tupali/librerias/gantt/code/modules/marker-clusters.src.js

1553 lines
76 KiB
JavaScript
Raw Permalink Normal View History

2020-05-23 20:45:54 +00:00
/**
* @license Highcharts JS v8.1.0 (2020-05-05)
*
* Marker clusters module for Highcharts
*
* (c) 2010-2019 Wojciech Chmiel
*
* License: www.highcharts.com/license
*/
'use strict';
(function (factory) {
if (typeof module === 'object' && module.exports) {
factory['default'] = factory;
module.exports = factory;
} else if (typeof define === 'function' && define.amd) {
define('highcharts/modules/marker-clusters', ['highcharts'], function (Highcharts) {
factory(Highcharts);
factory.Highcharts = Highcharts;
return factory;
});
} else {
factory(typeof Highcharts !== 'undefined' ? Highcharts : undefined);
}
}(function (Highcharts) {
var _modules = Highcharts ? Highcharts._modules : {};
function _registerModule(obj, path, args, fn) {
if (!obj.hasOwnProperty(path)) {
obj[path] = fn.apply(null, args);
}
}
_registerModule(_modules, 'modules/marker-clusters.src.js', [_modules['parts/Globals.js'], _modules['parts/Point.js'], _modules['parts/Utilities.js']], function (H, Point, U) {
/* *
*
* Marker clusters module.
*
* (c) 2010-2020 Torstein Honsi
*
* Author: Wojciech Chmiel
*
* License: www.highcharts.com/license
*
* !!!!!!! SOURCE GETS TRANSPILED BY TYPESCRIPT. EDIT TS FILE ONLY. !!!!!!!
*
* */
var __read = (this && this.__read) || function (o, n) {
var m = typeof Symbol === "function" && o[Symbol.iterator];
if (!m) return o;
var i = m.call(o), r, ar = [], e;
try {
while ((n === void 0 || n-- > 0) && !(r = i.next()).done) ar.push(r.value);
}
catch (error) { e = { error: error }; }
finally {
try {
if (r && !r.done && (m = i["return"])) m.call(i);
}
finally { if (e) throw e.error; }
}
return ar;
};
/**
* Function callback when a cluster is clicked.
*
* @callback Highcharts.MarkerClusterDrillCallbackFunction
*
* @param {Highcharts.Point} this
* The point where the event occured.
*
* @param {Highcharts.PointClickEventObject} event
* Event arguments.
*/
''; // detach doclets from following code
var addEvent = U.addEvent, animObject = U.animObject, defined = U.defined, error = U.error, isArray = U.isArray, isFunction = U.isFunction, isObject = U.isObject, isNumber = U.isNumber, merge = U.merge, objectEach = U.objectEach, relativeLength = U.relativeLength, syncTimeout = U.syncTimeout;
/* eslint-disable no-invalid-this */
var Series = H.Series, Scatter = H.seriesTypes.scatter, SvgRenderer = H.SVGRenderer, baseGeneratePoints = Series.prototype.generatePoints, stateIdCounter = 0,
// Points that ids are included in the oldPointsStateId array
// are hidden before animation. Other ones are destroyed.
oldPointsStateId = [];
/**
* Options for marker clusters, the concept of sampling the data
* values into larger blocks in order to ease readability and
* increase performance of the JavaScript charts.
*
* Note: marker clusters module is not working with `boost`
* and `draggable-points` modules.
*
* The marker clusters feature requires the marker-clusters.js
* file to be loaded, found in the modules directory of the download
* package, or online at [code.highcharts.com/modules/marker-clusters.js
* ](code.highcharts.com/modules/marker-clusters.js).
*
* @sample maps/marker-clusters/europe
* Maps marker clusters
* @sample highcharts/marker-clusters/basic
* Scatter marker clusters
* @sample maps/marker-clusters/optimized-kmeans
* Marker clusters with colorAxis
*
* @product highcharts highmaps
* @since 8.0.0
* @optionparent plotOptions.scatter.cluster
*
* @private
*/
var clusterDefaultOptions = {
/**
* Whether to enable the marker-clusters module.
*
* @sample maps/marker-clusters/basic
* Maps marker clusters
* @sample highcharts/marker-clusters/basic
* Scatter marker clusters
*/
enabled: false,
/**
* When set to `false` prevent cluster overlapping - this option
* works only when `layoutAlgorithm.type = "grid"`.
*
* @sample highcharts/marker-clusters/grid
* Prevent overlapping
*/
allowOverlap: true,
/**
* Options for the cluster marker animation.
* @type {boolean|Highcharts.AnimationOptionsObject}
* @default { "duration": 500 }
*/
animation: {
/** @ignore-option */
duration: 500
},
/**
* Zoom the plot area to the cluster points range when a cluster is clicked.
*/
drillToCluster: true,
/**
* The minimum amount of points to be combined into a cluster.
* This value has to be greater or equal to 2.
*
* @sample highcharts/marker-clusters/basic
* At least three points in the cluster
*/
minimumClusterSize: 2,
/**
* Options for layout algorithm. Inside there
* are options to change the type of the algorithm, gridSize,
* distance or iterations.
*/
layoutAlgorithm: {
/**
* Type of the algorithm used to combine points into a cluster.
* There are three available algorithms:
*
* 1) `grid` - grid-based clustering technique. Points are assigned
* to squares of set size depending on their position on the plot
* area. Points inside the grid square are combined into a cluster.
* The grid size can be controlled by `gridSize` property
* (grid size changes at certain zoom levels).
*
* 2) `kmeans` - based on K-Means clustering technique. In the
* first step, points are divided using the grid method (distance
* property is a grid size) to find the initial amount of clusters.
* Next, each point is classified by computing the distance between
* each cluster center and that point. When the closest cluster
* distance is lower than distance property set by a user the point
* is added to this cluster otherwise is classified as `noise`. The
* algorithm is repeated until each cluster center not change its
* previous position more than one pixel. This technique is more
* accurate but also more time consuming than the `grid` algorithm,
* especially for big datasets.
*
* 3) `optimizedKmeans` - based on K-Means clustering technique. This
* algorithm uses k-means algorithm only on the chart initialization
* or when chart extremes have greater range than on initialization.
* When a chart is redrawn the algorithm checks only clustered points
* distance from the cluster center and rebuild it when the point is
* spaced enough to be outside the cluster. It provides performance
* improvement and more stable clusters position yet can be used rather
* on small and sparse datasets.
*
* By default, the algorithm depends on visible quantity of points
* and `kmeansThreshold`. When there are more visible points than the
* `kmeansThreshold` the `grid` algorithm is used, otherwise `kmeans`.
*
* The custom clustering algorithm can be added by assigning a callback
* function as the type property. This function takes an array of
* `processedXData`, `processedYData`, `processedXData` indexes and
* `layoutAlgorithm` options as arguments and should return an object
* with grouped data.
*
* The algorithm should return an object like that:
* <pre>{
* clusterId1: [{
* x: 573,
* y: 285,
* index: 1 // point index in the data array
* }, {
* x: 521,
* y: 197,
* index: 2
* }],
* clusterId2: [{
* ...
* }]
* ...
* }</pre>
*
* `clusterId` (example above - unique id of a cluster or noise)
* is an array of points belonging to a cluster. If the
* array has only one point or fewer points than set in
* `cluster.minimumClusterSize` it won't be combined into a cluster.
*
* @sample maps/marker-clusters/optimized-kmeans
* Optimized K-Means algorithm
* @sample highcharts/marker-clusters/kmeans
* K-Means algorithm
* @sample highcharts/marker-clusters/grid
* Grid algorithm
* @sample maps/marker-clusters/custom-alg
* Custom algorithm
*
* @type {string|Function}
* @see [cluster.minimumClusterSize](#plotOptions.scatter.marker.cluster.minimumClusterSize)
* @apioption plotOptions.scatter.cluster.layoutAlgorithm.type
*/
/**
* When `type` is set to the `grid`,
* `gridSize` is a size of a grid square element either as a number
* defining pixels, or a percentage defining a percentage
* of the plot area width.
*
* @type {number|string}
*/
gridSize: 50,
/**
* When `type` is set to `kmeans`,
* `iterations` are the number of iterations that this algorithm will be
* repeated to find clusters positions.
*
* @type {number}
* @apioption plotOptions.scatter.cluster.layoutAlgorithm.iterations
*/
/**
* When `type` is set to `kmeans`,
* `distance` is a maximum distance between point and cluster center
* so that this point will be inside the cluster. The distance
* is either a number defining pixels or a percentage
* defining a percentage of the plot area width.
*
* @type {number|string}
*/
distance: 40,
/**
* When `type` is set to `undefined` and there are more visible points
* than the kmeansThreshold the `grid` algorithm is used to find
* clusters, otherwise `kmeans`. It ensures good performance on
* large datasets and better clusters arrangement after the zoom.
*/
kmeansThreshold: 100
},
/**
* Options for the cluster marker.
* @extends plotOptions.series.marker
* @excluding enabledThreshold, states
* @type {Highcharts.PointMarkerOptionsObject}
*/
marker: {
/** @internal */
symbol: 'cluster',
/** @internal */
radius: 15,
/** @internal */
lineWidth: 0,
/** @internal */
lineColor: '#ffffff'
},
/**
* Fires when the cluster point is clicked and `drillToCluster` is enabled.
* One parameter, `event`, is passed to the function. The default action
* is to zoom to the cluster points range. This can be prevented
* by calling `event.preventDefault()`.
*
* @type {Highcharts.MarkerClusterDrillCallbackFunction}
* @product highcharts highmaps
* @see [cluster.drillToCluster](#plotOptions.scatter.marker.cluster.drillToCluster)
* @apioption plotOptions.scatter.cluster.events.drillToCluster
*/
/**
* An array defining zones within marker clusters.
*
* In styled mode, the color zones are styled with the
* `.highcharts-cluster-zone-{n}` class, or custom
* classed from the `className`
* option.
*
* @sample highcharts/marker-clusters/basic
* Marker clusters zones
* @sample maps/marker-clusters/custom-alg
* Zones on maps
*
* @type {Array<*>}
* @product highcharts highmaps
* @apioption plotOptions.scatter.cluster.zones
*/
/**
* Styled mode only. A custom class name for the zone.
*
* @sample highcharts/css/color-zones/
* Zones styled by class name
*
* @type {string}
* @apioption plotOptions.scatter.cluster.zones.className
*/
/**
* Settings for the cluster marker belonging to the zone.
*
* @see [cluster.marker](#plotOptions.scatter.cluster.marker)
* @extends plotOptions.scatter.cluster.marker
* @product highcharts highmaps
* @apioption plotOptions.scatter.cluster.zones.marker
*/
/**
* The value where the zone starts.
*
* @type {number}
* @product highcharts highmaps
* @apioption plotOptions.scatter.cluster.zones.from
*/
/**
* The value where the zone ends.
*
* @type {number}
* @product highcharts highmaps
* @apioption plotOptions.scatter.cluster.zones.to
*/
/**
* The fill color of the cluster marker in hover state. When
* `undefined`, the series' or point's fillColor for normal
* state is used.
*
* @type {Highcharts.ColorType}
* @apioption plotOptions.scatter.cluster.states.hover.fillColor
*/
/**
* Options for the cluster data labels.
* @type {Highcharts.DataLabelsOptions}
*/
dataLabels: {
/** @internal */
enabled: true,
/** @internal */
format: '{point.clusterPointsAmount}',
/** @internal */
verticalAlign: 'middle',
/** @internal */
align: 'center',
/** @internal */
style: {
color: 'contrast'
},
/** @internal */
inside: true
}
};
(H.defaultOptions.plotOptions || {}).series = merge((H.defaultOptions.plotOptions || {}).series, {
cluster: clusterDefaultOptions,
tooltip: {
/**
* The HTML of the cluster point's in the tooltip. Works only with
* marker-clusters module and analogously to
* [pointFormat](#tooltip.pointFormat).
*
* The cluster tooltip can be also formatted using
* `tooltip.formatter` callback function and `point.isCluster` flag.
*
* @sample highcharts/marker-clusters/grid
* Format tooltip for cluster points.
*
* @sample maps/marker-clusters/europe/
* Format tooltip for clusters using tooltip.formatter
*
* @apioption tooltip.clusterFormat
*/
clusterFormat: '<span>Clustered points: ' +
'{point.clusterPointsAmount}</span><br/>'
}
});
// Utils.
/* eslint-disable require-jsdoc */
function getClusterPosition(points) {
var pointsLen = points.length, sumX = 0, sumY = 0, i;
for (i = 0; i < pointsLen; i++) {
sumX += points[i].x;
sumY += points[i].y;
}
return {
x: sumX / pointsLen,
y: sumY / pointsLen
};
}
// Prepare array with sorted data objects to be
// compared in getPointsState method.
function getDataState(clusteredData, stateDataLen) {
var state = [];
state.length = stateDataLen;
clusteredData.clusters.forEach(function (cluster) {
cluster.data.forEach(function (elem) {
state[elem.dataIndex] = elem;
});
});
clusteredData.noise.forEach(function (noise) {
state[noise.data[0].dataIndex] = noise.data[0];
});
return state;
}
function fadeInElement(elem, opacity, animation) {
elem
.attr({
opacity: opacity
})
.animate({
opacity: 1
}, animation);
}
function fadeInStatePoint(stateObj, opacity, animation, fadeinGraphic, fadeinDataLabel) {
if (stateObj.point) {
if (fadeinGraphic && stateObj.point.graphic) {
stateObj.point.graphic.show();
fadeInElement(stateObj.point.graphic, opacity, animation);
}
if (fadeinDataLabel && stateObj.point.dataLabel) {
stateObj.point.dataLabel.show();
fadeInElement(stateObj.point.dataLabel, opacity, animation);
}
}
}
function hideStatePoint(stateObj, hideGraphic, hideDataLabel) {
if (stateObj.point) {
if (hideGraphic && stateObj.point.graphic) {
stateObj.point.graphic.hide();
}
if (hideDataLabel && stateObj.point.dataLabel) {
stateObj.point.dataLabel.hide();
}
}
}
function destroyOldPoints(oldState) {
if (oldState) {
objectEach(oldState, function (state) {
if (state.point && state.point.destroy) {
state.point.destroy();
}
});
}
}
function fadeInNewPointAndDestoryOld(newPointObj, oldPoints, animation, opacity) {
// Fade in new point.
fadeInStatePoint(newPointObj, opacity, animation, true, true);
// Destroy old animated points.
oldPoints.forEach(function (p) {
if (p.point && p.point.destroy) {
p.point.destroy();
}
});
}
// Generate unique stateId for a state element.
function getStateId() {
return Math.random().toString(36).substring(2, 7) + '-' + stateIdCounter++;
}
// Useful for debugging.
// function drawGridLines(
// series: Highcharts.Series,
// options: Highcharts.MarkerClusterLayoutAlgorithmOptions
// ): void {
// var chart = series.chart,
// xAxis = series.xAxis,
// yAxis = series.yAxis,
// xAxisLen = series.xAxis.len,
// yAxisLen = series.yAxis.len,
// i, j, elem, text,
// currentX = 0,
// currentY = 0,
// scaledGridSize = 50,
// gridX = 0,
// gridY = 0,
// gridOffset = series.getGridOffset(),
// mapXSize, mapYSize;
// if (series.debugGridLines && series.debugGridLines.length) {
// series.debugGridLines.forEach(function (gridItem): void {
// if (gridItem && gridItem.destroy) {
// gridItem.destroy();
// }
// });
// }
// series.debugGridLines = [];
// scaledGridSize = series.getScaledGridSize(options);
// mapXSize = Math.abs(
// xAxis.toPixels(xAxis.dataMax || 0) -
// xAxis.toPixels(xAxis.dataMin || 0)
// );
// mapYSize = Math.abs(
// yAxis.toPixels(yAxis.dataMax || 0) -
// yAxis.toPixels(yAxis.dataMin || 0)
// );
// gridX = Math.ceil(mapXSize / scaledGridSize);
// gridY = Math.ceil(mapYSize / scaledGridSize);
// for (i = 0; i < gridX; i++) {
// currentX = i * scaledGridSize;
// if (
// gridOffset.plotLeft + currentX >= 0 &&
// gridOffset.plotLeft + currentX < xAxisLen
// ) {
// for (j = 0; j < gridY; j++) {
// currentY = j * scaledGridSize;
// if (
// gridOffset.plotTop + currentY >= 0 &&
// gridOffset.plotTop + currentY < yAxisLen
// ) {
// if (j % 2 === 0 && i % 2 === 0) {
// var rect = chart.renderer
// .rect(
// gridOffset.plotLeft + currentX,
// gridOffset.plotTop + currentY,
// scaledGridSize * 2,
// scaledGridSize * 2
// )
// .attr({
// stroke: series.color,
// 'stroke-width': '2px'
// })
// .add()
// .toFront();
// series.debugGridLines.push(rect);
// }
// elem = chart.renderer
// .rect(
// gridOffset.plotLeft + currentX,
// gridOffset.plotTop + currentY,
// scaledGridSize,
// scaledGridSize
// )
// .attr({
// stroke: series.color,
// opacity: 0.3,
// 'stroke-width': '1px'
// })
// .add()
// .toFront();
// text = chart.renderer
// .text(
// j + '-' + i,
// gridOffset.plotLeft + currentX + 2,
// gridOffset.plotTop + currentY + 7
// )
// .css({
// fill: 'rgba(0, 0, 0, 0.7)',
// fontSize: '7px'
// })
// .add()
// .toFront();
// series.debugGridLines.push(elem);
// series.debugGridLines.push(text);
// }
// }
// }
// }
// }
/* eslint-enable require-jsdoc */
// Cluster symbol.
SvgRenderer.prototype.symbols.cluster = function (x, y, width, height) {
var w = width / 2, h = height / 2, outerWidth = 1, space = 1, inner, outer1, outer2;
inner = this.arc(x + w, y + h, w - space * 4, h - space * 4, {
start: Math.PI * 0.5,
end: Math.PI * 2.5,
open: false
});
outer1 = this.arc(x + w, y + h, w - space * 3, h - space * 3, {
start: Math.PI * 0.5,
end: Math.PI * 2.5,
innerR: w - outerWidth * 2,
open: false
});
outer2 = this.arc(x + w, y + h, w - space, h - space, {
start: Math.PI * 0.5,
end: Math.PI * 2.5,
innerR: w,
open: false
});
return outer2.concat(outer1, inner);
};
Scatter.prototype.animateClusterPoint = function (clusterObj) {
var series = this, xAxis = series.xAxis, yAxis = series.yAxis, chart = series.chart, clusterOptions = series.options.cluster, animation = animObject((clusterOptions || {}).animation), animDuration = animation.duration || 500, pointsState = (series.markerClusterInfo || {}).pointsState, newState = (pointsState || {}).newState, oldState = (pointsState || {}).oldState, parentId, oldPointObj, newPointObj, oldPoints = [], newPointBBox, offset = 0, newX = 0, newY = 0, isOldPointGrahic = false, isCbHandled = false;
if (oldState && newState) {
newPointObj = newState[clusterObj.stateId];
newX = xAxis.toPixels(newPointObj.x) - chart.plotLeft;
newY = yAxis.toPixels(newPointObj.y) - chart.plotTop;
// Point has one ancestor.
if (newPointObj.parentsId.length === 1) {
parentId = (newState || {})[clusterObj.stateId].parentsId[0];
oldPointObj = oldState[parentId];
// If old and new poistions are the same do not animate.
if (newPointObj.point &&
newPointObj.point.graphic &&
oldPointObj &&
oldPointObj.point &&
oldPointObj.point.plotX &&
oldPointObj.point.plotY &&
oldPointObj.point.plotX !== newPointObj.point.plotX &&
oldPointObj.point.plotY !== newPointObj.point.plotY) {
newPointBBox = newPointObj.point.graphic.getBBox();
offset = newPointBBox.width / 2;
newPointObj.point.graphic.attr({
x: oldPointObj.point.plotX - offset,
y: oldPointObj.point.plotY - offset
});
newPointObj.point.graphic.animate({
x: newX - newPointObj.point.graphic.radius,
y: newY - newPointObj.point.graphic.radius
}, animation, function () {
isCbHandled = true;
// Destroy old point.
if (oldPointObj.point && oldPointObj.point.destroy) {
oldPointObj.point.destroy();
}
});
// Data label animation.
if (newPointObj.point.dataLabel &&
newPointObj.point.dataLabel.alignAttr &&
oldPointObj.point.dataLabel &&
oldPointObj.point.dataLabel.alignAttr) {
newPointObj.point.dataLabel.attr({
x: oldPointObj.point.dataLabel.alignAttr.x,
y: oldPointObj.point.dataLabel.alignAttr.y
});
newPointObj.point.dataLabel.animate({
x: newPointObj.point.dataLabel.alignAttr.x,
y: newPointObj.point.dataLabel.alignAttr.y
}, animation);
}
}
}
else if (newPointObj.parentsId.length === 0) {
// Point has no ancestors - new point.
// Hide new point.
hideStatePoint(newPointObj, true, true);
syncTimeout(function () {
// Fade in new point.
fadeInStatePoint(newPointObj, 0.1, animation, true, true);
}, animDuration / 2);
}
else {
// Point has many ancestors.
// Hide new point before animation.
hideStatePoint(newPointObj, true, true);
newPointObj.parentsId.forEach(function (elem) {
if (oldState && oldState[elem]) {
oldPointObj = oldState[elem];
oldPoints.push(oldPointObj);
if (oldPointObj.point &&
oldPointObj.point.graphic) {
isOldPointGrahic = true;
oldPointObj.point.graphic.show();
oldPointObj.point.graphic.animate({
x: newX - oldPointObj.point.graphic.radius,
y: newY - oldPointObj.point.graphic.radius,
opacity: 0.4
}, animation, function () {
isCbHandled = true;
fadeInNewPointAndDestoryOld(newPointObj, oldPoints, animation, 0.7);
});
if (oldPointObj.point.dataLabel &&
oldPointObj.point.dataLabel.y !== -9999 &&
newPointObj.point &&
newPointObj.point.dataLabel &&
newPointObj.point.dataLabel.alignAttr) {
oldPointObj.point.dataLabel.show();
oldPointObj.point.dataLabel.animate({
x: newPointObj.point.dataLabel.alignAttr.x,
y: newPointObj.point.dataLabel.alignAttr.y,
opacity: 0.4
}, animation);
}
}
}
});
// Make sure point is faded in.
syncTimeout(function () {
if (!isCbHandled) {
fadeInNewPointAndDestoryOld(newPointObj, oldPoints, animation, 0.85);
}
}, animDuration);
if (!isOldPointGrahic) {
syncTimeout(function () {
fadeInNewPointAndDestoryOld(newPointObj, oldPoints, animation, 0.1);
}, animDuration / 2);
}
}
}
};
Scatter.prototype.getGridOffset = function () {
var series = this, chart = series.chart, xAxis = series.xAxis, yAxis = series.yAxis, plotLeft = 0, plotTop = 0;
if (series.dataMinX && series.dataMaxX) {
plotLeft = xAxis.reversed ?
xAxis.toPixels(series.dataMaxX) : xAxis.toPixels(series.dataMinX);
}
else {
plotLeft = chart.plotLeft;
}
if (series.dataMinY && series.dataMaxY) {
plotTop = yAxis.reversed ?
yAxis.toPixels(series.dataMinY) : yAxis.toPixels(series.dataMaxY);
}
else {
plotTop = chart.plotTop;
}
return { plotLeft: plotLeft, plotTop: plotTop };
};
Scatter.prototype.getScaledGridSize = function (options) {
var series = this, xAxis = series.xAxis, search = true, k = 1, divider = 1, processedGridSize = options.processedGridSize ||
clusterDefaultOptions.layoutAlgorithm.gridSize, gridSize, scale, level;
if (!series.gridValueSize) {
series.gridValueSize = Math.abs(xAxis.toValue(processedGridSize) - xAxis.toValue(0));
}
gridSize = xAxis.toPixels(series.gridValueSize) - xAxis.toPixels(0);
scale = +(processedGridSize / gridSize).toFixed(14);
// Find the level and its divider.
while (search && scale !== 1) {
level = Math.pow(2, k);
if (scale > 0.75 && scale < 1.25) {
search = false;
}
else if (scale >= (1 / level) && scale < 2 * (1 / level)) {
search = false;
divider = level;
}
else if (scale <= level && scale > level / 2) {
search = false;
divider = 1 / level;
}
k++;
}
return (processedGridSize / divider) / scale;
};
Scatter.prototype.getRealExtremes = function () {
var _a, _b;
var series = this, chart = series.chart, xAxis = series.xAxis, yAxis = series.yAxis, realMinX = xAxis ? xAxis.toValue(chart.plotLeft) : 0, realMaxX = xAxis ?
xAxis.toValue(chart.plotLeft + chart.plotWidth) : 0, realMinY = yAxis ? yAxis.toValue(chart.plotTop) : 0, realMaxY = yAxis ?
yAxis.toValue(chart.plotTop + chart.plotHeight) : 0;
if (realMinX > realMaxX) {
_a = __read([realMinX, realMaxX], 2), realMaxX = _a[0], realMinX = _a[1];
}
if (realMinY > realMaxY) {
_b = __read([realMinY, realMaxY], 2), realMaxY = _b[0], realMinY = _b[1];
}
return {
minX: realMinX,
maxX: realMaxX,
minY: realMinY,
maxY: realMaxY
};
};
Scatter.prototype.onDrillToCluster = function (event) {
var point = event.point || event.target;
point.firePointEvent('drillToCluster', event, function (e) {
var _a, _b;
var point = e.point || e.target, series = point.series, xAxis = point.series.xAxis, yAxis = point.series.yAxis, chart = point.series.chart, clusterOptions = series.options.cluster, drillToCluster = (clusterOptions || {}).drillToCluster, offsetX, offsetY, sortedDataX, sortedDataY, minX, minY, maxX, maxY;
if (drillToCluster && point.clusteredData) {
sortedDataX = point.clusteredData.map(function (data) {
return data.x;
}).sort(function (a, b) { return a - b; });
sortedDataY = point.clusteredData.map(function (data) {
return data.y;
}).sort(function (a, b) { return a - b; });
minX = sortedDataX[0];
maxX = sortedDataX[sortedDataX.length - 1];
minY = sortedDataY[0];
maxY = sortedDataY[sortedDataY.length - 1];
offsetX = Math.abs((maxX - minX) * 0.1);
offsetY = Math.abs((maxY - minY) * 0.1);
chart.pointer.zoomX = true;
chart.pointer.zoomY = true;
// Swap when minus values.
if (minX > maxX) {
_a = __read([maxX, minX], 2), minX = _a[0], maxX = _a[1];
}
if (minY > maxY) {
_b = __read([maxY, minY], 2), minY = _b[0], maxY = _b[1];
}
chart.zoom({
originalEvent: e,
xAxis: [{
axis: xAxis,
min: minX - offsetX,
max: maxX + offsetX
}],
yAxis: [{
axis: yAxis,
min: minY - offsetY,
max: maxY + offsetY
}]
});
}
});
};
Scatter.prototype.getClusterDistancesFromPoint = function (clusters, pointX, pointY) {
var series = this, xAxis = series.xAxis, yAxis = series.yAxis, pointClusterDistance = [], j, distance;
for (j = 0; j < clusters.length; j++) {
distance = Math.sqrt(Math.pow(xAxis.toPixels(pointX) -
xAxis.toPixels(clusters[j].posX), 2) +
Math.pow(yAxis.toPixels(pointY) -
yAxis.toPixels(clusters[j].posY), 2));
pointClusterDistance.push({
clusterIndex: j,
distance: distance
});
}
return pointClusterDistance.sort(function (a, b) { return a.distance - b.distance; });
};
// Point state used when animation is enabled to compare
// and bind old points with new ones.
Scatter.prototype.getPointsState = function (clusteredData, oldMarkerClusterInfo, dataLength) {
var oldDataStateArr = oldMarkerClusterInfo ?
getDataState(oldMarkerClusterInfo, dataLength) : [], newDataStateArr = getDataState(clusteredData, dataLength), state = {}, newState, oldState, i;
// Clear global array before populate with new ids.
oldPointsStateId = [];
// Build points state structure.
clusteredData.clusters.forEach(function (cluster) {
state[cluster.stateId] = {
x: cluster.x,
y: cluster.y,
id: cluster.stateId,
point: cluster.point,
parentsId: []
};
});
clusteredData.noise.forEach(function (noise) {
state[noise.stateId] = {
x: noise.x,
y: noise.y,
id: noise.stateId,
point: noise.point,
parentsId: []
};
});
// Bind new and old state.
for (i = 0; i < newDataStateArr.length; i++) {
newState = newDataStateArr[i];
oldState = oldDataStateArr[i];
if (newState &&
oldState &&
newState.parentStateId &&
oldState.parentStateId &&
state[newState.parentStateId] &&
state[newState.parentStateId].parentsId.indexOf(oldState.parentStateId) === -1) {
state[newState.parentStateId].parentsId.push(oldState.parentStateId);
if (oldPointsStateId.indexOf(oldState.parentStateId) === -1) {
oldPointsStateId.push(oldState.parentStateId);
}
}
}
return state;
};
Scatter.prototype.markerClusterAlgorithms = {
grid: function (dataX, dataY, dataIndexes, options) {
var series = this, xAxis = series.xAxis, yAxis = series.yAxis, grid = {}, gridOffset = series.getGridOffset(), scaledGridSize, x, y, gridX, gridY, key, i;
// drawGridLines(series, options);
scaledGridSize = series.getScaledGridSize(options);
for (i = 0; i < dataX.length; i++) {
x = xAxis.toPixels(dataX[i]) - gridOffset.plotLeft;
y = yAxis.toPixels(dataY[i]) - gridOffset.plotTop;
gridX = Math.floor(x / scaledGridSize);
gridY = Math.floor(y / scaledGridSize);
key = gridY + '-' + gridX;
if (!grid[key]) {
grid[key] = [];
}
grid[key].push({
dataIndex: dataIndexes[i],
x: dataX[i],
y: dataY[i]
});
}
return grid;
},
kmeans: function (dataX, dataY, dataIndexes, options) {
var series = this, clusters = [], noise = [], group = {}, pointMaxDistance = options.processedDistance ||
clusterDefaultOptions.layoutAlgorithm.distance, iterations = options.iterations,
// Max pixel difference beetwen new and old cluster position.
maxClusterShift = 1, currentIteration = 0, repeat = true, pointX = 0, pointY = 0, tempPos, pointClusterDistance = [], groupedData, key, i, j;
options.processedGridSize = options.processedDistance;
// Use grid method to get groupedData object.
groupedData = series.markerClusterAlgorithms ?
series.markerClusterAlgorithms.grid.call(series, dataX, dataY, dataIndexes, options) : {};
// Find clusters amount and its start positions
// based on grid grouped data.
for (key in groupedData) {
if (groupedData[key].length > 1) {
tempPos = getClusterPosition(groupedData[key]);
clusters.push({
posX: tempPos.x,
posY: tempPos.y,
oldX: 0,
oldY: 0,
startPointsLen: groupedData[key].length,
points: []
});
}
}
// Start kmeans iteration process.
while (repeat) {
clusters.map(function (c) {
c.points.length = 0;
return c;
});
noise.length = 0;
for (i = 0; i < dataX.length; i++) {
pointX = dataX[i];
pointY = dataY[i];
pointClusterDistance = series.getClusterDistancesFromPoint(clusters, pointX, pointY);
if (pointClusterDistance.length &&
pointClusterDistance[0].distance < pointMaxDistance) {
clusters[pointClusterDistance[0].clusterIndex].points.push({
x: pointX,
y: pointY,
dataIndex: dataIndexes[i]
});
}
else {
noise.push({
x: pointX,
y: pointY,
dataIndex: dataIndexes[i]
});
}
}
// When cluster points array has only one point the
// point should be classified again.
for (j = 0; j < clusters.length; j++) {
if (clusters[j].points.length === 1) {
pointClusterDistance = series.getClusterDistancesFromPoint(clusters, clusters[j].points[0].x, clusters[j].points[0].y);
if (pointClusterDistance[1].distance < pointMaxDistance) {
// Add point to the next closest cluster.
clusters[pointClusterDistance[1].clusterIndex].points
.push(clusters[j].points[0]);
// Clear points array.
clusters[pointClusterDistance[0].clusterIndex]
.points.length = 0;
}
}
}
// Compute a new clusters position and check if it
// is different than the old one.
repeat = false;
for (j = 0; j < clusters.length; j++) {
tempPos = getClusterPosition(clusters[j].points);
clusters[j].oldX = clusters[j].posX;
clusters[j].oldY = clusters[j].posY;
clusters[j].posX = tempPos.x;
clusters[j].posY = tempPos.y;
// Repeat the algorithm if at least one cluster
// is shifted more than maxClusterShift property.
if (clusters[j].posX > clusters[j].oldX + maxClusterShift ||
clusters[j].posX < clusters[j].oldX - maxClusterShift ||
clusters[j].posY > clusters[j].oldY + maxClusterShift ||
clusters[j].posY < clusters[j].oldY - maxClusterShift) {
repeat = true;
}
}
// If iterations property is set repeat the algorithm
// specified amount of times.
if (iterations) {
repeat = currentIteration < iterations - 1;
}
currentIteration++;
}
clusters.forEach(function (cluster, i) {
group['cluster' + i] = cluster.points;
});
noise.forEach(function (noise, i) {
group['noise' + i] = [noise];
});
return group;
},
optimizedKmeans: function (processedXData, processedYData, dataIndexes, options) {
var series = this, xAxis = series.xAxis, yAxis = series.yAxis, pointMaxDistance = options.processedDistance ||
clusterDefaultOptions.layoutAlgorithm.gridSize, group = {}, extremes = series.getRealExtremes(), clusterMarkerOptions = (series.options.cluster || {}).marker, offset, distance, radius;
if (!series.markerClusterInfo || (series.initMaxX && series.initMaxX < extremes.maxX ||
series.initMinX && series.initMinX > extremes.minX ||
series.initMaxY && series.initMaxY < extremes.maxY ||
series.initMinY && series.initMinY > extremes.minY)) {
series.initMaxX = extremes.maxX;
series.initMinX = extremes.minX;
series.initMaxY = extremes.maxY;
series.initMinY = extremes.minY;
group = series.markerClusterAlgorithms ?
series.markerClusterAlgorithms.kmeans.call(series, processedXData, processedYData, dataIndexes, options) : {};
series.baseClusters = null;
}
else {
if (!series.baseClusters) {
series.baseClusters = {
clusters: series.markerClusterInfo.clusters,
noise: series.markerClusterInfo.noise
};
}
series.baseClusters.clusters.forEach(function (cluster) {
cluster.pointsOutside = [];
cluster.pointsInside = [];
cluster.data.forEach(function (dataPoint) {
distance = Math.sqrt(Math.pow(xAxis.toPixels(dataPoint.x) -
xAxis.toPixels(cluster.x), 2) +
Math.pow(yAxis.toPixels(dataPoint.y) -
yAxis.toPixels(cluster.y), 2));
if (cluster.clusterZone &&
cluster.clusterZone.marker &&
cluster.clusterZone.marker.radius) {
radius = cluster.clusterZone.marker.radius;
}
else if (clusterMarkerOptions &&
clusterMarkerOptions.radius) {
radius = clusterMarkerOptions.radius;
}
else {
radius = clusterDefaultOptions.marker.radius;
}
offset = pointMaxDistance - radius >= 0 ?
pointMaxDistance - radius : radius;
if (distance > radius + offset &&
defined(cluster.pointsOutside)) {
cluster.pointsOutside.push(dataPoint);
}
else if (defined(cluster.pointsInside)) {
cluster.pointsInside.push(dataPoint);
}
});
if (cluster.pointsInside.length) {
group[cluster.id] = cluster.pointsInside;
}
cluster.pointsOutside.forEach(function (p, i) {
group[cluster.id + '_noise' + i] = [p];
});
});
series.baseClusters.noise.forEach(function (noise) {
group[noise.id] = noise.data;
});
}
return group;
}
};
Scatter.prototype.preventClusterCollisions = function (props) {
var series = this, xAxis = series.xAxis, yAxis = series.yAxis, _a = __read(props.key.split('-').map(parseFloat), 2), gridY = _a[0], gridX = _a[1], gridSize = props.gridSize, groupedData = props.groupedData, defaultRadius = props.defaultRadius, clusterRadius = props.clusterRadius, gridXPx = gridX * gridSize, gridYPx = gridY * gridSize, xPixel = xAxis.toPixels(props.x), yPixel = yAxis.toPixels(props.y), gridsToCheckCollision = [], pointsLen = 0, radius = 0, clusterMarkerOptions = (series.options.cluster || {}).marker, zoneOptions = (series.options.cluster || {}).zones, gridOffset = series.getGridOffset(), nextXPixel, nextYPixel, signX, signY, cornerGridX, cornerGridY, i, j, itemX, itemY, nextClusterPos, maxDist, keys, x, y;
// Distance to the grid start.
xPixel -= gridOffset.plotLeft;
yPixel -= gridOffset.plotTop;
for (i = 1; i < 5; i++) {
signX = i % 2 ? -1 : 1;
signY = i < 3 ? -1 : 1;
cornerGridX = Math.floor((xPixel + signX * clusterRadius) / gridSize);
cornerGridY = Math.floor((yPixel + signY * clusterRadius) / gridSize);
keys = [
cornerGridY + '-' + cornerGridX,
cornerGridY + '-' + gridX,
gridY + '-' + cornerGridX
];
for (j = 0; j < keys.length; j++) {
if (gridsToCheckCollision.indexOf(keys[j]) === -1 &&
keys[j] !== props.key) {
gridsToCheckCollision.push(keys[j]);
}
}
}
gridsToCheckCollision.forEach(function (item) {
var _a;
if (groupedData[item]) {
// Cluster or noise position is already computed.
if (!groupedData[item].posX) {
nextClusterPos = getClusterPosition(groupedData[item]);
groupedData[item].posX = nextClusterPos.x;
groupedData[item].posY = nextClusterPos.y;
}
nextXPixel = xAxis.toPixels(groupedData[item].posX || 0) -
gridOffset.plotLeft;
nextYPixel = yAxis.toPixels(groupedData[item].posY || 0) -
gridOffset.plotTop;
_a = __read(item.split('-').map(parseFloat), 2), itemY = _a[0], itemX = _a[1];
if (zoneOptions) {
pointsLen = groupedData[item].length;
for (i = 0; i < zoneOptions.length; i++) {
if (pointsLen >= zoneOptions[i].from &&
pointsLen <= zoneOptions[i].to) {
if (defined((zoneOptions[i].marker || {}).radius)) {
radius = zoneOptions[i].marker.radius || 0;
}
else if (clusterMarkerOptions &&
clusterMarkerOptions.radius) {
radius = clusterMarkerOptions.radius;
}
else {
radius = clusterDefaultOptions.marker.radius;
}
}
}
}
if (groupedData[item].length > 1 &&
radius === 0 &&
clusterMarkerOptions &&
clusterMarkerOptions.radius) {
radius = clusterMarkerOptions.radius;
}
else if (groupedData[item].length === 1) {
radius = defaultRadius;
}
maxDist = clusterRadius + radius;
radius = 0;
if (itemX !== gridX &&
Math.abs(xPixel - nextXPixel) < maxDist) {
xPixel = itemX - gridX < 0 ? gridXPx + clusterRadius :
gridXPx + gridSize - clusterRadius;
}
if (itemY !== gridY &&
Math.abs(yPixel - nextYPixel) < maxDist) {
yPixel = itemY - gridY < 0 ? gridYPx + clusterRadius :
gridYPx + gridSize - clusterRadius;
}
}
});
x = xAxis.toValue(xPixel + gridOffset.plotLeft);
y = yAxis.toValue(yPixel + gridOffset.plotTop);
groupedData[props.key].posX = x;
groupedData[props.key].posY = y;
return { x: x, y: y };
};
// Check if user algorithm result is valid groupedDataObject.
Scatter.prototype.isValidGroupedDataObject = function (groupedData) {
var result = false, i;
if (!isObject(groupedData)) {
return false;
}
objectEach(groupedData, function (elem) {
result = true;
if (!isArray(elem) || !elem.length) {
result = false;
return;
}
for (i = 0; i < elem.length; i++) {
if (!isObject(elem[i]) || (!elem[i].x || !elem[i].y)) {
result = false;
return;
}
}
});
return result;
};
Scatter.prototype.getClusteredData = function (groupedData, options) {
var series = this, groupedXData = [], groupedYData = [], clusters = [], // Container for clusters.
noise = [], // Container for points not belonging to any cluster.
groupMap = [], index = 0,
// Prevent minimumClusterSize lower than 2.
minimumClusterSize = Math.max(2, options.minimumClusterSize || 2), stateId, point, points, pointUserOptions, pointsLen, marker, clusterPos, pointOptions, clusterTempPos, zoneOptions, clusterZone, clusterZoneClassName, i, k;
// Check if groupedData is valid when user uses a custom algorithm.
if (isFunction(options.layoutAlgorithm.type) &&
!series.isValidGroupedDataObject(groupedData)) {
error('Highcharts marker-clusters module: ' +
'The custom algorithm result is not valid!', false, series.chart);
return false;
}
for (k in groupedData) {
if (groupedData[k].length >= minimumClusterSize) {
points = groupedData[k];
stateId = getStateId();
pointsLen = points.length;
// Get zone options for cluster.
if (options.zones) {
for (i = 0; i < options.zones.length; i++) {
if (pointsLen >= options.zones[i].from &&
pointsLen <= options.zones[i].to) {
clusterZone = options.zones[i];
clusterZone.zoneIndex = i;
zoneOptions = options.zones[i].marker;
clusterZoneClassName = options.zones[i].className;
}
}
}
clusterTempPos = getClusterPosition(points);
if (options.layoutAlgorithm.type === 'grid' &&
!options.allowOverlap) {
marker = series.options.marker || {};
clusterPos = series.preventClusterCollisions({
x: clusterTempPos.x,
y: clusterTempPos.y,
key: k,
groupedData: groupedData,
gridSize: series.getScaledGridSize(options.layoutAlgorithm),
defaultRadius: marker.radius || 3 + (marker.lineWidth || 0),
clusterRadius: (zoneOptions && zoneOptions.radius) ?
zoneOptions.radius :
(options.marker || {}).radius ||
clusterDefaultOptions.marker.radius
});
}
else {
clusterPos = {
x: clusterTempPos.x,
y: clusterTempPos.y
};
}
for (i = 0; i < pointsLen; i++) {
points[i].parentStateId = stateId;
}
clusters.push({
x: clusterPos.x,
y: clusterPos.y,
id: k,
stateId: stateId,
index: index,
data: points,
clusterZone: clusterZone,
clusterZoneClassName: clusterZoneClassName
});
groupedXData.push(clusterPos.x);
groupedYData.push(clusterPos.y);
groupMap.push({
options: {
formatPrefix: 'cluster',
dataLabels: options.dataLabels,
marker: merge(options.marker, {
states: options.states
}, zoneOptions || {})
}
});
// Save cluster data points options.
if (series.options.data && series.options.data.length) {
for (i = 0; i < pointsLen; i++) {
if (isObject(series.options.data[points[i].dataIndex])) {
points[i].options =
series.options.data[points[i].dataIndex];
}
}
}
index++;
zoneOptions = null;
}
else {
for (i = 0; i < groupedData[k].length; i++) {
// Points not belonging to any cluster.
point = groupedData[k][i];
stateId = getStateId();
pointOptions = null;
pointUserOptions =
((series.options || {}).data || [])[point.dataIndex];
groupedXData.push(point.x);
groupedYData.push(point.y);
point.parentStateId = stateId;
noise.push({
x: point.x,
y: point.y,
id: k,
stateId: stateId,
index: index,
data: groupedData[k]
});
if (pointUserOptions &&
typeof pointUserOptions === 'object' &&
!isArray(pointUserOptions)) {
pointOptions = merge(pointUserOptions, { x: point.x, y: point.y });
}
else {
pointOptions = {
userOptions: pointUserOptions,
x: point.x,
y: point.y
};
}
groupMap.push({ options: pointOptions });
index++;
}
}
}
return {
clusters: clusters,
noise: noise,
groupedXData: groupedXData,
groupedYData: groupedYData,
groupMap: groupMap
};
};
// Destroy clustered data points.
Scatter.prototype.destroyClusteredData = function () {
var clusteredSeriesData = this.markerClusterSeriesData;
// Clear previous groups.
(clusteredSeriesData || []).forEach(function (point) {
if (point && point.destroy) {
point.destroy();
}
});
this.markerClusterSeriesData = null;
};
// Hide clustered data points.
Scatter.prototype.hideClusteredData = function () {
var series = this, clusteredSeriesData = this.markerClusterSeriesData, oldState = ((series.markerClusterInfo || {}).pointsState || {}).oldState || {}, oldPointsId = oldPointsStateId.map(function (elem) {
return (oldState[elem].point || {}).id || '';
});
(clusteredSeriesData || []).forEach(function (point) {
// If an old point is used in animation hide it, otherwise destroy.
if (point &&
oldPointsId.indexOf(point.id) !== -1) {
if (point.graphic) {
point.graphic.hide();
}
if (point.dataLabel) {
point.dataLabel.hide();
}
}
else {
if (point && point.destroy) {
point.destroy();
}
}
});
};
// Override the generatePoints method by adding a reference to grouped data.
Scatter.prototype.generatePoints = function () {
var series = this, chart = series.chart, xAxis = series.xAxis, yAxis = series.yAxis, clusterOptions = series.options.cluster, realExtremes = series.getRealExtremes(), visibleXData = [], visibleYData = [], visibleDataIndexes = [], oldPointsState, oldDataLen, oldMarkerClusterInfo, kmeansThreshold, cropDataOffsetX, cropDataOffsetY, seriesMinX, seriesMaxX, seriesMinY, seriesMaxY, type, algorithm, clusteredData, groupedData, layoutAlgOptions, point, i;
if (clusterOptions &&
clusterOptions.enabled &&
series.xData &&
series.yData &&
!chart.polar) {
type = clusterOptions.layoutAlgorithm.type;
layoutAlgOptions = clusterOptions.layoutAlgorithm;
// Get processed algorithm properties.
layoutAlgOptions.processedGridSize = relativeLength(layoutAlgOptions.gridSize ||
clusterDefaultOptions.layoutAlgorithm.gridSize, chart.plotWidth);
layoutAlgOptions.processedDistance = relativeLength(layoutAlgOptions.distance ||
clusterDefaultOptions.layoutAlgorithm.distance, chart.plotWidth);
kmeansThreshold = layoutAlgOptions.kmeansThreshold ||
clusterDefaultOptions.layoutAlgorithm.kmeansThreshold;
// Offset to prevent cluster size changes.
cropDataOffsetX = Math.abs(xAxis.toValue(layoutAlgOptions.processedGridSize / 2) -
xAxis.toValue(0));
cropDataOffsetY = Math.abs(yAxis.toValue(layoutAlgOptions.processedGridSize / 2) -
yAxis.toValue(0));
// Get only visible data.
for (i = 0; i < series.xData.length; i++) {
if (!series.dataMaxX) {
if (!defined(seriesMaxX) ||
!defined(seriesMinX) ||
!defined(seriesMaxY) ||
!defined(seriesMinY)) {
seriesMaxX = seriesMinX = series.xData[i];
seriesMaxY = seriesMinY = series.yData[i];
}
else if (isNumber(series.yData[i]) &&
isNumber(seriesMaxY) &&
isNumber(seriesMinY)) {
seriesMaxX = Math.max(series.xData[i], seriesMaxX);
seriesMinX = Math.min(series.xData[i], seriesMinX);
seriesMaxY = Math.max(series.yData[i] || seriesMaxY, seriesMaxY);
seriesMinY = Math.min(series.yData[i] || seriesMinY, seriesMinY);
}
}
// Crop data to visible ones with appropriate offset to prevent
// cluster size changes on the edge of the plot area.
if (series.xData[i] >= (realExtremes.minX - cropDataOffsetX) &&
series.xData[i] <= (realExtremes.maxX + cropDataOffsetX) &&
(series.yData[i] || realExtremes.minY) >=
(realExtremes.minY - cropDataOffsetY) &&
(series.yData[i] || realExtremes.maxY) <=
(realExtremes.maxY + cropDataOffsetY)) {
visibleXData.push(series.xData[i]);
visibleYData.push(series.yData[i]);
visibleDataIndexes.push(i);
}
}
// Save data max values.
if (defined(seriesMaxX) && defined(seriesMinX) &&
isNumber(seriesMaxY) && isNumber(seriesMinY)) {
series.dataMaxX = seriesMaxX;
series.dataMinX = seriesMinX;
series.dataMaxY = seriesMaxY;
series.dataMinY = seriesMinY;
}
if (isFunction(type)) {
algorithm = type;
}
else if (series.markerClusterAlgorithms) {
if (type && series.markerClusterAlgorithms[type]) {
algorithm = series.markerClusterAlgorithms[type];
}
else {
algorithm = visibleXData.length < kmeansThreshold ?
series.markerClusterAlgorithms.kmeans :
series.markerClusterAlgorithms.grid;
}
}
else {
algorithm = function () {
return false;
};
}
groupedData = algorithm.call(this, visibleXData, visibleYData, visibleDataIndexes, layoutAlgOptions);
clusteredData = groupedData ? series.getClusteredData(groupedData, clusterOptions) : groupedData;
// When animation is enabled get old points state.
if (clusterOptions.animation &&
series.markerClusterInfo &&
series.markerClusterInfo.pointsState &&
series.markerClusterInfo.pointsState.oldState) {
// Destroy old points.
destroyOldPoints(series.markerClusterInfo.pointsState.oldState);
oldPointsState = series.markerClusterInfo.pointsState.newState;
}
else {
oldPointsState = {};
}
// Save points old state info.
oldDataLen = series.xData.length;
oldMarkerClusterInfo = series.markerClusterInfo;
if (clusteredData) {
series.processedXData = clusteredData.groupedXData;
series.processedYData = clusteredData.groupedYData;
series.hasGroupedData = true;
series.markerClusterInfo = clusteredData;
series.groupMap = clusteredData.groupMap;
}
baseGeneratePoints.apply(this);
if (clusteredData && series.markerClusterInfo) {
// Mark cluster points. Safe point reference in the cluster object.
(series.markerClusterInfo.clusters || []).forEach(function (cluster) {
point = series.points[cluster.index];
point.isCluster = true;
point.clusteredData = cluster.data;
point.clusterPointsAmount = cluster.data.length;
cluster.point = point;
// Add zoom to cluster range.
addEvent(point, 'click', series.onDrillToCluster);
});
// Safe point reference in the noise object.
(series.markerClusterInfo.noise || []).forEach(function (noise) {
noise.point = series.points[noise.index];
});
// When animation is enabled save points state.
if (clusterOptions.animation &&
series.markerClusterInfo) {
series.markerClusterInfo.pointsState = {
oldState: oldPointsState,
newState: series.getPointsState(clusteredData, oldMarkerClusterInfo, oldDataLen)
};
}
// Record grouped data in order to let it be destroyed the next time
// processData runs.
if (!clusterOptions.animation) {
this.destroyClusteredData();
}
else {
this.hideClusteredData();
}
this.markerClusterSeriesData =
this.hasGroupedData ? this.points : null;
}
}
else {
baseGeneratePoints.apply(this);
}
};
// Handle animation.
addEvent(H.Chart, 'render', function () {
var chart = this;
(chart.series || []).forEach(function (series) {
if (series.markerClusterInfo) {
var options = series.options.cluster, pointsState = (series.markerClusterInfo || {}).pointsState, oldState = (pointsState || {}).oldState;
if ((options || {}).animation &&
series.markerClusterInfo &&
series.chart.pointer.pinchDown.length === 0 &&
(series.xAxis.eventArgs || {}).trigger !== 'pan' &&
oldState &&
Object.keys(oldState).length) {
series.markerClusterInfo.clusters.forEach(function (cluster) {
series.animateClusterPoint(cluster);
});
series.markerClusterInfo.noise.forEach(function (noise) {
series.animateClusterPoint(noise);
});
}
}
});
});
// Override point prototype to throw a warning when trying to update
// clustered point.
addEvent(Point, 'update', function () {
if (this.dataGroup) {
error('Highcharts marker-clusters module: ' +
'Running `Point.update` when point belongs to clustered series' +
' is not supported.', false, this.series.chart);
return false;
}
});
// Destroy grouped data on series destroy.
addEvent(Series, 'destroy', Scatter.prototype.destroyClusteredData);
// Add classes, change mouse cursor.
addEvent(Series, 'afterRender', function () {
var series = this, clusterZoomEnabled = (series.options.cluster || {}).drillToCluster;
if (series.markerClusterInfo && series.markerClusterInfo.clusters) {
series.markerClusterInfo.clusters.forEach(function (cluster) {
if (cluster.point && cluster.point.graphic) {
cluster.point.graphic.addClass('highcharts-cluster-point');
// Change cursor to pointer when drillToCluster is enabled.
if (clusterZoomEnabled && cluster.point) {
cluster.point.graphic.css({
cursor: 'pointer'
});
if (cluster.point.dataLabel) {
cluster.point.dataLabel.css({
cursor: 'pointer'
});
}
}
if (defined(cluster.clusterZone)) {
cluster.point.graphic.addClass(cluster.clusterZoneClassName ||
'highcharts-cluster-zone-' +
cluster.clusterZone.zoneIndex);
}
}
});
}
});
addEvent(Point, 'drillToCluster', function (event) {
var point = event.point || event.target, series = point.series, clusterOptions = series.options.cluster, onDrillToCluster = ((clusterOptions || {}).events || {}).drillToCluster;
if (isFunction(onDrillToCluster)) {
onDrillToCluster.call(this, event);
}
});
// Destroy the old tooltip after zoom.
addEvent(H.Axis, 'setExtremes', function () {
var chart = this.chart, animationDuration = 0, animation;
chart.series.forEach(function (series) {
if (series.markerClusterInfo) {
animation = animObject((series.options.cluster || {}).animation);
animationDuration = animation.duration || 0;
}
});
syncTimeout(function () {
if (chart.tooltip) {
chart.tooltip.destroy();
}
}, animationDuration);
});
});
_registerModule(_modules, 'masters/modules/marker-clusters.src.js', [], function () {
});
}));