react-code-dataset
/
next.js
/packages
/next
/src
/client
/components
/segment-cache-impl
/scheduler.ts
| import type { | |
| FlightRouterState, | |
| Segment as FlightRouterStateSegment, | |
| Segment, | |
| } from '../../../server/app-render/types' | |
| import { HasLoadingBoundary } from '../../../server/app-render/types' | |
| import { matchSegment } from '../match-segments' | |
| import { | |
| readOrCreateRouteCacheEntry, | |
| readOrCreateSegmentCacheEntry, | |
| fetchRouteOnCacheMiss, | |
| fetchSegmentOnCacheMiss, | |
| EntryStatus, | |
| type FulfilledRouteCacheEntry, | |
| type RouteCacheEntry, | |
| type SegmentCacheEntry, | |
| type RouteTree, | |
| fetchSegmentPrefetchesUsingDynamicRequest, | |
| type PendingSegmentCacheEntry, | |
| convertRouteTreeToFlightRouterState, | |
| readOrCreateRevalidatingSegmentEntry, | |
| upsertSegmentEntry, | |
| type FulfilledSegmentCacheEntry, | |
| upgradeToPendingSegment, | |
| waitForSegmentCacheEntry, | |
| resetRevalidatingSegmentEntry, | |
| getSegmentKeypathForTask, | |
| } from './cache' | |
| import type { RouteCacheKey } from './cache-key' | |
| import { | |
| getCurrentCacheVersion, | |
| PrefetchPriority, | |
| FetchStrategy, | |
| type PrefetchTaskFetchStrategy, | |
| } from '../segment-cache' | |
| import { | |
| addSearchParamsIfPageSegment, | |
| PAGE_SEGMENT_KEY, | |
| } from '../../../shared/lib/segment' | |
| const scheduleMicrotask = | |
| typeof queueMicrotask === 'function' | |
| ? queueMicrotask | |
| : (fn: () => unknown) => | |
| Promise.resolve() | |
| .then(fn) | |
| .catch((error) => | |
| setTimeout(() => { | |
| throw error | |
| }) | |
| ) | |
| export type PrefetchTask = { | |
| key: RouteCacheKey | |
| /** | |
| * The FlightRouterState at the time the task was initiated. This is needed | |
| * when falling back to the non-PPR behavior, which only prefetches up to | |
| * the first loading boundary. | |
| */ | |
| treeAtTimeOfPrefetch: FlightRouterState | |
| /** | |
| * The cache version at the time the task was initiated. This is used to | |
| * determine if the cache was invalidated since the task was initiated. | |
| */ | |
| cacheVersion: number | |
| /** | |
| * Whether to prefetch dynamic data, in addition to static data. This is | |
| * used by `<Link prefetch={true}>`. | |
| * | |
| * Note that a task with `FetchStrategy.PPR` might need to use | |
| * `FetchStrategy.LoadingBoundary` instead if we find out that a route | |
| * does not support PPR after doing the initial route prefetch. | |
| */ | |
| fetchStrategy: PrefetchTaskFetchStrategy | |
| /** | |
| * sortId is an incrementing counter | |
| * | |
| * Newer prefetches are prioritized over older ones, so that as new links | |
| * enter the viewport, they are not starved by older links that are no | |
| * longer relevant. In the future, we can add additional prioritization | |
| * heuristics, like removing prefetches once a link leaves the viewport. | |
| * | |
| * The sortId is assigned when the prefetch is initiated, and reassigned if | |
| * the same task is prefetched again (effectively bumping it to the top of | |
| * the queue). | |
| * | |
| * TODO: We can add additional fields here to indicate what kind of prefetch | |
| * it is. For example, was it initiated by a link? Or was it an imperative | |
| * call? If it was initiated by a link, we can remove it from the queue when | |
| * the link leaves the viewport, but if it was an imperative call, then we | |
| * should keep it in the queue until it's fulfilled. | |
| * | |
| * We can also add priority levels. For example, hovering over a link could | |
| * increase the priority of its prefetch. | |
| */ | |
| sortId: number | |
| /** | |
| * The priority of the task. Like sortId, this affects the task's position in | |
| * the queue, so it must never be updated without resifting the heap. | |
| */ | |
| priority: PrefetchPriority | |
| /** | |
| * The phase of the task. Tasks are split into multiple phases so that their | |
| * priority can be adjusted based on what kind of work they're doing. | |
| * Concretely, prefetching the route tree is higher priority than prefetching | |
| * segment data. | |
| */ | |
| phase: PrefetchPhase | |
| /** | |
| * Temporary state for tracking the currently running task. This is currently | |
| * used to track whether a task deferred some work to run background at | |
| * priority, but we might need it for additional state in the future. | |
| */ | |
| hasBackgroundWork: boolean | |
| /** | |
| * True if the prefetch was cancelled. | |
| */ | |
| isCanceled: boolean | |
| /** | |
| * The callback passed to `router.prefetch`, if given. | |
| */ | |
| onInvalidate: null | (() => void) | |
| /** | |
| * The index of the task in the heap's backing array. Used to efficiently | |
| * change the priority of a task by re-sifting it, which requires knowing | |
| * where it is in the array. This is only used internally by the heap | |
| * algorithm. The naive alternative is indexOf every time a task is queued, | |
| * which has O(n) complexity. | |
| * | |
| * We also use this field to check whether a task is currently in the queue. | |
| */ | |
| _heapIndex: number | |
| } | |
| const enum PrefetchTaskExitStatus { | |
| /** | |
| * The task yielded because there are too many requests in progress. | |
| */ | |
| InProgress, | |
| /** | |
| * The task is blocked. It needs more data before it can proceed. | |
| * | |
| * Currently the only reason this happens is we're still waiting to receive a | |
| * route tree from the server, because we can't start prefetching the segments | |
| * until we know what to prefetch. | |
| */ | |
| Blocked, | |
| /** | |
| * There's nothing left to prefetch. | |
| */ | |
| Done, | |
| } | |
| /** | |
| * Prefetch tasks are processed in two phases: first the route tree is fetched, | |
| * then the segments. We use this to priortize tasks that have not yet fetched | |
| * the route tree. | |
| */ | |
| const enum PrefetchPhase { | |
| RouteTree = 1, | |
| Segments = 0, | |
| } | |
| export type PrefetchSubtaskResult<T> = { | |
| /** | |
| * A promise that resolves when the network connection is closed. | |
| */ | |
| closed: Promise<void> | |
| value: T | |
| } | |
| const taskHeap: Array<PrefetchTask> = [] | |
| let inProgressRequests = 0 | |
| let sortIdCounter = 0 | |
| let didScheduleMicrotask = false | |
| // The most recently hovered (or touched, etc) link, i.e. the most recent task | |
| // scheduled at Intent priority. There's only ever a single task at Intent | |
| // priority at a time. We reserve special network bandwidth for this task only. | |
| let mostRecentlyHoveredLink: PrefetchTask | null = null | |
| /** | |
| * Initiates a prefetch task for the given URL. If a prefetch for the same URL | |
| * is already in progress, this will bump it to the top of the queue. | |
| * | |
| * This is not a user-facing function. By the time this is called, the href is | |
| * expected to be validated and normalized. | |
| * | |
| * @param key The RouteCacheKey to prefetch. | |
| * @param treeAtTimeOfPrefetch The app's current FlightRouterState | |
| * @param fetchStrategy Whether to prefetch dynamic data, in addition to | |
| * static data. This is used by `<Link prefetch={true}>`. | |
| */ | |
| export function schedulePrefetchTask( | |
| key: RouteCacheKey, | |
| treeAtTimeOfPrefetch: FlightRouterState, | |
| fetchStrategy: PrefetchTaskFetchStrategy, | |
| priority: PrefetchPriority, | |
| onInvalidate: null | (() => void) | |
| ): PrefetchTask { | |
| // Spawn a new prefetch task | |
| const task: PrefetchTask = { | |
| key, | |
| treeAtTimeOfPrefetch, | |
| cacheVersion: getCurrentCacheVersion(), | |
| priority, | |
| phase: PrefetchPhase.RouteTree, | |
| hasBackgroundWork: false, | |
| fetchStrategy, | |
| sortId: sortIdCounter++, | |
| isCanceled: false, | |
| onInvalidate, | |
| _heapIndex: -1, | |
| } | |
| trackMostRecentlyHoveredLink(task) | |
| heapPush(taskHeap, task) | |
| // Schedule an async task to process the queue. | |
| // | |
| // The main reason we process the queue in an async task is for batching. | |
| // It's common for a single JS task/event to trigger multiple prefetches. | |
| // By deferring to a microtask, we only process the queue once per JS task. | |
| // If they have different priorities, it also ensures they are processed in | |
| // the optimal order. | |
| ensureWorkIsScheduled() | |
| return task | |
| } | |
| export function cancelPrefetchTask(task: PrefetchTask): void { | |
| // Remove the prefetch task from the queue. If the task already completed, | |
| // then this is a no-op. | |
| // | |
| // We must also explicitly mark the task as canceled so that a blocked task | |
| // does not get added back to the queue when it's pinged by the network. | |
| task.isCanceled = true | |
| heapDelete(taskHeap, task) | |
| } | |
| export function reschedulePrefetchTask( | |
| task: PrefetchTask, | |
| treeAtTimeOfPrefetch: FlightRouterState, | |
| fetchStrategy: PrefetchTaskFetchStrategy, | |
| priority: PrefetchPriority | |
| ): void { | |
| // Bump the prefetch task to the top of the queue, as if it were a fresh | |
| // task. This is essentially the same as canceling the task and scheduling | |
| // a new one, except it reuses the original object. | |
| // | |
| // The primary use case is to increase the priority of a Link-initated | |
| // prefetch on hover. | |
| // Un-cancel the task, in case it was previously canceled. | |
| task.isCanceled = false | |
| task.phase = PrefetchPhase.RouteTree | |
| // Assign a new sort ID to move it ahead of all other tasks at the same | |
| // priority level. (Higher sort IDs are processed first.) | |
| task.sortId = sortIdCounter++ | |
| task.priority = | |
| // If this task is the most recently hovered link, maintain its | |
| // Intent priority, even if the rescheduled priority is lower. | |
| task === mostRecentlyHoveredLink ? PrefetchPriority.Intent : priority | |
| task.treeAtTimeOfPrefetch = treeAtTimeOfPrefetch | |
| task.fetchStrategy = fetchStrategy | |
| trackMostRecentlyHoveredLink(task) | |
| if (task._heapIndex !== -1) { | |
| // The task is already in the queue. | |
| heapResift(taskHeap, task) | |
| } else { | |
| heapPush(taskHeap, task) | |
| } | |
| ensureWorkIsScheduled() | |
| } | |
| export function isPrefetchTaskDirty( | |
| task: PrefetchTask, | |
| nextUrl: string | null, | |
| tree: FlightRouterState | |
| ): boolean { | |
| // This is used to quickly bail out of a prefetch task if the result is | |
| // guaranteed to not have changed since the task was initiated. This is | |
| // strictly an optimization — theoretically, if it always returned true, no | |
| // behavior should change because a full prefetch task will effectively | |
| // perform the same checks. | |
| const currentCacheVersion = getCurrentCacheVersion() | |
| return ( | |
| task.cacheVersion !== currentCacheVersion || | |
| task.treeAtTimeOfPrefetch !== tree || | |
| task.key.nextUrl !== nextUrl | |
| ) | |
| } | |
| function trackMostRecentlyHoveredLink(task: PrefetchTask) { | |
| // Track the mostly recently hovered link, i.e. the most recently scheduled | |
| // task at Intent priority. There must only be one such task at a time. | |
| if ( | |
| task.priority === PrefetchPriority.Intent && | |
| task !== mostRecentlyHoveredLink | |
| ) { | |
| if (mostRecentlyHoveredLink !== null) { | |
| // Bump the previously hovered link's priority down to Default. | |
| if (mostRecentlyHoveredLink.priority !== PrefetchPriority.Background) { | |
| mostRecentlyHoveredLink.priority = PrefetchPriority.Default | |
| heapResift(taskHeap, mostRecentlyHoveredLink) | |
| } | |
| } | |
| mostRecentlyHoveredLink = task | |
| } | |
| } | |
| function ensureWorkIsScheduled() { | |
| if (didScheduleMicrotask) { | |
| // Already scheduled a task to process the queue | |
| return | |
| } | |
| didScheduleMicrotask = true | |
| scheduleMicrotask(processQueueInMicrotask) | |
| } | |
| /** | |
| * Checks if we've exceeded the maximum number of concurrent prefetch requests, | |
| * to avoid saturating the browser's internal network queue. This is a | |
| * cooperative limit — prefetch tasks should check this before issuing | |
| * new requests. | |
| */ | |
| function hasNetworkBandwidth(task: PrefetchTask): boolean { | |
| // TODO: Also check if there's an in-progress navigation. We should never | |
| // add prefetch requests to the network queue if an actual navigation is | |
| // taking place, to ensure there's sufficient bandwidth for render-blocking | |
| // data and resources. | |
| // TODO: Consider reserving some amount of bandwidth for static prefetches. | |
| if (task.priority === PrefetchPriority.Intent) { | |
| // The most recently hovered link is allowed to exceed the default limit. | |
| // | |
| // The goal is to always have enough bandwidth to start a new prefetch | |
| // request when hovering over a link. | |
| // | |
| // However, because we don't abort in-progress requests, it's still possible | |
| // we'll run out of bandwidth. When links are hovered in quick succession, | |
| // there could be multiple hover requests running simultaneously. | |
| return inProgressRequests < 12 | |
| } | |
| // The default limit is lower than the limit for a hovered link. | |
| return inProgressRequests < 4 | |
| } | |
| function spawnPrefetchSubtask<T>( | |
| prefetchSubtask: Promise<PrefetchSubtaskResult<T> | null> | |
| ): Promise<T | null> { | |
| // When the scheduler spawns an async task, we don't await its result. | |
| // Instead, the async task writes its result directly into the cache, then | |
| // pings the scheduler to continue. | |
| // | |
| // We process server responses streamingly, so the prefetch subtask will | |
| // likely resolve before we're finished receiving all the data. The subtask | |
| // result includes a promise that resolves once the network connection is | |
| // closed. The scheduler uses this to control network bandwidth by tracking | |
| // and limiting the number of concurrent requests. | |
| inProgressRequests++ | |
| return prefetchSubtask.then((result) => { | |
| if (result === null) { | |
| // The prefetch task errored before it could start processing the | |
| // network stream. Assume the connection is closed. | |
| onPrefetchConnectionClosed() | |
| return null | |
| } | |
| // Wait for the connection to close before freeing up more bandwidth. | |
| result.closed.then(onPrefetchConnectionClosed) | |
| return result.value | |
| }) | |
| } | |
| function onPrefetchConnectionClosed(): void { | |
| inProgressRequests-- | |
| // Notify the scheduler that we have more bandwidth, and can continue | |
| // processing tasks. | |
| ensureWorkIsScheduled() | |
| } | |
| /** | |
| * Notify the scheduler that we've received new data for an in-progress | |
| * prefetch. The corresponding task will be added back to the queue (unless the | |
| * task has been canceled in the meantime). | |
| */ | |
| export function pingPrefetchTask(task: PrefetchTask) { | |
| // "Ping" a prefetch that's already in progress to notify it of new data. | |
| if ( | |
| // Check if prefetch was canceled. | |
| task.isCanceled || | |
| // Check if prefetch is already queued. | |
| task._heapIndex !== -1 | |
| ) { | |
| return | |
| } | |
| // Add the task back to the queue. | |
| heapPush(taskHeap, task) | |
| ensureWorkIsScheduled() | |
| } | |
| function processQueueInMicrotask() { | |
| didScheduleMicrotask = false | |
| // We aim to minimize how often we read the current time. Since nearly all | |
| // functions in the prefetch scheduler are synchronous, we can read the time | |
| // once and pass it as an argument wherever it's needed. | |
| const now = Date.now() | |
| // Process the task queue until we run out of network bandwidth. | |
| let task = heapPeek(taskHeap) | |
| while (task !== null && hasNetworkBandwidth(task)) { | |
| task.cacheVersion = getCurrentCacheVersion() | |
| const route = readOrCreateRouteCacheEntry(now, task) | |
| const exitStatus = pingRootRouteTree(now, task, route) | |
| // The `hasBackgroundWork` field is only valid for a single attempt. Reset | |
| // it immediately upon exit. | |
| const hasBackgroundWork = task.hasBackgroundWork | |
| task.hasBackgroundWork = false | |
| switch (exitStatus) { | |
| case PrefetchTaskExitStatus.InProgress: | |
| // The task yielded because there are too many requests in progress. | |
| // Stop processing tasks until we have more bandwidth. | |
| return | |
| case PrefetchTaskExitStatus.Blocked: | |
| // The task is blocked. It needs more data before it can proceed. | |
| // Keep the task out of the queue until the server responds. | |
| heapPop(taskHeap) | |
| // Continue to the next task | |
| task = heapPeek(taskHeap) | |
| continue | |
| case PrefetchTaskExitStatus.Done: | |
| if (task.phase === PrefetchPhase.RouteTree) { | |
| // Finished prefetching the route tree. Proceed to prefetching | |
| // the segments. | |
| task.phase = PrefetchPhase.Segments | |
| heapResift(taskHeap, task) | |
| } else if (hasBackgroundWork) { | |
| // The task spawned additional background work. Reschedule the task | |
| // at background priority. | |
| task.priority = PrefetchPriority.Background | |
| heapResift(taskHeap, task) | |
| } else { | |
| // The prefetch is complete. Continue to the next task. | |
| heapPop(taskHeap) | |
| } | |
| task = heapPeek(taskHeap) | |
| continue | |
| default: | |
| exitStatus satisfies never | |
| } | |
| } | |
| } | |
| /** | |
| * Check this during a prefetch task to determine if background work can be | |
| * performed. If so, it evaluates to `true`. Otherwise, it returns `false`, | |
| * while also scheduling a background task to run later. Usage: | |
| * | |
| * @example | |
| * if (background(task)) { | |
| * // Perform background-pri work | |
| * } | |
| */ | |
| function background(task: PrefetchTask): boolean { | |
| if (task.priority === PrefetchPriority.Background) { | |
| return true | |
| } | |
| task.hasBackgroundWork = true | |
| return false | |
| } | |
| function pingRootRouteTree( | |
| now: number, | |
| task: PrefetchTask, | |
| route: RouteCacheEntry | |
| ): PrefetchTaskExitStatus { | |
| switch (route.status) { | |
| case EntryStatus.Empty: { | |
| // Route is not yet cached, and there's no request already in progress. | |
| // Spawn a task to request the route, load it into the cache, and ping | |
| // the task to continue. | |
| // TODO: There are multiple strategies in the <Link> API for prefetching | |
| // a route. Currently we've only implemented the main one: per-segment, | |
| // static-data only. | |
| // | |
| // There's also <Link prefetch={true}> which prefetches both static *and* | |
| // dynamic data. Similarly, we need to fallback to the old, per-page | |
| // behavior if PPR is disabled for a route (via the incremental opt-in). | |
| // | |
| // Those cases will be handled here. | |
| spawnPrefetchSubtask(fetchRouteOnCacheMiss(route, task)) | |
| // If the request takes longer than a minute, a subsequent request should | |
| // retry instead of waiting for this one. When the response is received, | |
| // this value will be replaced by a new value based on the stale time sent | |
| // from the server. | |
| // TODO: We should probably also manually abort the fetch task, to reclaim | |
| // server bandwidth. | |
| route.staleAt = now + 60 * 1000 | |
| // Upgrade to Pending so we know there's already a request in progress | |
| route.status = EntryStatus.Pending | |
| // Intentional fallthrough to the Pending branch | |
| } | |
| case EntryStatus.Pending: { | |
| // Still pending. We can't start prefetching the segments until the route | |
| // tree has loaded. Add the task to the set of blocked tasks so that it | |
| // is notified when the route tree is ready. | |
| const blockedTasks = route.blockedTasks | |
| if (blockedTasks === null) { | |
| route.blockedTasks = new Set([task]) | |
| } else { | |
| blockedTasks.add(task) | |
| } | |
| return PrefetchTaskExitStatus.Blocked | |
| } | |
| case EntryStatus.Rejected: { | |
| // Route tree failed to load. Treat as a 404. | |
| return PrefetchTaskExitStatus.Done | |
| } | |
| case EntryStatus.Fulfilled: { | |
| if (task.phase !== PrefetchPhase.Segments) { | |
| // Do not prefetch segment data until we've entered the segment phase. | |
| return PrefetchTaskExitStatus.Done | |
| } | |
| // Recursively fill in the segment tree. | |
| if (!hasNetworkBandwidth(task)) { | |
| // Stop prefetching segments until there's more bandwidth. | |
| return PrefetchTaskExitStatus.InProgress | |
| } | |
| const tree = route.tree | |
| // A task's fetch strategy gets set to `PPR` for any "auto" prefetch. | |
| // If it turned out that the route isn't PPR-enabled, we need to use `LoadingBoundary` instead. | |
| const fetchStrategy = | |
| task.fetchStrategy === FetchStrategy.PPR | |
| ? route.isPPREnabled | |
| ? FetchStrategy.PPR | |
| : FetchStrategy.LoadingBoundary | |
| : task.fetchStrategy | |
| switch (fetchStrategy) { | |
| case FetchStrategy.PPR: | |
| // Individually prefetch the static shell for each segment. This is | |
| // the default prefetching behavior for static routes, or when PPR is | |
| // enabled. It will not include any dynamic data. | |
| return pingPPRRouteTree(now, task, route, tree) | |
| case FetchStrategy.Full: | |
| case FetchStrategy.LoadingBoundary: { | |
| // Prefetch multiple segments using a single dynamic request. | |
| const spawnedEntries = new Map<string, PendingSegmentCacheEntry>() | |
| const dynamicRequestTree = diffRouteTreeAgainstCurrent( | |
| now, | |
| task, | |
| route, | |
| task.treeAtTimeOfPrefetch, | |
| tree, | |
| spawnedEntries, | |
| fetchStrategy | |
| ) | |
| const needsDynamicRequest = spawnedEntries.size > 0 | |
| if (needsDynamicRequest) { | |
| // Perform a dynamic prefetch request and populate the cache with | |
| // the result | |
| spawnPrefetchSubtask( | |
| fetchSegmentPrefetchesUsingDynamicRequest( | |
| task, | |
| route, | |
| fetchStrategy, | |
| dynamicRequestTree, | |
| spawnedEntries | |
| ) | |
| ) | |
| } | |
| return PrefetchTaskExitStatus.Done | |
| } | |
| default: | |
| fetchStrategy satisfies never | |
| } | |
| break | |
| } | |
| default: { | |
| route satisfies never | |
| } | |
| } | |
| return PrefetchTaskExitStatus.Done | |
| } | |
| function pingPPRRouteTree( | |
| now: number, | |
| task: PrefetchTask, | |
| route: FulfilledRouteCacheEntry, | |
| tree: RouteTree | |
| ): PrefetchTaskExitStatus.InProgress | PrefetchTaskExitStatus.Done { | |
| const segment = readOrCreateSegmentCacheEntry(now, task, route, tree.key) | |
| pingPerSegment(now, task, route, segment, task.key, tree.key) | |
| if (tree.slots !== null) { | |
| if (!hasNetworkBandwidth(task)) { | |
| // Stop prefetching segments until there's more bandwidth. | |
| return PrefetchTaskExitStatus.InProgress | |
| } | |
| // Recursively ping the children. | |
| for (const parallelRouteKey in tree.slots) { | |
| const childTree = tree.slots[parallelRouteKey] | |
| const childExitStatus = pingPPRRouteTree(now, task, route, childTree) | |
| if (childExitStatus === PrefetchTaskExitStatus.InProgress) { | |
| // Child yielded without finishing. | |
| return PrefetchTaskExitStatus.InProgress | |
| } | |
| } | |
| } | |
| // This segment and all its children have finished prefetching. | |
| return PrefetchTaskExitStatus.Done | |
| } | |
| function diffRouteTreeAgainstCurrent( | |
| now: number, | |
| task: PrefetchTask, | |
| route: FulfilledRouteCacheEntry, | |
| oldTree: FlightRouterState, | |
| newTree: RouteTree, | |
| spawnedEntries: Map<string, PendingSegmentCacheEntry>, | |
| fetchStrategy: FetchStrategy.Full | FetchStrategy.LoadingBoundary | |
| ): FlightRouterState { | |
| // This is a single recursive traversal that does multiple things: | |
| // - Finds the parts of the target route (newTree) that are not part of | |
| // of the current page (oldTree) by diffing them, using the same algorithm | |
| // as a real navigation. | |
| // - Constructs a request tree (FlightRouterState) that describes which | |
| // segments need to be prefetched and which ones are already cached. | |
| // - Creates a set of pending cache entries for the segments that need to | |
| // be prefetched, so that a subsequent prefetch task does not request the | |
| // same segments again. | |
| const oldTreeChildren = oldTree[1] | |
| const newTreeChildren = newTree.slots | |
| let requestTreeChildren: Record<string, FlightRouterState> = {} | |
| if (newTreeChildren !== null) { | |
| for (const parallelRouteKey in newTreeChildren) { | |
| const newTreeChild = newTreeChildren[parallelRouteKey] | |
| const newTreeChildSegment = newTreeChild.segment | |
| const oldTreeChild: FlightRouterState | void = | |
| oldTreeChildren[parallelRouteKey] | |
| const oldTreeChildSegment: FlightRouterStateSegment | void = | |
| oldTreeChild?.[0] | |
| if ( | |
| oldTreeChildSegment !== undefined && | |
| doesCurrentSegmentMatchCachedSegment( | |
| route, | |
| newTreeChildSegment, | |
| oldTreeChildSegment | |
| ) | |
| ) { | |
| // This segment is already part of the current route. Keep traversing. | |
| const requestTreeChild = diffRouteTreeAgainstCurrent( | |
| now, | |
| task, | |
| route, | |
| oldTreeChild, | |
| newTreeChild, | |
| spawnedEntries, | |
| fetchStrategy | |
| ) | |
| requestTreeChildren[parallelRouteKey] = requestTreeChild | |
| } else { | |
| // This segment is not part of the current route. We're entering a | |
| // part of the tree that we need to prefetch (unless everything is | |
| // already cached). | |
| switch (fetchStrategy) { | |
| case FetchStrategy.LoadingBoundary: { | |
| // When PPR is disabled, we can't prefetch per segment. We must | |
| // fallback to the old prefetch behavior and send a dynamic request. | |
| // Only routes that include a loading boundary can be prefetched in | |
| // this way. | |
| // | |
| // This is simlar to a "full" prefetch, but we're much more | |
| // conservative about which segments to include in the request. | |
| // | |
| // The server will only render up to the first loading boundary | |
| // inside new part of the tree. If there's no loading boundary | |
| // anywhere in the tree, the server will never return any data, so | |
| // we can skip the request. | |
| const subtreeHasLoadingBoundary = | |
| newTreeChild.hasLoadingBoundary !== | |
| HasLoadingBoundary.SubtreeHasNoLoadingBoundary | |
| const requestTreeChild = subtreeHasLoadingBoundary | |
| ? pingPPRDisabledRouteTreeUpToLoadingBoundary( | |
| now, | |
| task, | |
| route, | |
| newTreeChild, | |
| null, | |
| spawnedEntries | |
| ) | |
| : // There's no loading boundary within this tree. Bail out. | |
| convertRouteTreeToFlightRouterState(newTreeChild) | |
| requestTreeChildren[parallelRouteKey] = requestTreeChild | |
| break | |
| } | |
| case FetchStrategy.Full: { | |
| // This is a "full" prefetch. Fetch all the data in the tree, both | |
| // static and dynamic. We issue roughly the same request that we | |
| // would during a real navigation. The goal is that once the | |
| // navigation occurs, the router should not have to fetch any | |
| // additional data. | |
| // | |
| // Although the response will include dynamic data, opting into a | |
| // Full prefetch — via <Link prefetch={true}> — implicitly | |
| // instructs the cache to treat the response as "static", or non- | |
| // dynamic, since the whole point is to cache it for | |
| // future navigations. | |
| // | |
| // Construct a tree (currently a FlightRouterState) that represents | |
| // which segments need to be prefetched and which ones are already | |
| // cached. If the tree is empty, then we can exit. Otherwise, we'll | |
| // send the request tree to the server and use the response to | |
| // populate the segment cache. | |
| const requestTreeChild = pingRouteTreeAndIncludeDynamicData( | |
| now, | |
| task, | |
| route, | |
| newTreeChild, | |
| false, | |
| spawnedEntries | |
| ) | |
| requestTreeChildren[parallelRouteKey] = requestTreeChild | |
| break | |
| } | |
| default: | |
| fetchStrategy satisfies never | |
| } | |
| } | |
| } | |
| } | |
| const requestTree: FlightRouterState = [ | |
| newTree.segment, | |
| requestTreeChildren, | |
| null, | |
| null, | |
| newTree.isRootLayout, | |
| ] | |
| return requestTree | |
| } | |
| function pingPPRDisabledRouteTreeUpToLoadingBoundary( | |
| now: number, | |
| task: PrefetchTask, | |
| route: FulfilledRouteCacheEntry, | |
| tree: RouteTree, | |
| refetchMarkerContext: 'refetch' | 'inside-shared-layout' | null, | |
| spawnedEntries: Map<string, PendingSegmentCacheEntry> | |
| ): FlightRouterState { | |
| // This function is similar to pingRouteTreeAndIncludeDynamicData, except the | |
| // server is only going to return a minimal loading state — it will stop | |
| // rendering at the first loading boundary. Whereas a Full prefetch is | |
| // intentionally aggressive and tries to pretfetch all the data that will be | |
| // needed for a navigation, a LoadingBoundary prefetch is much more | |
| // conservative. For example, it will omit from the request tree any segment | |
| // that is already cached, regardles of whether it's partial or full. By | |
| // contrast, a Full prefetch will refetch partial segments. | |
| // "inside-shared-layout" tells the server where to start looking for a | |
| // loading boundary. | |
| let refetchMarker: 'refetch' | 'inside-shared-layout' | null = | |
| refetchMarkerContext === null ? 'inside-shared-layout' : null | |
| const segment = readOrCreateSegmentCacheEntry(now, task, route, tree.key) | |
| switch (segment.status) { | |
| case EntryStatus.Empty: { | |
| // This segment is not cached. Add a refetch marker so the server knows | |
| // to start rendering here. | |
| // TODO: Instead of a "refetch" marker, we could just omit this subtree's | |
| // FlightRouterState from the request tree. I think this would probably | |
| // already work even without any updates to the server. For consistency, | |
| // though, I'll send the full tree and we'll look into this later as part | |
| // of a larger redesign of the request protocol. | |
| // Add the pending cache entry to the result map. | |
| spawnedEntries.set( | |
| tree.key, | |
| upgradeToPendingSegment( | |
| segment, | |
| // Set the fetch strategy to LoadingBoundary to indicate that the server | |
| // might not include it in the pending response. If another route is able | |
| // to issue a per-segment request, we'll do that in the background. | |
| FetchStrategy.LoadingBoundary | |
| ) | |
| ) | |
| if (refetchMarkerContext !== 'refetch') { | |
| refetchMarker = refetchMarkerContext = 'refetch' | |
| } else { | |
| // There's already a parent with a refetch marker, so we don't need | |
| // to add another one. | |
| } | |
| break | |
| } | |
| case EntryStatus.Fulfilled: { | |
| // The segment is already cached. | |
| const segmentHasLoadingBoundary = | |
| tree.hasLoadingBoundary === HasLoadingBoundary.SegmentHasLoadingBoundary | |
| if (segmentHasLoadingBoundary) { | |
| // This segment has a loading boundary, which means the server won't | |
| // render its children. So there's nothing left to prefetch along this | |
| // path. We can bail out. | |
| return convertRouteTreeToFlightRouterState(tree) | |
| } | |
| // NOTE: If the cached segment were fetched using PPR, then it might be | |
| // partial. We could get a more complete version of the segment by | |
| // including it in this non-PPR request. | |
| // | |
| // We're intentionally choosing not to, though, because it's generally | |
| // better to avoid doing a dynamic prefetch whenever possible. | |
| break | |
| } | |
| case EntryStatus.Pending: { | |
| // There's another prefetch currently in progress. Don't add the refetch | |
| // marker yet, so the server knows it can skip rendering this segment. | |
| break | |
| } | |
| case EntryStatus.Rejected: { | |
| // The segment failed to load. We shouldn't issue another request until | |
| // the stale time has elapsed. | |
| break | |
| } | |
| default: | |
| segment satisfies never | |
| } | |
| const requestTreeChildren: Record<string, FlightRouterState> = {} | |
| if (tree.slots !== null) { | |
| for (const parallelRouteKey in tree.slots) { | |
| const childTree = tree.slots[parallelRouteKey] | |
| requestTreeChildren[parallelRouteKey] = | |
| pingPPRDisabledRouteTreeUpToLoadingBoundary( | |
| now, | |
| task, | |
| route, | |
| childTree, | |
| refetchMarkerContext, | |
| spawnedEntries | |
| ) | |
| } | |
| } | |
| const requestTree: FlightRouterState = [ | |
| tree.segment, | |
| requestTreeChildren, | |
| null, | |
| refetchMarker, | |
| tree.isRootLayout, | |
| ] | |
| return requestTree | |
| } | |
| function pingRouteTreeAndIncludeDynamicData( | |
| now: number, | |
| task: PrefetchTask, | |
| route: FulfilledRouteCacheEntry, | |
| tree: RouteTree, | |
| isInsideRefetchingParent: boolean, | |
| spawnedEntries: Map<string, PendingSegmentCacheEntry> | |
| ): FlightRouterState { | |
| // The tree we're constructing is the same shape as the tree we're navigating | |
| // to. But even though this is a "new" tree, some of the individual segments | |
| // may be cached as a result of other route prefetches. | |
| // | |
| // So we need to find the first uncached segment along each path add an | |
| // explicit "refetch" marker so the server knows where to start rendering. | |
| // Once the server starts rendering along a path, it keeps rendering the | |
| // entire subtree. | |
| const segment = readOrCreateSegmentCacheEntry(now, task, route, tree.key) | |
| let spawnedSegment: PendingSegmentCacheEntry | null = null | |
| switch (segment.status) { | |
| case EntryStatus.Empty: { | |
| // This segment is not cached. Include it in the request. | |
| spawnedSegment = upgradeToPendingSegment(segment, FetchStrategy.Full) | |
| break | |
| } | |
| case EntryStatus.Fulfilled: { | |
| // The segment is already cached. | |
| if (segment.isPartial) { | |
| // The cached segment contians dynamic holes. Since this is a Full | |
| // prefetch, we need to include it in the request. | |
| spawnedSegment = pingFullSegmentRevalidation( | |
| now, | |
| task, | |
| route, | |
| segment, | |
| tree.key | |
| ) | |
| } | |
| break | |
| } | |
| case EntryStatus.Pending: | |
| case EntryStatus.Rejected: { | |
| // There's either another prefetch currently in progress, or the previous | |
| // attempt failed. If it wasn't a Full prefetch, fetch it again. | |
| if (segment.fetchStrategy !== FetchStrategy.Full) { | |
| spawnedSegment = pingFullSegmentRevalidation( | |
| now, | |
| task, | |
| route, | |
| segment, | |
| tree.key | |
| ) | |
| } | |
| break | |
| } | |
| default: | |
| segment satisfies never | |
| } | |
| const requestTreeChildren: Record<string, FlightRouterState> = {} | |
| if (tree.slots !== null) { | |
| for (const parallelRouteKey in tree.slots) { | |
| const childTree = tree.slots[parallelRouteKey] | |
| requestTreeChildren[parallelRouteKey] = | |
| pingRouteTreeAndIncludeDynamicData( | |
| now, | |
| task, | |
| route, | |
| childTree, | |
| isInsideRefetchingParent || spawnedSegment !== null, | |
| spawnedEntries | |
| ) | |
| } | |
| } | |
| if (spawnedSegment !== null) { | |
| // Add the pending entry to the result map. | |
| spawnedEntries.set(tree.key, spawnedSegment) | |
| } | |
| // Don't bother to add a refetch marker if one is already present in a parent. | |
| const refetchMarker = | |
| !isInsideRefetchingParent && spawnedSegment !== null ? 'refetch' : null | |
| const requestTree: FlightRouterState = [ | |
| tree.segment, | |
| requestTreeChildren, | |
| null, | |
| refetchMarker, | |
| tree.isRootLayout, | |
| ] | |
| return requestTree | |
| } | |
| function pingPerSegment( | |
| now: number, | |
| task: PrefetchTask, | |
| route: FulfilledRouteCacheEntry, | |
| segment: SegmentCacheEntry, | |
| routeKey: RouteCacheKey, | |
| segmentKey: string | |
| ): void { | |
| switch (segment.status) { | |
| case EntryStatus.Empty: | |
| // Upgrade to Pending so we know there's already a request in progress | |
| spawnPrefetchSubtask( | |
| fetchSegmentOnCacheMiss( | |
| route, | |
| upgradeToPendingSegment(segment, FetchStrategy.PPR), | |
| routeKey, | |
| segmentKey | |
| ) | |
| ) | |
| break | |
| case EntryStatus.Pending: { | |
| // There's already a request in progress. Depending on what kind of | |
| // request it is, we may want to revalidate it. | |
| switch (segment.fetchStrategy) { | |
| case FetchStrategy.PPR: | |
| case FetchStrategy.Full: | |
| // There's already a request in progress. Don't do anything. | |
| break | |
| case FetchStrategy.LoadingBoundary: | |
| // There's a pending request, but because it's using the old | |
| // prefetching strategy, we can't be sure if it will be fulfilled by | |
| // the response — it might be inside the loading boundary. Perform | |
| // a revalidation, but because it's speculative, wait to do it at | |
| // background priority. | |
| if (background(task)) { | |
| // TODO: Instead of speculatively revalidating, consider including | |
| // `hasLoading` in the route tree prefetch response. | |
| pingPPRSegmentRevalidation( | |
| now, | |
| task, | |
| segment, | |
| route, | |
| routeKey, | |
| segmentKey | |
| ) | |
| } | |
| break | |
| default: | |
| segment.fetchStrategy satisfies never | |
| } | |
| break | |
| } | |
| case EntryStatus.Rejected: { | |
| // The existing entry in the cache was rejected. Depending on how it | |
| // was originally fetched, we may or may not want to revalidate it. | |
| switch (segment.fetchStrategy) { | |
| case FetchStrategy.PPR: | |
| case FetchStrategy.Full: | |
| // The previous attempt to fetch this entry failed. Don't attempt to | |
| // fetch it again until the entry expires. | |
| break | |
| case FetchStrategy.LoadingBoundary: | |
| // There's a rejected entry, but it was fetched using the loading | |
| // boundary strategy. So the reason it wasn't returned by the server | |
| // might just be because it was inside a loading boundary. Or because | |
| // there was a dynamic rewrite. Revalidate it using the per- | |
| // segment strategy. | |
| // | |
| // Because a rejected segment will definitely prevent the segment (and | |
| // all of its children) from rendering, we perform this revalidation | |
| // immediately instead of deferring it to a background task. | |
| pingPPRSegmentRevalidation( | |
| now, | |
| task, | |
| segment, | |
| route, | |
| routeKey, | |
| segmentKey | |
| ) | |
| break | |
| default: | |
| segment.fetchStrategy satisfies never | |
| } | |
| break | |
| } | |
| case EntryStatus.Fulfilled: | |
| // Segment is already cached. There's nothing left to prefetch. | |
| break | |
| default: | |
| segment satisfies never | |
| } | |
| // Segments do not have dependent tasks, so once the prefetch is initiated, | |
| // there's nothing else for us to do (except write the server data into the | |
| // entry, which is handled by `fetchSegmentOnCacheMiss`). | |
| } | |
| function pingPPRSegmentRevalidation( | |
| now: number, | |
| task: PrefetchTask, | |
| currentSegment: SegmentCacheEntry, | |
| route: FulfilledRouteCacheEntry, | |
| routeKey: RouteCacheKey, | |
| segmentKey: string | |
| ): void { | |
| const revalidatingSegment = readOrCreateRevalidatingSegmentEntry( | |
| now, | |
| currentSegment | |
| ) | |
| switch (revalidatingSegment.status) { | |
| case EntryStatus.Empty: | |
| // Spawn a prefetch request and upsert the segment into the cache | |
| // upon completion. | |
| upsertSegmentOnCompletion( | |
| task, | |
| route, | |
| segmentKey, | |
| spawnPrefetchSubtask( | |
| fetchSegmentOnCacheMiss( | |
| route, | |
| upgradeToPendingSegment(revalidatingSegment, FetchStrategy.PPR), | |
| routeKey, | |
| segmentKey | |
| ) | |
| ) | |
| ) | |
| break | |
| case EntryStatus.Pending: | |
| // There's already a revalidation in progress. | |
| break | |
| case EntryStatus.Fulfilled: | |
| case EntryStatus.Rejected: | |
| // A previous revalidation attempt finished, but we chose not to replace | |
| // the existing entry in the cache. Don't try again until or unless the | |
| // revalidation entry expires. | |
| break | |
| default: | |
| revalidatingSegment satisfies never | |
| } | |
| } | |
| function pingFullSegmentRevalidation( | |
| now: number, | |
| task: PrefetchTask, | |
| route: FulfilledRouteCacheEntry, | |
| currentSegment: SegmentCacheEntry, | |
| segmentKey: string | |
| ): PendingSegmentCacheEntry | null { | |
| const revalidatingSegment = readOrCreateRevalidatingSegmentEntry( | |
| now, | |
| currentSegment | |
| ) | |
| if (revalidatingSegment.status === EntryStatus.Empty) { | |
| // During a Full prefetch, a single dynamic request is made for all the | |
| // segments that we need. So we don't initiate a request here directly. By | |
| // returning a pending entry from this function, it signals to the caller | |
| // that this segment should be included in the request that's sent to | |
| // the server. | |
| const pendingSegment = upgradeToPendingSegment( | |
| revalidatingSegment, | |
| FetchStrategy.Full | |
| ) | |
| upsertSegmentOnCompletion( | |
| task, | |
| route, | |
| segmentKey, | |
| waitForSegmentCacheEntry(pendingSegment) | |
| ) | |
| return pendingSegment | |
| } else { | |
| // There's already a revalidation in progress. | |
| const nonEmptyRevalidatingSegment = revalidatingSegment | |
| if (nonEmptyRevalidatingSegment.fetchStrategy !== FetchStrategy.Full) { | |
| // The existing revalidation was not fetched using the Full strategy. | |
| // Reset it and start a new revalidation. | |
| const emptySegment = resetRevalidatingSegmentEntry( | |
| nonEmptyRevalidatingSegment | |
| ) | |
| const pendingSegment = upgradeToPendingSegment( | |
| emptySegment, | |
| FetchStrategy.Full | |
| ) | |
| upsertSegmentOnCompletion( | |
| task, | |
| route, | |
| segmentKey, | |
| waitForSegmentCacheEntry(pendingSegment) | |
| ) | |
| return pendingSegment | |
| } | |
| switch (nonEmptyRevalidatingSegment.status) { | |
| case EntryStatus.Pending: | |
| // There's already an in-progress prefetch that includes this segment. | |
| return null | |
| case EntryStatus.Fulfilled: | |
| case EntryStatus.Rejected: | |
| // A previous revalidation attempt finished, but we chose not to replace | |
| // the existing entry in the cache. Don't try again until or unless the | |
| // revalidation entry expires. | |
| return null | |
| default: | |
| nonEmptyRevalidatingSegment satisfies never | |
| return null | |
| } | |
| } | |
| } | |
| const noop = () => {} | |
| function upsertSegmentOnCompletion( | |
| task: PrefetchTask, | |
| route: FulfilledRouteCacheEntry, | |
| key: string, | |
| promise: Promise<FulfilledSegmentCacheEntry | null> | |
| ) { | |
| // Wait for a segment to finish loading, then upsert it into the cache | |
| promise.then((fulfilled) => { | |
| if (fulfilled !== null) { | |
| // Received new data. Attempt to replace the existing entry in the cache. | |
| const keypath = getSegmentKeypathForTask(task, route, key) | |
| upsertSegmentEntry(Date.now(), keypath, fulfilled) | |
| } | |
| }, noop) | |
| } | |
| function doesCurrentSegmentMatchCachedSegment( | |
| route: FulfilledRouteCacheEntry, | |
| currentSegment: Segment, | |
| cachedSegment: Segment | |
| ): boolean { | |
| if (cachedSegment === PAGE_SEGMENT_KEY) { | |
| // In the FlightRouterState stored by the router, the page segment has the | |
| // rendered search params appended to the name of the segment. In the | |
| // prefetch cache, however, this is stored separately. So, when comparing | |
| // the router's current FlightRouterState to the cached FlightRouterState, | |
| // we need to make sure we compare both parts of the segment. | |
| // TODO: This is not modeled clearly. We use the same type, | |
| // FlightRouterState, for both the CacheNode tree _and_ the prefetch cache | |
| // _and_ the server response format, when conceptually those are three | |
| // different things and treated in different ways. We should encode more of | |
| // this information into the type design so mistakes are less likely. | |
| return ( | |
| currentSegment === | |
| addSearchParamsIfPageSegment( | |
| PAGE_SEGMENT_KEY, | |
| Object.fromEntries(new URLSearchParams(route.renderedSearch)) | |
| ) | |
| ) | |
| } | |
| // Non-page segments are compared using the same function as the server | |
| return matchSegment(cachedSegment, currentSegment) | |
| } | |
| // ----------------------------------------------------------------------------- | |
| // The remainder of the module is a MinHeap implementation. Try not to put any | |
| // logic below here unless it's related to the heap algorithm. We can extract | |
| // this to a separate module if/when we need multiple kinds of heaps. | |
| // ----------------------------------------------------------------------------- | |
| function compareQueuePriority(a: PrefetchTask, b: PrefetchTask) { | |
| // Since the queue is a MinHeap, this should return a positive number if b is | |
| // higher priority than a, and a negative number if a is higher priority | |
| // than b. | |
| // `priority` is an integer, where higher numbers are higher priority. | |
| const priorityDiff = b.priority - a.priority | |
| if (priorityDiff !== 0) { | |
| return priorityDiff | |
| } | |
| // If the priority is the same, check which phase the prefetch is in — is it | |
| // prefetching the route tree, or the segments? Route trees are prioritized. | |
| const phaseDiff = b.phase - a.phase | |
| if (phaseDiff !== 0) { | |
| return phaseDiff | |
| } | |
| // Finally, check the insertion order. `sortId` is an incrementing counter | |
| // assigned to prefetches. We want to process the newest prefetches first. | |
| return b.sortId - a.sortId | |
| } | |
| function heapPush(heap: Array<PrefetchTask>, node: PrefetchTask): void { | |
| const index = heap.length | |
| heap.push(node) | |
| node._heapIndex = index | |
| heapSiftUp(heap, node, index) | |
| } | |
| function heapPeek(heap: Array<PrefetchTask>): PrefetchTask | null { | |
| return heap.length === 0 ? null : heap[0] | |
| } | |
| function heapPop(heap: Array<PrefetchTask>): PrefetchTask | null { | |
| if (heap.length === 0) { | |
| return null | |
| } | |
| const first = heap[0] | |
| first._heapIndex = -1 | |
| const last = heap.pop() as PrefetchTask | |
| if (last !== first) { | |
| heap[0] = last | |
| last._heapIndex = 0 | |
| heapSiftDown(heap, last, 0) | |
| } | |
| return first | |
| } | |
| function heapDelete(heap: Array<PrefetchTask>, node: PrefetchTask): void { | |
| const index = node._heapIndex | |
| if (index !== -1) { | |
| node._heapIndex = -1 | |
| if (heap.length !== 0) { | |
| const last = heap.pop() as PrefetchTask | |
| if (last !== node) { | |
| heap[index] = last | |
| last._heapIndex = index | |
| heapSiftDown(heap, last, index) | |
| } | |
| } | |
| } | |
| } | |
| function heapResift(heap: Array<PrefetchTask>, node: PrefetchTask): void { | |
| const index = node._heapIndex | |
| if (index !== -1) { | |
| if (index === 0) { | |
| heapSiftDown(heap, node, 0) | |
| } else { | |
| const parentIndex = (index - 1) >>> 1 | |
| const parent = heap[parentIndex] | |
| if (compareQueuePriority(parent, node) > 0) { | |
| // The parent is larger. Sift up. | |
| heapSiftUp(heap, node, index) | |
| } else { | |
| // The parent is smaller (or equal). Sift down. | |
| heapSiftDown(heap, node, index) | |
| } | |
| } | |
| } | |
| } | |
| function heapSiftUp( | |
| heap: Array<PrefetchTask>, | |
| node: PrefetchTask, | |
| i: number | |
| ): void { | |
| let index = i | |
| while (index > 0) { | |
| const parentIndex = (index - 1) >>> 1 | |
| const parent = heap[parentIndex] | |
| if (compareQueuePriority(parent, node) > 0) { | |
| // The parent is larger. Swap positions. | |
| heap[parentIndex] = node | |
| node._heapIndex = parentIndex | |
| heap[index] = parent | |
| parent._heapIndex = index | |
| index = parentIndex | |
| } else { | |
| // The parent is smaller. Exit. | |
| return | |
| } | |
| } | |
| } | |
| function heapSiftDown( | |
| heap: Array<PrefetchTask>, | |
| node: PrefetchTask, | |
| i: number | |
| ): void { | |
| let index = i | |
| const length = heap.length | |
| const halfLength = length >>> 1 | |
| while (index < halfLength) { | |
| const leftIndex = (index + 1) * 2 - 1 | |
| const left = heap[leftIndex] | |
| const rightIndex = leftIndex + 1 | |
| const right = heap[rightIndex] | |
| // If the left or right node is smaller, swap with the smaller of those. | |
| if (compareQueuePriority(left, node) < 0) { | |
| if (rightIndex < length && compareQueuePriority(right, left) < 0) { | |
| heap[index] = right | |
| right._heapIndex = index | |
| heap[rightIndex] = node | |
| node._heapIndex = rightIndex | |
| index = rightIndex | |
| } else { | |
| heap[index] = left | |
| left._heapIndex = index | |
| heap[leftIndex] = node | |
| node._heapIndex = leftIndex | |
| index = leftIndex | |
| } | |
| } else if (rightIndex < length && compareQueuePriority(right, node) < 0) { | |
| heap[index] = right | |
| right._heapIndex = index | |
| heap[rightIndex] = node | |
| node._heapIndex = rightIndex | |
| index = rightIndex | |
| } else { | |
| // Neither child is smaller. Exit. | |
| return | |
| } | |
| } | |
| } | |