedulerTracing) {
      // Determine which interactions this batch of work currently includes,
      // So that we can accurately attribute time spent working on it,
      // And so that cascading work triggered during the render phase will be associated with it.
      const interactions: Set<Interaction> = new Set();
      root.pendingInteractionMap.forEach(
        (scheduledInteractions, scheduledExpirationTime) => {
          if (scheduledExpirationTime <= expirationTime) {
            scheduledInteractions.forEach(interaction =>
              interactions.add(interaction),
            );
          }
        },
      );

      // Store the current set of interactions on the FiberRoot for a few reasons:
      // We can re-use it in hot functions like renderRoot() without having to recalculate it.
      // We will also use it in commitWork() to pass to any Profiler onRender() hooks.
      // This also provides DevTools with a way to access it when the onCommitRoot() hook is called.
      root.memoizedInteractions = interactions;

      if (interactions.size > 0) {
        const subscriber = __subscriberRef.current;
        if (subscriber !== null) {
          const threadID = computeThreadID(
            expirationTime,
            root.interactionThreadID,
          );
          try {
            subscriber.onWorkStarted(interactions, threadID);
          } catch (error) {
            // Work thrown by an interaction tracing subscriber should be rethrown,
            // But only once it's safe (to avoid leaveing the scheduler in an invalid state).
            // Store the error for now and we'll re-throw in finishRendering().
            if (!hasUnhandledError) {
              hasUnhandledError = true;
              unhandledError = error;
            }
          }
        }
      }
    }
  }

  let prevInteractions: Set<Interaction> = (null: any);
  if (enableSchedulerTracing) {
    // We're about to start new traced work.
    // Restore pending interactions so cascading work triggered during the render phase will be accounted for.
    prevInteractions = __interactionsRef.current;
    __interactionsRef.current = root.memoizedInteractions;
  }

  let didFatal = false;

  startWorkLoopTimer(nextUnitOfWork);

  // 上面初始化工作做完之后，就开始 workLoop
  // 如果有 catch 就会有一大段处理逻辑, 这里先跳过
  // 
  do {
    try {
      workLoop(isYieldy);
    } catch (thrownValue) {
      if (nextUnitOfWork === null) {
        // This is a fatal error.
        didFatal = true;
        onUncaughtError(thrownValue);
      } else {
        if (__DEV__) {
          // Reset global debug state
          // We assume this is defined in DEV
          (resetCurrentlyProcessingQueue: any)();
        }

        const failedUnitOfWork: Fiber = nextUnitOfWork;
        if (__DEV__ && replayFailedUnitOfWorkWithInvokeGuardedCallback) {
          replayUnitOfWork(failedUnitOfWork, thrownValue, isYieldy);
        }

        // TODO: we already know this isn't true in some cases.
        // At least this shows a nicer error message until we figure out the cause.
        // https://github.com/facebook/react/issues/12449#issuecomment-386727431
        invariant(
          nextUnitOfWork !== null,
          'Failed to replay rendering after an error. This ' +
            'is likely caused by a bug in React. Please file an issue ' +
            'with a reproducing case to help us find it.',
        );

        const sourceFiber: Fiber = nextUnitOfWork;
        let returnFiber = sourceFiber.return;
        if (returnFiber === null) {
          // This is the root. The root could capture its own errors. However,
          // we don't know if it errors before or after we pushed the host
          // context. This information is needed to avoid a stack mismatch.
          // Because we're not sure, treat this as a fatal error. We could track
          // which phase it fails in, but doesn't seem worth it. At least
          // for now.
          didFatal = true;
          onUncaughtError(thrownValue);
        } else {
          throwException(
            root,
            returnFiber,
            sourceFiber,
            thrownValue,
            nextRenderExpirationTime,
          );
          nextUnitOfWork = completeUnitOfWork(sourceFiber);
          continue;
        }
      }
    }
    break;
  } while (true);

  if (enableSchedulerTracing) {
    // Traced work is done for now; restore the previous interactions.
    __interactionsRef.current = prevInteractions;
  }

  // We're done performing work. Time to clean up.
  isWorking = false;
  ReactCurrentOwner.currentDispatcher = null;
  resetContextDependences();

  // 在处理完 workLoop 后这里会有各种不同的判断
  // Yield back to main thread.
  // 这里代表有致命的错误
  if (didFatal) {
    const didCompleteRoot = false;
    stopWorkLoopTimer(interruptedBy, didCompleteRoot);
    interruptedBy = null;
    // There was a fatal error.
    if (__DEV__) {
      resetStackAfterFatalErrorInDev();
    }
    // `nextRoot` points to the in-progress root. A non-null value indicates
    // that we're in the middle of an async render. Set it to null to indicate
    // there's no more work to be done in the current batch.
    nextRoot = null;
    onFatal(root);
    return;
  }

  // 正常流程走完，这个if一定会匹配
  // 因为已经跳出 workLoop 了，说明一定有 react没有意识到的错误，所以调用 onYield
  if (nextUnitOfWork !== null) {
    // There's still remaining async work in this tree, but we ran out of time
    // in the current frame. Yield back to the renderer. Unless we're
    // interrupted by a higher priority update, we'll continue later from where
    // we left off.
    const didCompleteRoot = false;
    stopWorkLoopTimer(interruptedBy, didCompleteRoot);
    interruptedBy = null;
    onYield(root);
    return;
  }

  // We completed the whole tree.
  const didCompleteRoot = true;
  stopWorkLoopTimer(interruptedBy, didCompleteRoot);
  const rootWorkInProgress = root.current.alternate;
  invariant(
    rootWorkInProgress !== null,
    'Finished root should have a work-in-progress. This error is likely ' +
      'caused by a bug in React. Please file an issue.',
  );

  // `nextRoot` points to the in-progress root. A non-null value indicates
  // that we're in the middle of an async render. Set it to null to indicate
  // there's no more work to be done in the current batch.
  nextRoot = null;
  interruptedBy = null;

  // 这里也是
  if (nextRenderDidError) {
    // There was an error
    if (hasLowerPriorityWork(root, expirationTime)) {
      // There's lower priority work. If so, it may have the effect of fixing
      // the exception that was just thrown. Exit without committing. This is
      // similar to a suspend, but without a timeout because we're not waiting
      // for a promise to resolve. React will restart at the lower
      // priority level.
      markSuspendedPriorityLevel(root, expirationTime);
      const suspendedExpirationTime = expirationTime;
      const rootExpirationTime = root.expirationTime;
      onSuspend(
        root,
        rootWorkInProgress,
        suspendedExpirationTime,
        rootExpirationTime,
        -1, // Indicates no timeout
      );
      return;
    } else if (
      // There's no lower priority work, but we're rendering asynchronously.
      // Synchronsouly attempt to render the same level one more time. This is
      // similar to a suspend, but without a timeout because we're not waiting
      // for a promise to resolve.
      !root.didError &&
      !isExpired
    ) {
      root.didError = true;
      const suspendedExpirationTime = (root.nextExpirationTimeToWorkOn = expirationTime);
      const rootExpirationTime = (root.expirationTime = Sync);
      onSuspend(
        root,
        rootWorkInProgress,
        suspendedExpirationTime,
        rootExpirationTime,
        -1, // Indicates no timeout
      );
      return;
    }
  }

  // 注意这里的错误
  if (!isExpired && nextLatestAbsoluteTimeoutMs !== -1) {
    // The tree was suspended.
    const suspendedExpirationTime = expirationTime;
    markSuspendedPriorityLevel(root, suspendedExpirationTime);

    // Find the earliest uncommitted expiration time in the tree, including
    // work that is suspended. The timeout threshold cannot be longer than
    // the overall expiration.
    const earliestExpirationTime = findEarliestOutstandingPriorityLevel(
      root,
      expirationTime,
    );
    const earliestExpirationTimeMs = expirationTimeToMs(earliestExpirationTime);
    if (earliestExpirationTimeMs < nextLatestAbsoluteTimeoutMs) {
      nextLatestAbsoluteTimeoutMs = earliestExpirationTimeMs;
    }

    // Subtract the current time from the absolute timeout to get the number
    // of milliseconds until the timeout. In other words, convert an absolute
    // timestamp to a relative time. This is the value that is passed
    // to `setTimeout`.
    const currentTimeMs = expirationTimeToMs(requestCurrentTime());
    let msUntilTimeout = nextLatestAbsoluteTimeoutMs - currentTimeMs;
    msUntilTimeout = msUntilTimeout < 0 ? 0 : msUntilTimeout;

    // TODO: Account for the Just Noticeable Difference

    const rootExpirationTime = root.expirationTime;
    onSuspend(
      root,
      rootWorkInProgress,
      suspendedExpirationTime,
      rootExpirationTime,
      msUntilTimeout,
    );
    return;
  }

  // Ready to commit.
  onComplete(root, rootWorkInProgress, expirationTime);
}

• renderRoot 代码会相对比较长，要把代码的区块进行一个区分
• 一些原版英文注释，和我添加的中文注释如上

现在来看下 workLoop 的源码

function workLoop(isYieldy) {
  if (!isYieldy) {
    // Flush work without yielding
    while (nextUnitOfWork !== null) {
      nextUnitOfWork = performUnitOfWork(nextUnitOfWork);
    }
  } else {
    // Flush asynchronous work until the deadline runs out of time.
    while (nextUnitOfWork !== null && !shouldYield()) {
      nextUnitOfWork = performUnitOfWork(nextUnitOfWork);
    }
  }
}

• 它接收一个 isYieldy 作为参数
• 这个参数意味着是否可以被中断
• Sync的任务和已超时的异步任务都是不可中断的
• 如果是不可中断的，只要有 nextUnitOfWork
• 就会继续调用 performUnitOfWork
• 如果是可以中断的，就通过判断 !shouldYield()
• 来看当前时间片中是否还有足够的时间继续渲染下一个节点

再来看下 performUnitOfWork

function performUnitOfWork(workInProgress: Fiber): Fiber | null {
  // The current, flushed, state of this fiber is the alternate.
  // Ideally nothing should rely on this, but relying on it here
  // means that we don't need an additional field on the work in
  // progress.
  const current = workInProgress.alternate;

  // See if beginning this work spawns more work.
  startWorkTimer(workInProgress);
  if (__DEV__) {
    ReactCurrentFiber.setCurrentFiber(workInProgress);
  }

  if (__DEV__ && replayFailedUnitOfWorkWithInvokeGuardedCallback) {
    stashedWorkInProgressProperties = assignFiberPropertiesInDEV(
      stashedWorkInProgressProperties,
      workInProgress,
    );
  }

  let next;
  if (enableProfilerTimer) {
    if (workInProgress.mode & ProfileMode) {
      startProfilerTimer(workInProgress);
    }

    next = beginWork(current, workInProgress, nextRenderExpirationTime);
    workInProgress.memoizedProps = workInProgress.pendingProps;

    if (workInProgress.mode & ProfileMode) {
      // Record the render duration assuming we didn't bailout (or error).
      stopProfilerTimerIfRunningAndRecordDelta(workInProgress, true);
    }
  } else {
    next = beginWork(current, workInProgress, nextRenderExpirationTime);
    workInProgress.memoizedProps = workInProgress.pendingProps;
  }

  if (__DEV__) {
    ReactCurrentFiber.resetCurrentFiber();
    if (isReplayingFailedUnitOfWork) {
      // Currently replaying a failed unit of work. This should be unreachable,
      // because the render phase is meant to be idempotent, and it should
      // have thrown again. Since it didn't, rethrow the original error, so
      // React's internal stack is not misaligned.
      rethrowOriginalError();
    }
  }
  if (__DEV__ && ReactFiberInstrumentation.debugTool) {
    ReactFiberInstrumentation.debugTool.onBeginWork(workInProgress);
  }

  if (next === null) {
    // If this doesn't spawn new work, complete the current work.
    next = completeUnitOfWork(workInProgress);
  }

  ReactCurrentOwner.current = null;

  return next;
}

• 它声明了一个 next 变量，next = beginWork(…), 这里涉及到对每个节点的更新
  • 更新完一个节点之后，它会返回它的下一个节点
  • 会更新 workInProgress.memoizedProps, 节点已经更新完了
  • 最新的 props 已经变成目前正在用的 props
  • 先跳过
• 跳过 DEV 的代码
• 如果 next === null 说明这个节点已经更新到子树的叶子节点了
  • 这棵子树就可以结束了
  • 结束就调用 completeUnitOfWork
  • 它也会返回它的下一个节点
• 最后，return next
  • 在 workLoop 函数中可看到，它会赋值给 nextUnitOfWork

    // 参考其中一个 while
    while (nextUnitOfWork !== null) {
      nextUnitOfWork = performUnitOfWork(nextUnitOfWork);
    }
    

  • 所以，真正到 nextUnitOfWork 为 null 的情况是它到了根节点，即 FiberRoot 节点
  • 它的 return 是 null，这时就跳出了 while 循环了