/** * Copyright (c) Facebook, Inc. and its affiliates. * * This source code is licensed under the MIT license found in the * LICENSE file in the root directory of this source tree. * * @flow * @format */ 'use strict'; const invariant = require('invariant'); /** * Used to find the indices of the frames that overlap the given offsets. Useful for finding the * items that bound different windows of content, such as the visible area or the buffered overscan * area. */ function elementsThatOverlapOffsets( offsets: Array, itemCount: number, getFrameMetrics: ( index: number, ) => { length: number, offset: number, ... }, ): Array { const out = []; let outLength = 0; for (let ii = 0; ii < itemCount; ii++) { const frame = getFrameMetrics(ii); const trailingOffset = frame.offset + frame.length; for (let kk = 0; kk < offsets.length; kk++) { if (out[kk] == null && trailingOffset >= offsets[kk]) { out[kk] = ii; outLength++; if (kk === offsets.length - 1) { invariant( outLength === offsets.length, 'bad offsets input, should be in increasing order: %s', JSON.stringify(offsets), ); return out; } } } } return out; } /** * Computes the number of elements in the `next` range that are new compared to the `prev` range. * Handy for calculating how many new items will be rendered when the render window changes so we * can restrict the number of new items render at once so that content can appear on the screen * faster. */ function newRangeCount( prev: { first: number, last: number, ... }, next: { first: number, last: number, ... }, ): number { return ( next.last - next.first + 1 - Math.max( 0, 1 + Math.min(next.last, prev.last) - Math.max(next.first, prev.first), ) ); } /** * Custom logic for determining which items should be rendered given the current frame and scroll * metrics, as well as the previous render state. The algorithm may evolve over time, but generally * prioritizes the visible area first, then expands that with overscan regions ahead and behind, * biased in the direction of scroll. */ function computeWindowedRenderLimits( props: { data: any, getItemCount: (data: any) => number, maxToRenderPerBatch: number, windowSize: number, ... }, prev: { first: number, last: number, ... }, getFrameMetricsApprox: ( index: number, ) => { length: number, offset: number, ... }, scrollMetrics: { dt: number, offset: number, velocity: number, visibleLength: number, ... }, ): { first: number, last: number, ... } { const {data, getItemCount, maxToRenderPerBatch, windowSize} = props; const itemCount = getItemCount(data); if (itemCount === 0) { return prev; } const {offset, velocity, visibleLength} = scrollMetrics; // Start with visible area, then compute maximum overscan region by expanding from there, biased // in the direction of scroll. Total overscan area is capped, which should cap memory consumption // too. const visibleBegin = Math.max(0, offset); const visibleEnd = visibleBegin + visibleLength; const overscanLength = (windowSize - 1) * visibleLength; // Considering velocity seems to introduce more churn than it's worth. const leadFactor = 0.5; // Math.max(0, Math.min(1, velocity / 25 + 0.5)); const fillPreference = velocity > 1 ? 'after' : velocity < -1 ? 'before' : 'none'; const overscanBegin = Math.max( 0, visibleBegin - (1 - leadFactor) * overscanLength, ); const overscanEnd = Math.max(0, visibleEnd + leadFactor * overscanLength); const lastItemOffset = getFrameMetricsApprox(itemCount - 1).offset; if (lastItemOffset < overscanBegin) { // Entire list is before our overscan window return { first: Math.max(0, itemCount - 1 - maxToRenderPerBatch), last: itemCount - 1, }; } // Find the indices that correspond to the items at the render boundaries we're targeting. let [overscanFirst, first, last, overscanLast] = elementsThatOverlapOffsets( [overscanBegin, visibleBegin, visibleEnd, overscanEnd], props.getItemCount(props.data), getFrameMetricsApprox, ); overscanFirst = overscanFirst == null ? 0 : overscanFirst; first = first == null ? Math.max(0, overscanFirst) : first; overscanLast = overscanLast == null ? itemCount - 1 : overscanLast; last = last == null ? Math.min(overscanLast, first + maxToRenderPerBatch - 1) : last; const visible = {first, last}; // We want to limit the number of new cells we're rendering per batch so that we can fill the // content on the screen quickly. If we rendered the entire overscan window at once, the user // could be staring at white space for a long time waiting for a bunch of offscreen content to // render. let newCellCount = newRangeCount(prev, visible); while (true) { if (first <= overscanFirst && last >= overscanLast) { // If we fill the entire overscan range, we're done. break; } const maxNewCells = newCellCount >= maxToRenderPerBatch; const firstWillAddMore = first <= prev.first || first > prev.last; const firstShouldIncrement = first > overscanFirst && (!maxNewCells || !firstWillAddMore); const lastWillAddMore = last >= prev.last || last < prev.first; const lastShouldIncrement = last < overscanLast && (!maxNewCells || !lastWillAddMore); if (maxNewCells && !firstShouldIncrement && !lastShouldIncrement) { // We only want to stop if we've hit maxNewCells AND we cannot increment first or last // without rendering new items. This let's us preserve as many already rendered items as // possible, reducing render churn and keeping the rendered overscan range as large as // possible. break; } if ( firstShouldIncrement && !(fillPreference === 'after' && lastShouldIncrement && lastWillAddMore) ) { if (firstWillAddMore) { newCellCount++; } first--; } if ( lastShouldIncrement && !(fillPreference === 'before' && firstShouldIncrement && firstWillAddMore) ) { if (lastWillAddMore) { newCellCount++; } last++; } } if ( !( last >= first && first >= 0 && last < itemCount && first >= overscanFirst && last <= overscanLast && first <= visible.first && last >= visible.last ) ) { throw new Error( 'Bad window calculation ' + JSON.stringify({ first, last, itemCount, overscanFirst, overscanLast, visible, }), ); } return {first, last}; } const VirtualizeUtils = { computeWindowedRenderLimits, elementsThatOverlapOffsets, newRangeCount, }; module.exports = VirtualizeUtils;