forked from qwerty/tupali
2241 lines
98 KiB
JavaScript
2241 lines
98 KiB
JavaScript
/**
|
|
* @license Highcharts JS v8.1.0 (2020-05-05)
|
|
*
|
|
* Data module
|
|
*
|
|
* (c) 2012-2019 Torstein Honsi
|
|
*
|
|
* 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/data', ['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, 'mixins/ajax.js', [_modules['parts/Globals.js'], _modules['parts/Utilities.js']], function (H, U) {
|
|
/* *
|
|
*
|
|
* (c) 2010-2017 Christer Vasseng, Torstein Honsi
|
|
*
|
|
* License: www.highcharts.com/license
|
|
*
|
|
* !!!!!!! SOURCE GETS TRANSPILED BY TYPESCRIPT. EDIT TS FILE ONLY. !!!!!!!
|
|
*
|
|
* */
|
|
var merge = U.merge, objectEach = U.objectEach;
|
|
/**
|
|
* @interface Highcharts.AjaxSettingsObject
|
|
*/ /**
|
|
* The payload to send.
|
|
*
|
|
* @name Highcharts.AjaxSettingsObject#data
|
|
* @type {string|Highcharts.Dictionary<any>}
|
|
*/ /**
|
|
* The data type expected.
|
|
* @name Highcharts.AjaxSettingsObject#dataType
|
|
* @type {"json"|"xml"|"text"|"octet"}
|
|
*/ /**
|
|
* Function to call on error.
|
|
* @name Highcharts.AjaxSettingsObject#error
|
|
* @type {Function}
|
|
*/ /**
|
|
* The headers; keyed on header name.
|
|
* @name Highcharts.AjaxSettingsObject#headers
|
|
* @type {Highcharts.Dictionary<string>}
|
|
*/ /**
|
|
* Function to call on success.
|
|
* @name Highcharts.AjaxSettingsObject#success
|
|
* @type {Function}
|
|
*/ /**
|
|
* The verb to use.
|
|
* @name Highcharts.AjaxSettingsObject#type
|
|
* @type {"GET"|"POST"|"UPDATE"|"DELETE"}
|
|
*/ /**
|
|
* The URL to call.
|
|
* @name Highcharts.AjaxSettingsObject#url
|
|
* @type {string}
|
|
*/
|
|
/**
|
|
* Perform an Ajax call.
|
|
*
|
|
* @function Highcharts.ajax
|
|
*
|
|
* @param {Partial<Highcharts.AjaxSettingsObject>} attr
|
|
* The Ajax settings to use.
|
|
*
|
|
* @return {false|undefined}
|
|
* Returns false, if error occured.
|
|
*/
|
|
H.ajax = function (attr) {
|
|
var options = merge(true, {
|
|
url: false,
|
|
type: 'get',
|
|
dataType: 'json',
|
|
success: false,
|
|
error: false,
|
|
data: false,
|
|
headers: {}
|
|
}, attr), headers = {
|
|
json: 'application/json',
|
|
xml: 'application/xml',
|
|
text: 'text/plain',
|
|
octet: 'application/octet-stream'
|
|
}, r = new XMLHttpRequest();
|
|
/**
|
|
* @private
|
|
* @param {XMLHttpRequest} xhr - Internal request object.
|
|
* @param {string|Error} err - Occured error.
|
|
* @return {void}
|
|
*/
|
|
function handleError(xhr, err) {
|
|
if (options.error) {
|
|
options.error(xhr, err);
|
|
}
|
|
else {
|
|
// @todo Maybe emit a highcharts error event here
|
|
}
|
|
}
|
|
if (!options.url) {
|
|
return false;
|
|
}
|
|
r.open(options.type.toUpperCase(), options.url, true);
|
|
if (!options.headers['Content-Type']) {
|
|
r.setRequestHeader('Content-Type', headers[options.dataType] || headers.text);
|
|
}
|
|
objectEach(options.headers, function (val, key) {
|
|
r.setRequestHeader(key, val);
|
|
});
|
|
// @todo lacking timeout handling
|
|
r.onreadystatechange = function () {
|
|
var res;
|
|
if (r.readyState === 4) {
|
|
if (r.status === 200) {
|
|
res = r.responseText;
|
|
if (options.dataType === 'json') {
|
|
try {
|
|
res = JSON.parse(res);
|
|
}
|
|
catch (e) {
|
|
return handleError(r, e);
|
|
}
|
|
}
|
|
return options.success && options.success(res);
|
|
}
|
|
handleError(r, r.responseText);
|
|
}
|
|
};
|
|
try {
|
|
options.data = JSON.stringify(options.data);
|
|
}
|
|
catch (e) {
|
|
// empty
|
|
}
|
|
r.send(options.data || true);
|
|
};
|
|
/**
|
|
* Get a JSON resource over XHR, also supporting CORS without preflight.
|
|
*
|
|
* @function Highcharts.getJSON
|
|
* @param {string} url
|
|
* The URL to load.
|
|
* @param {Function} success
|
|
* The success callback. For error handling, use the `Highcharts.ajax`
|
|
* function instead.
|
|
* @return {void}
|
|
*/
|
|
H.getJSON = function (url, success) {
|
|
H.ajax({
|
|
url: url,
|
|
success: success,
|
|
dataType: 'json',
|
|
headers: {
|
|
// Override the Content-Type to avoid preflight problems with CORS
|
|
// in the Highcharts demos
|
|
'Content-Type': 'text/plain'
|
|
}
|
|
});
|
|
};
|
|
|
|
});
|
|
_registerModule(_modules, 'modules/data.src.js', [_modules['parts/Globals.js'], _modules['parts/Utilities.js'], _modules['parts/Globals.js'], _modules['parts/Point.js']], function (Highcharts, U, H, Point) {
|
|
/* *
|
|
*
|
|
* Data module
|
|
*
|
|
* (c) 2012-2020 Torstein Honsi
|
|
*
|
|
* License: www.highcharts.com/license
|
|
*
|
|
* !!!!!!! SOURCE GETS TRANSPILED BY TYPESCRIPT. EDIT TS FILE ONLY. !!!!!!!
|
|
*
|
|
* */
|
|
/**
|
|
* Callback function to modify the CSV before parsing it by the data module.
|
|
*
|
|
* @callback Highcharts.DataBeforeParseCallbackFunction
|
|
*
|
|
* @param {string} csv
|
|
* The CSV to modify.
|
|
*
|
|
* @return {string}
|
|
* The CSV to parse.
|
|
*/
|
|
/**
|
|
* Callback function that gets called after parsing data.
|
|
*
|
|
* @callback Highcharts.DataCompleteCallbackFunction
|
|
*
|
|
* @param {Highcharts.Options} chartOptions
|
|
* The chart options that were used.
|
|
*/
|
|
/**
|
|
* Callback function that returns the correspondig Date object to a match.
|
|
*
|
|
* @callback Highcharts.DataDateFormatCallbackFunction
|
|
*
|
|
* @param {Array<number>} match
|
|
*
|
|
* @return {number}
|
|
*/
|
|
/**
|
|
* Structure for alternative date formats to parse.
|
|
*
|
|
* @interface Highcharts.DataDateFormatObject
|
|
*/ /**
|
|
* @name Highcharts.DataDateFormatObject#alternative
|
|
* @type {string|undefined}
|
|
*/ /**
|
|
* @name Highcharts.DataDateFormatObject#parser
|
|
* @type {Highcharts.DataDateFormatCallbackFunction}
|
|
*/ /**
|
|
* @name Highcharts.DataDateFormatObject#regex
|
|
* @type {global.RegExp}
|
|
*/
|
|
/**
|
|
* Possible types for a data item in a column or row.
|
|
*
|
|
* @typedef {number|string|null} Highcharts.DataValueType
|
|
*/
|
|
/**
|
|
* Callback function to parse string representations of dates into
|
|
* JavaScript timestamps (milliseconds since 1.1.1970).
|
|
*
|
|
* @callback Highcharts.DataParseDateCallbackFunction
|
|
*
|
|
* @param {string} dateValue
|
|
*
|
|
* @return {number}
|
|
* Timestamp (milliseconds since 1.1.1970) as integer for Date class.
|
|
*/
|
|
/**
|
|
* Callback function to access the parsed columns, the two-dimentional
|
|
* input data array directly, before they are interpreted into series
|
|
* data and categories.
|
|
*
|
|
* @callback Highcharts.DataParsedCallbackFunction
|
|
*
|
|
* @param {Array<Array<*>>} columns
|
|
* The parsed columns by the data module.
|
|
*
|
|
* @return {boolean|undefined}
|
|
* Return `false` to stop completion, or call `this.complete()` to
|
|
* continue async.
|
|
*/
|
|
var addEvent = U.addEvent, defined = U.defined, extend = U.extend, fireEvent = U.fireEvent, isNumber = U.isNumber, merge = U.merge, objectEach = U.objectEach, pick = U.pick, splat = U.splat;
|
|
// Utilities
|
|
var Chart = H.Chart, win = H.win, doc = win.document;
|
|
/**
|
|
* The Data module provides a simplified interface for adding data to
|
|
* a chart from sources like CVS, HTML tables or grid views. See also
|
|
* the [tutorial article on the Data module](
|
|
* https://www.highcharts.com/docs/working-with-data/data-module).
|
|
*
|
|
* It requires the `modules/data.js` file to be loaded.
|
|
*
|
|
* Please note that the default way of adding data in Highcharts, without
|
|
* the need of a module, is through the [series._type_.data](#series.line.data)
|
|
* option.
|
|
*
|
|
* @sample {highcharts} highcharts/demo/column-parsed/
|
|
* HTML table
|
|
* @sample {highcharts} highcharts/data/csv/
|
|
* CSV
|
|
*
|
|
* @since 4.0
|
|
* @requires modules/data
|
|
* @apioption data
|
|
*/
|
|
/**
|
|
* A callback function to modify the CSV before parsing it. Return the modified
|
|
* string.
|
|
*
|
|
* @sample {highcharts} highcharts/demo/line-ajax/
|
|
* Modify CSV before parse
|
|
*
|
|
* @type {Highcharts.DataBeforeParseCallbackFunction}
|
|
* @since 6.1
|
|
* @apioption data.beforeParse
|
|
*/
|
|
/**
|
|
* A two-dimensional array representing the input data on tabular form.
|
|
* This input can be used when the data is already parsed, for example
|
|
* from a grid view component. Each cell can be a string or number.
|
|
* If not switchRowsAndColumns is set, the columns are interpreted as
|
|
* series.
|
|
*
|
|
* @see [data.rows](#data.rows)
|
|
*
|
|
* @sample {highcharts} highcharts/data/columns/
|
|
* Columns
|
|
*
|
|
* @type {Array<Array<Highcharts.DataValueType>>}
|
|
* @since 4.0
|
|
* @apioption data.columns
|
|
*/
|
|
/**
|
|
* The callback that is evaluated when the data is finished loading,
|
|
* optionally from an external source, and parsed. The first argument
|
|
* passed is a finished chart options object, containing the series.
|
|
* These options can be extended with additional options and passed
|
|
* directly to the chart constructor.
|
|
*
|
|
* @see [data.parsed](#data.parsed)
|
|
*
|
|
* @sample {highcharts} highcharts/data/complete/
|
|
* Modify data on complete
|
|
*
|
|
* @type {Highcharts.DataCompleteCallbackFunction}
|
|
* @since 4.0
|
|
* @apioption data.complete
|
|
*/
|
|
/**
|
|
* A comma delimited string to be parsed. Related options are [startRow](
|
|
* #data.startRow), [endRow](#data.endRow), [startColumn](#data.startColumn)
|
|
* and [endColumn](#data.endColumn) to delimit what part of the table
|
|
* is used. The [lineDelimiter](#data.lineDelimiter) and [itemDelimiter](
|
|
* #data.itemDelimiter) options define the CSV delimiter formats.
|
|
*
|
|
* The built-in CSV parser doesn't support all flavours of CSV, so in
|
|
* some cases it may be necessary to use an external CSV parser. See
|
|
* [this example](https://jsfiddle.net/highcharts/u59176h4/) of parsing
|
|
* CSV through the MIT licensed [Papa Parse](http://papaparse.com/)
|
|
* library.
|
|
*
|
|
* @sample {highcharts} highcharts/data/csv/
|
|
* Data from CSV
|
|
*
|
|
* @type {string}
|
|
* @since 4.0
|
|
* @apioption data.csv
|
|
*/
|
|
/**
|
|
* Which of the predefined date formats in Date.prototype.dateFormats
|
|
* to use to parse date values. Defaults to a best guess based on what
|
|
* format gives valid and ordered dates. Valid options include: `YYYY/mm/dd`,
|
|
* `dd/mm/YYYY`, `mm/dd/YYYY`, `dd/mm/YY`, `mm/dd/YY`.
|
|
*
|
|
* @see [data.parseDate](#data.parseDate)
|
|
*
|
|
* @sample {highcharts} highcharts/data/dateformat-auto/
|
|
* Best guess date format
|
|
*
|
|
* @type {string}
|
|
* @since 4.0
|
|
* @validvalue ["YYYY/mm/dd", "dd/mm/YYYY", "mm/dd/YYYY", "dd/mm/YYYY",
|
|
* "dd/mm/YY", "mm/dd/YY"]
|
|
* @apioption data.dateFormat
|
|
*/
|
|
/**
|
|
* The decimal point used for parsing numbers in the CSV.
|
|
*
|
|
* If both this and data.delimiter is set to `undefined`, the parser will
|
|
* attempt to deduce the decimal point automatically.
|
|
*
|
|
* @sample {highcharts} highcharts/data/delimiters/
|
|
* Comma as decimal point
|
|
*
|
|
* @type {string}
|
|
* @default .
|
|
* @since 4.1.0
|
|
* @apioption data.decimalPoint
|
|
*/
|
|
/**
|
|
* In tabular input data, the last column (indexed by 0) to use. Defaults
|
|
* to the last column containing data.
|
|
*
|
|
* @sample {highcharts} highcharts/data/start-end/
|
|
* Limited data
|
|
*
|
|
* @type {number}
|
|
* @since 4.0
|
|
* @apioption data.endColumn
|
|
*/
|
|
/**
|
|
* In tabular input data, the last row (indexed by 0) to use. Defaults
|
|
* to the last row containing data.
|
|
*
|
|
* @sample {highcharts} highcharts/data/start-end/
|
|
* Limited data
|
|
*
|
|
* @type {number}
|
|
* @since 4.0.4
|
|
* @apioption data.endRow
|
|
*/
|
|
/**
|
|
* Whether to use the first row in the data set as series names.
|
|
*
|
|
* @sample {highcharts} highcharts/data/start-end/
|
|
* Don't get series names from the CSV
|
|
* @sample {highstock} highcharts/data/start-end/
|
|
* Don't get series names from the CSV
|
|
*
|
|
* @type {boolean}
|
|
* @default true
|
|
* @since 4.1.0
|
|
* @product highcharts highstock gantt
|
|
* @apioption data.firstRowAsNames
|
|
*/
|
|
/**
|
|
* The key for a Google Spreadsheet to load. See [general information
|
|
* on GS](https://developers.google.com/gdata/samples/spreadsheet_sample).
|
|
*
|
|
* @sample {highcharts} highcharts/data/google-spreadsheet/
|
|
* Load a Google Spreadsheet
|
|
*
|
|
* @type {string}
|
|
* @since 4.0
|
|
* @apioption data.googleSpreadsheetKey
|
|
*/
|
|
/**
|
|
* The Google Spreadsheet worksheet to use in combination with
|
|
* [googleSpreadsheetKey](#data.googleSpreadsheetKey). The available id's from
|
|
* your sheet can be read from `https://spreadsheets.google.com/feeds/worksheets/{key}/public/basic`.
|
|
*
|
|
* @sample {highcharts} highcharts/data/google-spreadsheet/
|
|
* Load a Google Spreadsheet
|
|
*
|
|
* @type {string}
|
|
* @since 4.0
|
|
* @apioption data.googleSpreadsheetWorksheet
|
|
*/
|
|
/**
|
|
* Item or cell delimiter for parsing CSV. Defaults to the tab character
|
|
* `\t` if a tab character is found in the CSV string, if not it defaults
|
|
* to `,`.
|
|
*
|
|
* If this is set to false or undefined, the parser will attempt to deduce
|
|
* the delimiter automatically.
|
|
*
|
|
* @sample {highcharts} highcharts/data/delimiters/
|
|
* Delimiters
|
|
*
|
|
* @type {string}
|
|
* @since 4.0
|
|
* @apioption data.itemDelimiter
|
|
*/
|
|
/**
|
|
* Line delimiter for parsing CSV.
|
|
*
|
|
* @sample {highcharts} highcharts/data/delimiters/
|
|
* Delimiters
|
|
*
|
|
* @type {string}
|
|
* @default \n
|
|
* @since 4.0
|
|
* @apioption data.lineDelimiter
|
|
*/
|
|
/**
|
|
* A callback function to access the parsed columns, the two-dimentional
|
|
* input data array directly, before they are interpreted into series
|
|
* data and categories. Return `false` to stop completion, or call
|
|
* `this.complete()` to continue async.
|
|
*
|
|
* @see [data.complete](#data.complete)
|
|
*
|
|
* @sample {highcharts} highcharts/data/parsed/
|
|
* Modify data after parse
|
|
*
|
|
* @type {Highcharts.DataParsedCallbackFunction}
|
|
* @since 4.0
|
|
* @apioption data.parsed
|
|
*/
|
|
/**
|
|
* A callback function to parse string representations of dates into
|
|
* JavaScript timestamps. Should return an integer timestamp on success.
|
|
*
|
|
* @see [dateFormat](#data.dateFormat)
|
|
*
|
|
* @type {Highcharts.DataParseDateCallbackFunction}
|
|
* @since 4.0
|
|
* @apioption data.parseDate
|
|
*/
|
|
/**
|
|
* The same as the columns input option, but defining rows intead of
|
|
* columns.
|
|
*
|
|
* @see [data.columns](#data.columns)
|
|
*
|
|
* @sample {highcharts} highcharts/data/rows/
|
|
* Data in rows
|
|
*
|
|
* @type {Array<Array<Highcharts.DataValueType>>}
|
|
* @since 4.0
|
|
* @apioption data.rows
|
|
*/
|
|
/**
|
|
* An array containing dictionaries for each series. A dictionary exists of
|
|
* Point property names as the key and the CSV column index as the value.
|
|
*
|
|
* @sample {highcharts} highcharts/data/seriesmapping-label/
|
|
* Label from data set
|
|
*
|
|
* @type {Array<Highcharts.Dictionary<number>>}
|
|
* @since 4.0.4
|
|
* @apioption data.seriesMapping
|
|
*/
|
|
/**
|
|
* In tabular input data, the first column (indexed by 0) to use.
|
|
*
|
|
* @sample {highcharts} highcharts/data/start-end/
|
|
* Limited data
|
|
*
|
|
* @type {number}
|
|
* @default 0
|
|
* @since 4.0
|
|
* @apioption data.startColumn
|
|
*/
|
|
/**
|
|
* In tabular input data, the first row (indexed by 0) to use.
|
|
*
|
|
* @sample {highcharts} highcharts/data/start-end/
|
|
* Limited data
|
|
*
|
|
* @type {number}
|
|
* @default 0
|
|
* @since 4.0
|
|
* @apioption data.startRow
|
|
*/
|
|
/**
|
|
* Switch rows and columns of the input data, so that `this.columns`
|
|
* effectively becomes the rows of the data set, and the rows are interpreted
|
|
* as series.
|
|
*
|
|
* @sample {highcharts} highcharts/data/switchrowsandcolumns/
|
|
* Switch rows and columns
|
|
*
|
|
* @type {boolean}
|
|
* @default false
|
|
* @since 4.0
|
|
* @apioption data.switchRowsAndColumns
|
|
*/
|
|
/**
|
|
* An HTML table or the id of such to be parsed as input data. Related
|
|
* options are `startRow`, `endRow`, `startColumn` and `endColumn` to
|
|
* delimit what part of the table is used.
|
|
*
|
|
* @sample {highcharts} highcharts/demo/column-parsed/
|
|
* Parsed table
|
|
*
|
|
* @type {string|global.HTMLElement}
|
|
* @since 4.0
|
|
* @apioption data.table
|
|
*/
|
|
/**
|
|
* An URL to a remote CSV dataset. Will be fetched when the chart is created
|
|
* using Ajax.
|
|
*
|
|
* @sample highcharts/data/livedata-columns
|
|
* Categorized bar chart with CSV and live polling
|
|
* @sample highcharts/data/livedata-csv
|
|
* Time based line chart with CSV and live polling
|
|
*
|
|
* @type {string}
|
|
* @apioption data.csvURL
|
|
*/
|
|
/**
|
|
* A URL to a remote JSON dataset, structured as a row array.
|
|
* Will be fetched when the chart is created using Ajax.
|
|
*
|
|
* @sample highcharts/data/livedata-rows
|
|
* Rows with live polling
|
|
*
|
|
* @type {string}
|
|
* @apioption data.rowsURL
|
|
*/
|
|
/**
|
|
* A URL to a remote JSON dataset, structured as a column array.
|
|
* Will be fetched when the chart is created using Ajax.
|
|
*
|
|
* @sample highcharts/data/livedata-columns
|
|
* Columns with live polling
|
|
*
|
|
* @type {string}
|
|
* @apioption data.columnsURL
|
|
*/
|
|
/**
|
|
* Sets the refresh rate for data polling when importing remote dataset by
|
|
* setting [data.csvURL](data.csvURL), [data.rowsURL](data.rowsURL),
|
|
* [data.columnsURL](data.columnsURL), or
|
|
* [data.googleSpreadsheetKey](data.googleSpreadsheetKey).
|
|
*
|
|
* Note that polling must be enabled by setting
|
|
* [data.enablePolling](data.enablePolling) to true.
|
|
*
|
|
* The value is the number of seconds between pollings.
|
|
* It cannot be set to less than 1 second.
|
|
*
|
|
* @sample highcharts/demo/live-data
|
|
* Live data with user set refresh rate
|
|
*
|
|
* @default 1
|
|
* @type {number}
|
|
* @apioption data.dataRefreshRate
|
|
*/
|
|
/**
|
|
* Enables automatic refetching of remote datasets every _n_ seconds (defined by
|
|
* setting [data.dataRefreshRate](data.dataRefreshRate)).
|
|
*
|
|
* Only works when either [data.csvURL](data.csvURL),
|
|
* [data.rowsURL](data.rowsURL), [data.columnsURL](data.columnsURL), or
|
|
* [data.googleSpreadsheetKey](data.googleSpreadsheetKey).
|
|
*
|
|
* @sample highcharts/demo/live-data
|
|
* Live data
|
|
* @sample highcharts/data/livedata-columns
|
|
* Categorized bar chart with CSV and live polling
|
|
*
|
|
* @type {boolean}
|
|
* @default false
|
|
* @apioption data.enablePolling
|
|
*/
|
|
/* eslint-disable valid-jsdoc */
|
|
/**
|
|
* The Data class
|
|
*
|
|
* @requires module:modules/data
|
|
*
|
|
* @class
|
|
* @name Highcharts.Data
|
|
*
|
|
* @param {Highcharts.DataOptions} dataOptions
|
|
*
|
|
* @param {Highcharts.Options} [chartOptions]
|
|
*
|
|
* @param {Highcharts.Chart} [chart]
|
|
*/
|
|
var Data = /** @class */ (function () {
|
|
function Data(dataOptions, chartOptions, chart) {
|
|
this.chart = void 0;
|
|
this.chartOptions = void 0;
|
|
this.firstRowAsNames = void 0;
|
|
this.rawColumns = void 0;
|
|
this.options = void 0;
|
|
/**
|
|
* A collection of available date formats, extendable from the outside to
|
|
* support custom date formats.
|
|
*
|
|
* @name Highcharts.Data#dateFormats
|
|
* @type {Highcharts.Dictionary<Highcharts.DataDateFormatObject>}
|
|
*/
|
|
this.dateFormats = {
|
|
'YYYY/mm/dd': {
|
|
regex: /^([0-9]{4})[\-\/\.]([0-9]{1,2})[\-\/\.]([0-9]{1,2})$/,
|
|
parser: function (match) {
|
|
return (match ?
|
|
Date.UTC(+match[1], match[2] - 1, +match[3]) :
|
|
NaN);
|
|
}
|
|
},
|
|
'dd/mm/YYYY': {
|
|
regex: /^([0-9]{1,2})[\-\/\.]([0-9]{1,2})[\-\/\.]([0-9]{4})$/,
|
|
parser: function (match) {
|
|
return (match ?
|
|
Date.UTC(+match[3], match[2] - 1, +match[1]) :
|
|
NaN);
|
|
},
|
|
alternative: 'mm/dd/YYYY' // different format with the same regex
|
|
},
|
|
'mm/dd/YYYY': {
|
|
regex: /^([0-9]{1,2})[\-\/\.]([0-9]{1,2})[\-\/\.]([0-9]{4})$/,
|
|
parser: function (match) {
|
|
return (match ?
|
|
Date.UTC(+match[3], match[1] - 1, +match[2]) :
|
|
NaN);
|
|
}
|
|
},
|
|
'dd/mm/YY': {
|
|
regex: /^([0-9]{1,2})[\-\/\.]([0-9]{1,2})[\-\/\.]([0-9]{2})$/,
|
|
parser: function (match) {
|
|
if (!match) {
|
|
return NaN;
|
|
}
|
|
var year = +match[3], d = new Date();
|
|
if (year > (d.getFullYear() - 2000)) {
|
|
year += 1900;
|
|
}
|
|
else {
|
|
year += 2000;
|
|
}
|
|
return Date.UTC(year, match[2] - 1, +match[1]);
|
|
},
|
|
alternative: 'mm/dd/YY' // different format with the same regex
|
|
},
|
|
'mm/dd/YY': {
|
|
regex: /^([0-9]{1,2})[\-\/\.]([0-9]{1,2})[\-\/\.]([0-9]{2})$/,
|
|
parser: function (match) {
|
|
return (match ?
|
|
Date.UTC(+match[3] + 2000, match[1] - 1, +match[2]) :
|
|
NaN);
|
|
}
|
|
}
|
|
};
|
|
this.init(dataOptions, chartOptions, chart);
|
|
}
|
|
/**
|
|
* Initialize the Data object with the given options
|
|
*
|
|
* @private
|
|
* @function Highcharts.Data#init
|
|
* @param {Highcharts.DataOptions} options
|
|
* @param {Highcharts.Options} [chartOptions]
|
|
* @param {Highcharts.Chart} [chart]
|
|
*/
|
|
Data.prototype.init = function (options, chartOptions, chart) {
|
|
var decimalPoint = options.decimalPoint, hasData;
|
|
if (chartOptions) {
|
|
this.chartOptions = chartOptions;
|
|
}
|
|
if (chart) {
|
|
this.chart = chart;
|
|
}
|
|
if (decimalPoint !== '.' && decimalPoint !== ',') {
|
|
decimalPoint = void 0;
|
|
}
|
|
this.options = options;
|
|
this.columns = (options.columns ||
|
|
this.rowsToColumns(options.rows) ||
|
|
[]);
|
|
this.firstRowAsNames = pick(options.firstRowAsNames, this.firstRowAsNames, true);
|
|
this.decimalRegex = (decimalPoint &&
|
|
new RegExp('^(-?[0-9]+)' + decimalPoint + '([0-9]+)$'));
|
|
// This is a two-dimensional array holding the raw, trimmed string
|
|
// values with the same organisation as the columns array. It makes it
|
|
// possible for example to revert from interpreted timestamps to
|
|
// string-based categories.
|
|
this.rawColumns = [];
|
|
// No need to parse or interpret anything
|
|
if (this.columns.length) {
|
|
this.dataFound();
|
|
hasData = true;
|
|
}
|
|
if (this.hasURLOption(options)) {
|
|
clearTimeout(this.liveDataTimeout);
|
|
hasData = false;
|
|
}
|
|
if (!hasData) {
|
|
// Fetch live data
|
|
hasData = this.fetchLiveData();
|
|
}
|
|
if (!hasData) {
|
|
// Parse a CSV string if options.csv is given. The parseCSV function
|
|
// returns a columns array, if it has no length, we have no data
|
|
hasData = Boolean(this.parseCSV().length);
|
|
}
|
|
if (!hasData) {
|
|
// Parse a HTML table if options.table is given
|
|
hasData = Boolean(this.parseTable().length);
|
|
}
|
|
if (!hasData) {
|
|
// Parse a Google Spreadsheet
|
|
hasData = this.parseGoogleSpreadsheet();
|
|
}
|
|
if (!hasData && options.afterComplete) {
|
|
options.afterComplete();
|
|
}
|
|
};
|
|
Data.prototype.hasURLOption = function (options) {
|
|
return Boolean(options &&
|
|
(options.rowsURL || options.csvURL || options.columnsURL));
|
|
};
|
|
/**
|
|
* Get the column distribution. For example, a line series takes a single
|
|
* column for Y values. A range series takes two columns for low and high
|
|
* values respectively, and an OHLC series takes four columns.
|
|
*
|
|
* @function Highcharts.Data#getColumnDistribution
|
|
*/
|
|
Data.prototype.getColumnDistribution = function () {
|
|
var chartOptions = this.chartOptions, options = this.options, xColumns = [], getValueCount = function (type) {
|
|
return (H.seriesTypes[type || 'line'].prototype
|
|
.pointArrayMap ||
|
|
[0]).length;
|
|
}, getPointArrayMap = function (type) {
|
|
return H.seriesTypes[type || 'line']
|
|
.prototype.pointArrayMap;
|
|
}, globalType = (chartOptions &&
|
|
chartOptions.chart &&
|
|
chartOptions.chart.type), individualCounts = [], seriesBuilders = [], seriesIndex = 0,
|
|
// If no series mapping is defined, check if the series array is
|
|
// defined with types.
|
|
seriesMapping = ((options && options.seriesMapping) ||
|
|
(chartOptions &&
|
|
chartOptions.series &&
|
|
chartOptions.series.map(function () {
|
|
return { x: 0 };
|
|
})) ||
|
|
[]), i;
|
|
((chartOptions && chartOptions.series) || []).forEach(function (series) {
|
|
individualCounts.push(getValueCount(series.type || globalType));
|
|
});
|
|
// Collect the x-column indexes from seriesMapping
|
|
seriesMapping.forEach(function (mapping) {
|
|
xColumns.push(mapping.x || 0);
|
|
});
|
|
// If there are no defined series with x-columns, use the first column
|
|
// as x column
|
|
if (xColumns.length === 0) {
|
|
xColumns.push(0);
|
|
}
|
|
// Loop all seriesMappings and constructs SeriesBuilders from
|
|
// the mapping options.
|
|
seriesMapping.forEach(function (mapping) {
|
|
var builder = new SeriesBuilder(), numberOfValueColumnsNeeded = individualCounts[seriesIndex] ||
|
|
getValueCount(globalType), seriesArr = (chartOptions && chartOptions.series) || [], series = seriesArr[seriesIndex] || {}, defaultPointArrayMap = getPointArrayMap(series.type || globalType), pointArrayMap = defaultPointArrayMap || ['y'];
|
|
if (
|
|
// User-defined x.mapping
|
|
defined(mapping.x) ||
|
|
// All non cartesian don't need 'x'
|
|
series.isCartesian ||
|
|
// Except pie series:
|
|
!defaultPointArrayMap) {
|
|
// Add an x reader from the x property or from an undefined
|
|
// column if the property is not set. It will then be auto
|
|
// populated later.
|
|
builder.addColumnReader(mapping.x, 'x');
|
|
}
|
|
// Add all column mappings
|
|
objectEach(mapping, function (val, name) {
|
|
if (name !== 'x') {
|
|
builder.addColumnReader(val, name);
|
|
}
|
|
});
|
|
// Add missing columns
|
|
for (i = 0; i < numberOfValueColumnsNeeded; i++) {
|
|
if (!builder.hasReader(pointArrayMap[i])) {
|
|
// Create and add a column reader for the next free column
|
|
// index
|
|
builder.addColumnReader(void 0, pointArrayMap[i]);
|
|
}
|
|
}
|
|
seriesBuilders.push(builder);
|
|
seriesIndex++;
|
|
});
|
|
var globalPointArrayMap = getPointArrayMap(globalType);
|
|
if (typeof globalPointArrayMap === 'undefined') {
|
|
globalPointArrayMap = ['y'];
|
|
}
|
|
this.valueCount = {
|
|
global: getValueCount(globalType),
|
|
xColumns: xColumns,
|
|
individual: individualCounts,
|
|
seriesBuilders: seriesBuilders,
|
|
globalPointArrayMap: globalPointArrayMap
|
|
};
|
|
};
|
|
/**
|
|
* When the data is parsed into columns, either by CSV, table, GS or direct
|
|
* input, continue with other operations.
|
|
*
|
|
* @private
|
|
* @function Highcharts.Data#dataFound
|
|
*/
|
|
Data.prototype.dataFound = function () {
|
|
if (this.options.switchRowsAndColumns) {
|
|
this.columns = this.rowsToColumns(this.columns);
|
|
}
|
|
// Interpret the info about series and columns
|
|
this.getColumnDistribution();
|
|
// Interpret the values into right types
|
|
this.parseTypes();
|
|
// Handle columns if a handleColumns callback is given
|
|
if (this.parsed() !== false) {
|
|
// Complete if a complete callback is given
|
|
this.complete();
|
|
}
|
|
};
|
|
/**
|
|
* Parse a CSV input string
|
|
*
|
|
* @function Highcharts.Data#parseCSV
|
|
*
|
|
* @param {Highcharts.DataOptions} [inOptions]
|
|
*
|
|
* @return {Array<Array<Highcharts.DataValueType>>}
|
|
*/
|
|
Data.prototype.parseCSV = function (inOptions) {
|
|
var self = this, options = inOptions || this.options, csv = options.csv, columns, startRow = (typeof options.startRow !== 'undefined' && options.startRow ?
|
|
options.startRow :
|
|
0), endRow = options.endRow || Number.MAX_VALUE, startColumn = (typeof options.startColumn !== 'undefined' &&
|
|
options.startColumn) ? options.startColumn : 0, endColumn = options.endColumn || Number.MAX_VALUE, itemDelimiter, lines, rowIt = 0,
|
|
// activeRowNo = 0,
|
|
dataTypes = [],
|
|
// We count potential delimiters in the prepass, and use the
|
|
// result as the basis of half-intelligent guesses.
|
|
potDelimiters = {
|
|
',': 0,
|
|
';': 0,
|
|
'\t': 0
|
|
};
|
|
columns = this.columns = [];
|
|
/*
|
|
This implementation is quite verbose. It will be shortened once
|
|
it's stable and passes all the test.
|
|
|
|
It's also not written with speed in mind, instead everything is
|
|
very seggregated, and there a several redundant loops.
|
|
This is to make it easier to stabilize the code initially.
|
|
|
|
We do a pre-pass on the first 4 rows to make some intelligent
|
|
guesses on the set. Guessed delimiters are in this pass counted.
|
|
|
|
Auto detecting delimiters
|
|
- If we meet a quoted string, the next symbol afterwards
|
|
(that's not \s, \t) is the delimiter
|
|
- If we meet a date, the next symbol afterwards is the delimiter
|
|
|
|
Date formats
|
|
- If we meet a column with date formats, check all of them to
|
|
see if one of the potential months crossing 12. If it does,
|
|
we now know the format
|
|
|
|
It would make things easier to guess the delimiter before
|
|
doing the actual parsing.
|
|
|
|
General rules:
|
|
- Quoting is allowed, e.g: "Col 1",123,321
|
|
- Quoting is optional, e.g.: Col1,123,321
|
|
- Doubble quoting is escaping, e.g. "Col ""Hello world""",123
|
|
- Spaces are considered part of the data: Col1 ,123
|
|
- New line is always the row delimiter
|
|
- Potential column delimiters are , ; \t
|
|
- First row may optionally contain headers
|
|
- The last row may or may not have a row delimiter
|
|
- Comments are optionally supported, in which case the comment
|
|
must start at the first column, and the rest of the line will
|
|
be ignored
|
|
*/
|
|
/**
|
|
* Parse a single row.
|
|
* @private
|
|
*/
|
|
function parseRow(columnStr, rowNumber, noAdd, callbacks) {
|
|
var i = 0, c = '', cl = '', cn = '', token = '', actualColumn = 0, column = 0;
|
|
/**
|
|
* @private
|
|
*/
|
|
function read(j) {
|
|
c = columnStr[j];
|
|
cl = columnStr[j - 1];
|
|
cn = columnStr[j + 1];
|
|
}
|
|
/**
|
|
* @private
|
|
*/
|
|
function pushType(type) {
|
|
if (dataTypes.length < column + 1) {
|
|
dataTypes.push([type]);
|
|
}
|
|
if (dataTypes[column][dataTypes[column].length - 1] !== type) {
|
|
dataTypes[column].push(type);
|
|
}
|
|
}
|
|
/**
|
|
* @private
|
|
*/
|
|
function push() {
|
|
if (startColumn > actualColumn || actualColumn > endColumn) {
|
|
// Skip this column, but increment the column count (#7272)
|
|
++actualColumn;
|
|
token = '';
|
|
return;
|
|
}
|
|
if (!isNaN(parseFloat(token)) && isFinite(token)) {
|
|
token = parseFloat(token);
|
|
pushType('number');
|
|
}
|
|
else if (!isNaN(Date.parse(token))) {
|
|
token = token.replace(/\//g, '-');
|
|
pushType('date');
|
|
}
|
|
else {
|
|
pushType('string');
|
|
}
|
|
if (columns.length < column + 1) {
|
|
columns.push([]);
|
|
}
|
|
if (!noAdd) {
|
|
// Don't push - if there's a varrying amount of columns
|
|
// for each row, pushing will skew everything down n slots
|
|
columns[column][rowNumber] = token;
|
|
}
|
|
token = '';
|
|
++column;
|
|
++actualColumn;
|
|
}
|
|
if (!columnStr.trim().length) {
|
|
return;
|
|
}
|
|
if (columnStr.trim()[0] === '#') {
|
|
return;
|
|
}
|
|
for (; i < columnStr.length; i++) {
|
|
read(i);
|
|
// Quoted string
|
|
if (c === '#') {
|
|
// The rest of the row is a comment
|
|
push();
|
|
return;
|
|
}
|
|
if (c === '"') {
|
|
read(++i);
|
|
while (i < columnStr.length) {
|
|
if (c === '"' && cl !== '"' && cn !== '"') {
|
|
break;
|
|
}
|
|
if (c !== '"' || (c === '"' && cl !== '"')) {
|
|
token += c;
|
|
}
|
|
read(++i);
|
|
}
|
|
// Perform "plugin" handling
|
|
}
|
|
else if (callbacks && callbacks[c]) {
|
|
if (callbacks[c](c, token)) {
|
|
push();
|
|
}
|
|
// Delimiter - push current token
|
|
}
|
|
else if (c === itemDelimiter) {
|
|
push();
|
|
// Actual column data
|
|
}
|
|
else {
|
|
token += c;
|
|
}
|
|
}
|
|
push();
|
|
}
|
|
/**
|
|
* Attempt to guess the delimiter. We do a separate parse pass here
|
|
* because we need to count potential delimiters softly without making
|
|
* any assumptions.
|
|
* @private
|
|
*/
|
|
function guessDelimiter(lines) {
|
|
var points = 0, commas = 0, guessed = false;
|
|
lines.some(function (columnStr, i) {
|
|
var inStr = false, c, cn, cl, token = '';
|
|
// We should be able to detect dateformats within 13 rows
|
|
if (i > 13) {
|
|
return true;
|
|
}
|
|
for (var j = 0; j < columnStr.length; j++) {
|
|
c = columnStr[j];
|
|
cn = columnStr[j + 1];
|
|
cl = columnStr[j - 1];
|
|
if (c === '#') {
|
|
// Skip the rest of the line - it's a comment
|
|
return;
|
|
}
|
|
if (c === '"') {
|
|
if (inStr) {
|
|
if (cl !== '"' && cn !== '"') {
|
|
while (cn === ' ' && j < columnStr.length) {
|
|
cn = columnStr[++j];
|
|
}
|
|
// After parsing a string, the next non-blank
|
|
// should be a delimiter if the CSV is properly
|
|
// formed.
|
|
if (typeof potDelimiters[cn] !== 'undefined') {
|
|
potDelimiters[cn]++;
|
|
}
|
|
inStr = false;
|
|
}
|
|
}
|
|
else {
|
|
inStr = true;
|
|
}
|
|
}
|
|
else if (typeof potDelimiters[c] !== 'undefined') {
|
|
token = token.trim();
|
|
if (!isNaN(Date.parse(token))) {
|
|
potDelimiters[c]++;
|
|
}
|
|
else if (isNaN(token) ||
|
|
!isFinite(token)) {
|
|
potDelimiters[c]++;
|
|
}
|
|
token = '';
|
|
}
|
|
else {
|
|
token += c;
|
|
}
|
|
if (c === ',') {
|
|
commas++;
|
|
}
|
|
if (c === '.') {
|
|
points++;
|
|
}
|
|
}
|
|
});
|
|
// Count the potential delimiters.
|
|
// This could be improved by checking if the number of delimiters
|
|
// equals the number of columns - 1
|
|
if (potDelimiters[';'] > potDelimiters[',']) {
|
|
guessed = ';';
|
|
}
|
|
else if (potDelimiters[','] > potDelimiters[';']) {
|
|
guessed = ',';
|
|
}
|
|
else {
|
|
// No good guess could be made..
|
|
guessed = ',';
|
|
}
|
|
// Try to deduce the decimal point if it's not explicitly set.
|
|
// If both commas or points is > 0 there is likely an issue
|
|
if (!options.decimalPoint) {
|
|
if (points > commas) {
|
|
options.decimalPoint = '.';
|
|
}
|
|
else {
|
|
options.decimalPoint = ',';
|
|
}
|
|
// Apply a new decimal regex based on the presumed decimal sep.
|
|
self.decimalRegex = new RegExp('^(-?[0-9]+)' +
|
|
options.decimalPoint +
|
|
'([0-9]+)$');
|
|
}
|
|
return guessed;
|
|
}
|
|
/**
|
|
* Tries to guess the date format
|
|
* - Check if either month candidate exceeds 12
|
|
* - Check if year is missing (use current year)
|
|
* - Check if a shortened year format is used (e.g. 1/1/99)
|
|
* - If no guess can be made, the user must be prompted
|
|
* data is the data to deduce a format based on
|
|
* @private
|
|
*/
|
|
function deduceDateFormat(data, limit) {
|
|
var format = 'YYYY/mm/dd', thing, guessedFormat = [], calculatedFormat, i = 0, madeDeduction = false,
|
|
// candidates = {},
|
|
stable = [], max = [], j;
|
|
if (!limit || limit > data.length) {
|
|
limit = data.length;
|
|
}
|
|
for (; i < limit; i++) {
|
|
if (typeof data[i] !== 'undefined' &&
|
|
data[i] && data[i].length) {
|
|
thing = data[i]
|
|
.trim()
|
|
.replace(/\//g, ' ')
|
|
.replace(/\-/g, ' ')
|
|
.replace(/\./g, ' ')
|
|
.split(' ');
|
|
guessedFormat = [
|
|
'',
|
|
'',
|
|
''
|
|
];
|
|
for (j = 0; j < thing.length; j++) {
|
|
if (j < guessedFormat.length) {
|
|
thing[j] = parseInt(thing[j], 10);
|
|
if (thing[j]) {
|
|
max[j] = (!max[j] || max[j] < thing[j]) ?
|
|
thing[j] :
|
|
max[j];
|
|
if (typeof stable[j] !== 'undefined') {
|
|
if (stable[j] !== thing[j]) {
|
|
stable[j] = false;
|
|
}
|
|
}
|
|
else {
|
|
stable[j] = thing[j];
|
|
}
|
|
if (thing[j] > 31) {
|
|
if (thing[j] < 100) {
|
|
guessedFormat[j] = 'YY';
|
|
}
|
|
else {
|
|
guessedFormat[j] = 'YYYY';
|
|
}
|
|
// madeDeduction = true;
|
|
}
|
|
else if (thing[j] > 12 &&
|
|
thing[j] <= 31) {
|
|
guessedFormat[j] = 'dd';
|
|
madeDeduction = true;
|
|
}
|
|
else if (!guessedFormat[j].length) {
|
|
guessedFormat[j] = 'mm';
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
if (madeDeduction) {
|
|
// This handles a few edge cases with hard to guess dates
|
|
for (j = 0; j < stable.length; j++) {
|
|
if (stable[j] !== false) {
|
|
if (max[j] > 12 &&
|
|
guessedFormat[j] !== 'YY' &&
|
|
guessedFormat[j] !== 'YYYY') {
|
|
guessedFormat[j] = 'YY';
|
|
}
|
|
}
|
|
else if (max[j] > 12 && guessedFormat[j] === 'mm') {
|
|
guessedFormat[j] = 'dd';
|
|
}
|
|
}
|
|
// If the middle one is dd, and the last one is dd,
|
|
// the last should likely be year.
|
|
if (guessedFormat.length === 3 &&
|
|
guessedFormat[1] === 'dd' &&
|
|
guessedFormat[2] === 'dd') {
|
|
guessedFormat[2] = 'YY';
|
|
}
|
|
calculatedFormat = guessedFormat.join('/');
|
|
// If the caculated format is not valid, we need to present an
|
|
// error.
|
|
if (!(options.dateFormats || self.dateFormats)[calculatedFormat]) {
|
|
// This should emit an event instead
|
|
fireEvent('deduceDateFailed');
|
|
return format;
|
|
}
|
|
return calculatedFormat;
|
|
}
|
|
return format;
|
|
}
|
|
/**
|
|
* @todo
|
|
* Figure out the best axis types for the data
|
|
* - If the first column is a number, we're good
|
|
* - If the first column is a date, set to date/time
|
|
* - If the first column is a string, set to categories
|
|
* @private
|
|
*/
|
|
function deduceAxisTypes() {
|
|
}
|
|
if (csv && options.beforeParse) {
|
|
csv = options.beforeParse.call(this, csv);
|
|
}
|
|
if (csv) {
|
|
lines = csv
|
|
.replace(/\r\n/g, '\n') // Unix
|
|
.replace(/\r/g, '\n') // Mac
|
|
.split(options.lineDelimiter || '\n');
|
|
if (!startRow || startRow < 0) {
|
|
startRow = 0;
|
|
}
|
|
if (!endRow || endRow >= lines.length) {
|
|
endRow = lines.length - 1;
|
|
}
|
|
if (options.itemDelimiter) {
|
|
itemDelimiter = options.itemDelimiter;
|
|
}
|
|
else {
|
|
itemDelimiter = null;
|
|
itemDelimiter = guessDelimiter(lines);
|
|
}
|
|
var offset = 0;
|
|
for (rowIt = startRow; rowIt <= endRow; rowIt++) {
|
|
if (lines[rowIt][0] === '#') {
|
|
offset++;
|
|
}
|
|
else {
|
|
parseRow(lines[rowIt], rowIt - startRow - offset);
|
|
}
|
|
}
|
|
// //Make sure that there's header columns for everything
|
|
// columns.forEach(function (col) {
|
|
// });
|
|
deduceAxisTypes();
|
|
if ((!options.columnTypes || options.columnTypes.length === 0) &&
|
|
dataTypes.length &&
|
|
dataTypes[0].length &&
|
|
dataTypes[0][1] === 'date' &&
|
|
!options.dateFormat) {
|
|
options.dateFormat = deduceDateFormat(columns[0]);
|
|
}
|
|
// lines.forEach(function (line, rowNo) {
|
|
// var trimmed = self.trim(line),
|
|
// isComment = trimmed.indexOf('#') === 0,
|
|
// isBlank = trimmed === '',
|
|
// items;
|
|
// if (
|
|
// rowNo >= startRow &&
|
|
// rowNo <= endRow &&
|
|
// !isComment && !isBlank
|
|
// ) {
|
|
// items = line.split(itemDelimiter);
|
|
// items.forEach(function (item, colNo) {
|
|
// if (colNo >= startColumn && colNo <= endColumn) {
|
|
// if (!columns[colNo - startColumn]) {
|
|
// columns[colNo - startColumn] = [];
|
|
// }
|
|
// columns[colNo - startColumn][activeRowNo] = item;
|
|
// }
|
|
// });
|
|
// activeRowNo += 1;
|
|
// }
|
|
// });
|
|
//
|
|
this.dataFound();
|
|
}
|
|
return columns;
|
|
};
|
|
/**
|
|
* Parse a HTML table
|
|
*
|
|
* @function Highcharts.Data#parseTable
|
|
*
|
|
* @return {Array<Array<Highcharts.DataValueType>>}
|
|
*/
|
|
Data.prototype.parseTable = function () {
|
|
var options = this.options, table = options.table, columns = this.columns || [], startRow = options.startRow || 0, endRow = options.endRow || Number.MAX_VALUE, startColumn = options.startColumn || 0, endColumn = options.endColumn || Number.MAX_VALUE;
|
|
if (table) {
|
|
if (typeof table === 'string') {
|
|
table = doc.getElementById(table);
|
|
}
|
|
[].forEach.call(table.getElementsByTagName('tr'), function (tr, rowNo) {
|
|
if (rowNo >= startRow && rowNo <= endRow) {
|
|
[].forEach.call(tr.children, function (item, colNo) {
|
|
var row = columns[colNo - startColumn];
|
|
var i = 1;
|
|
if ((item.tagName === 'TD' ||
|
|
item.tagName === 'TH') &&
|
|
colNo >= startColumn &&
|
|
colNo <= endColumn) {
|
|
if (!columns[colNo - startColumn]) {
|
|
columns[colNo - startColumn] = [];
|
|
}
|
|
columns[colNo - startColumn][rowNo - startRow] = item.innerHTML;
|
|
// Loop over all previous indices and make sure
|
|
// they are nulls, not undefined.
|
|
while (rowNo - startRow >= i &&
|
|
row[rowNo - startRow - i] === void 0) {
|
|
row[rowNo - startRow - i] = null;
|
|
i++;
|
|
}
|
|
}
|
|
});
|
|
}
|
|
});
|
|
this.dataFound(); // continue
|
|
}
|
|
return columns;
|
|
};
|
|
/**
|
|
* Fetch or refetch live data
|
|
*
|
|
* @function Highcharts.Data#fetchLiveData
|
|
*
|
|
* @return {boolean}
|
|
* The URLs that were tried can be found in the options
|
|
*/
|
|
Data.prototype.fetchLiveData = function () {
|
|
var data = this, chart = this.chart, options = this.options, maxRetries = 3, currentRetries = 0, pollingEnabled = options.enablePolling, updateIntervalMs = (options.dataRefreshRate || 2) * 1000, originalOptions = merge(options);
|
|
if (!this.hasURLOption(options)) {
|
|
return false;
|
|
}
|
|
// Do not allow polling more than once a second
|
|
if (updateIntervalMs < 1000) {
|
|
updateIntervalMs = 1000;
|
|
}
|
|
delete options.csvURL;
|
|
delete options.rowsURL;
|
|
delete options.columnsURL;
|
|
/**
|
|
* @private
|
|
*/
|
|
function performFetch(initialFetch) {
|
|
/**
|
|
* Helper function for doing the data fetch + polling.
|
|
* @private
|
|
*/
|
|
function request(url, done, tp) {
|
|
if (!url || url.indexOf('http') !== 0) {
|
|
if (url && options.error) {
|
|
options.error('Invalid URL');
|
|
}
|
|
return false;
|
|
}
|
|
if (initialFetch) {
|
|
clearTimeout(data.liveDataTimeout);
|
|
chart.liveDataURL = url;
|
|
}
|
|
/**
|
|
* @private
|
|
*/
|
|
function poll() {
|
|
// Poll
|
|
if (pollingEnabled && chart.liveDataURL === url) {
|
|
// We need to stop doing this if the URL has changed
|
|
data.liveDataTimeout =
|
|
setTimeout(performFetch, updateIntervalMs);
|
|
}
|
|
}
|
|
Highcharts.ajax({
|
|
url: url,
|
|
dataType: tp || 'json',
|
|
success: function (res) {
|
|
if (chart && chart.series) {
|
|
done(res);
|
|
}
|
|
poll();
|
|
},
|
|
error: function (xhr, text) {
|
|
if (++currentRetries < maxRetries) {
|
|
poll();
|
|
}
|
|
return options.error && options.error(text, xhr);
|
|
}
|
|
});
|
|
return true;
|
|
}
|
|
if (!request(originalOptions.csvURL, function (res) {
|
|
chart.update({
|
|
data: {
|
|
csv: res
|
|
}
|
|
});
|
|
}, 'text')) {
|
|
if (!request(originalOptions.rowsURL, function (res) {
|
|
chart.update({
|
|
data: {
|
|
rows: res
|
|
}
|
|
});
|
|
})) {
|
|
request(originalOptions.columnsURL, function (res) {
|
|
chart.update({
|
|
data: {
|
|
columns: res
|
|
}
|
|
});
|
|
});
|
|
}
|
|
}
|
|
}
|
|
performFetch(true);
|
|
return this.hasURLOption(options);
|
|
};
|
|
/**
|
|
* Parse a Google spreadsheet.
|
|
*
|
|
* @function Highcharts.Data#parseGoogleSpreadsheet
|
|
*
|
|
* @return {boolean}
|
|
* Always returns false, because it is an intermediate fetch.
|
|
*/
|
|
Data.prototype.parseGoogleSpreadsheet = function () {
|
|
var data = this, options = this.options, googleSpreadsheetKey = options.googleSpreadsheetKey, chart = this.chart,
|
|
// use sheet 1 as the default rather than od6
|
|
// as the latter sometimes cause issues (it looks like it can
|
|
// be renamed in some cases, ref. a fogbugz case).
|
|
worksheet = options.googleSpreadsheetWorksheet || 1, startRow = options.startRow || 0, endRow = options.endRow || Number.MAX_VALUE, startColumn = options.startColumn || 0, endColumn = options.endColumn || Number.MAX_VALUE, refreshRate = (options.dataRefreshRate || 2) * 1000;
|
|
if (refreshRate < 4000) {
|
|
refreshRate = 4000;
|
|
}
|
|
/**
|
|
* Fetch the actual spreadsheet using XMLHttpRequest.
|
|
* @private
|
|
*/
|
|
function fetchSheet(fn) {
|
|
var url = [
|
|
'https://spreadsheets.google.com/feeds/cells',
|
|
googleSpreadsheetKey,
|
|
worksheet,
|
|
'public/values?alt=json'
|
|
].join('/');
|
|
Highcharts.ajax({
|
|
url: url,
|
|
dataType: 'json',
|
|
success: function (json) {
|
|
fn(json);
|
|
if (options.enablePolling) {
|
|
setTimeout(function () {
|
|
fetchSheet(fn);
|
|
}, (options.dataRefreshRate || 2) * 1000);
|
|
}
|
|
},
|
|
error: function (xhr, text) {
|
|
return options.error && options.error(text, xhr);
|
|
}
|
|
});
|
|
}
|
|
if (googleSpreadsheetKey) {
|
|
delete options.googleSpreadsheetKey;
|
|
fetchSheet(function (json) {
|
|
// Prepare the data from the spreadsheat
|
|
var columns = [], cells = json.feed.entry, cell, cellCount = (cells || []).length, colCount = 0, rowCount = 0, val, gr, gc, cellInner, i;
|
|
if (!cells || cells.length === 0) {
|
|
return false;
|
|
}
|
|
// First, find the total number of columns and rows that
|
|
// are actually filled with data
|
|
for (i = 0; i < cellCount; i++) {
|
|
cell = cells[i];
|
|
colCount = Math.max(colCount, cell.gs$cell.col);
|
|
rowCount = Math.max(rowCount, cell.gs$cell.row);
|
|
}
|
|
// Set up arrays containing the column data
|
|
for (i = 0; i < colCount; i++) {
|
|
if (i >= startColumn && i <= endColumn) {
|
|
// Create new columns with the length of either
|
|
// end-start or rowCount
|
|
columns[i - startColumn] = [];
|
|
}
|
|
}
|
|
// Loop over the cells and assign the value to the right
|
|
// place in the column arrays
|
|
for (i = 0; i < cellCount; i++) {
|
|
cell = cells[i];
|
|
gr = cell.gs$cell.row - 1; // rows start at 1
|
|
gc = cell.gs$cell.col - 1; // columns start at 1
|
|
// If both row and col falls inside start and end set the
|
|
// transposed cell value in the newly created columns
|
|
if (gc >= startColumn && gc <= endColumn &&
|
|
gr >= startRow && gr <= endRow) {
|
|
cellInner = cell.gs$cell || cell.content;
|
|
val = null;
|
|
if (cellInner.numericValue) {
|
|
if (cellInner.$t.indexOf('/') >= 0 ||
|
|
cellInner.$t.indexOf('-') >= 0) {
|
|
// This is a date - for future reference.
|
|
val = cellInner.$t;
|
|
}
|
|
else if (cellInner.$t.indexOf('%') > 0) {
|
|
// Percentage
|
|
val = parseFloat(cellInner.numericValue) * 100;
|
|
}
|
|
else {
|
|
val = parseFloat(cellInner.numericValue);
|
|
}
|
|
}
|
|
else if (cellInner.$t && cellInner.$t.length) {
|
|
val = cellInner.$t;
|
|
}
|
|
columns[gc - startColumn][gr - startRow] = val;
|
|
}
|
|
}
|
|
// Insert null for empty spreadsheet cells (#5298)
|
|
columns.forEach(function (column) {
|
|
for (i = 0; i < column.length; i++) {
|
|
if (typeof column[i] === 'undefined') {
|
|
column[i] = null;
|
|
}
|
|
}
|
|
});
|
|
if (chart && chart.series) {
|
|
chart.update({
|
|
data: {
|
|
columns: columns
|
|
}
|
|
});
|
|
}
|
|
else { // #8245
|
|
data.columns = columns;
|
|
data.dataFound();
|
|
}
|
|
});
|
|
}
|
|
// This is an intermediate fetch, so always return false.
|
|
return false;
|
|
};
|
|
/**
|
|
* Trim a string from whitespaces.
|
|
*
|
|
* @function Highcharts.Data#trim
|
|
*
|
|
* @param {string} str
|
|
* String to trim
|
|
*
|
|
* @param {boolean} [inside=false]
|
|
* Remove all spaces between numbers.
|
|
*
|
|
* @return {string}
|
|
* Trimed string
|
|
*/
|
|
Data.prototype.trim = function (str, inside) {
|
|
if (typeof str === 'string') {
|
|
str = str.replace(/^\s+|\s+$/g, '');
|
|
// Clear white space insdie the string, like thousands separators
|
|
if (inside && /^[0-9\s]+$/.test(str)) {
|
|
str = str.replace(/\s/g, '');
|
|
}
|
|
if (this.decimalRegex) {
|
|
str = str.replace(this.decimalRegex, '$1.$2');
|
|
}
|
|
}
|
|
return str;
|
|
};
|
|
/**
|
|
* Parse numeric cells in to number types and date types in to true dates.
|
|
*
|
|
* @function Highcharts.Data#parseTypes
|
|
*/
|
|
Data.prototype.parseTypes = function () {
|
|
var columns = this.columns, col = columns.length;
|
|
while (col--) {
|
|
this.parseColumn(columns[col], col);
|
|
}
|
|
};
|
|
/**
|
|
* Parse a single column. Set properties like .isDatetime and .isNumeric.
|
|
*
|
|
* @function Highcharts.Data#parseColumn
|
|
*
|
|
* @param {Array<Highcharts.DataValueType>} column
|
|
* Column to parse
|
|
*
|
|
* @param {number} col
|
|
* Column index
|
|
*/
|
|
Data.prototype.parseColumn = function (column, col) {
|
|
var rawColumns = this.rawColumns, columns = this.columns, row = column.length, val, floatVal, trimVal, trimInsideVal, firstRowAsNames = this.firstRowAsNames, isXColumn = this.valueCount.xColumns.indexOf(col) !== -1, dateVal, backup = [], diff, chartOptions = this.chartOptions, descending, columnTypes = this.options.columnTypes || [], columnType = columnTypes[col], forceCategory = isXColumn && ((chartOptions &&
|
|
chartOptions.xAxis &&
|
|
splat(chartOptions.xAxis)[0].type === 'category') || columnType === 'string');
|
|
if (!rawColumns[col]) {
|
|
rawColumns[col] = [];
|
|
}
|
|
while (row--) {
|
|
val = backup[row] || column[row];
|
|
trimVal = this.trim(val);
|
|
trimInsideVal = this.trim(val, true);
|
|
floatVal = parseFloat(trimInsideVal);
|
|
// Set it the first time
|
|
if (typeof rawColumns[col][row] === 'undefined') {
|
|
rawColumns[col][row] = trimVal;
|
|
}
|
|
// Disable number or date parsing by setting the X axis type to
|
|
// category
|
|
if (forceCategory || (row === 0 && firstRowAsNames)) {
|
|
column[row] = '' + trimVal;
|
|
}
|
|
else if (+trimInsideVal === floatVal) { // is numeric
|
|
column[row] = floatVal;
|
|
// If the number is greater than milliseconds in a year, assume
|
|
// datetime
|
|
if (floatVal > 365 * 24 * 3600 * 1000 &&
|
|
columnType !== 'float') {
|
|
column.isDatetime = true;
|
|
}
|
|
else {
|
|
column.isNumeric = true;
|
|
}
|
|
if (typeof column[row + 1] !== 'undefined') {
|
|
descending = floatVal > column[row + 1];
|
|
}
|
|
// String, continue to determine if it is a date string or really a
|
|
// string
|
|
}
|
|
else {
|
|
if (trimVal && trimVal.length) {
|
|
dateVal = this.parseDate(val);
|
|
}
|
|
// Only allow parsing of dates if this column is an x-column
|
|
if (isXColumn && isNumber(dateVal) && columnType !== 'float') {
|
|
backup[row] = val;
|
|
column[row] = dateVal;
|
|
column.isDatetime = true;
|
|
// Check if the dates are uniformly descending or ascending.
|
|
// If they are not, chances are that they are a different
|
|
// time format, so check for alternative.
|
|
if (typeof column[row + 1] !== 'undefined') {
|
|
diff = dateVal > column[row + 1];
|
|
if (diff !== descending &&
|
|
typeof descending !== 'undefined') {
|
|
if (this.alternativeFormat) {
|
|
this.dateFormat = this.alternativeFormat;
|
|
row = column.length;
|
|
this.alternativeFormat =
|
|
this.dateFormats[this.dateFormat]
|
|
.alternative;
|
|
}
|
|
else {
|
|
column.unsorted = true;
|
|
}
|
|
}
|
|
descending = diff;
|
|
}
|
|
}
|
|
else { // string
|
|
column[row] = trimVal === '' ? null : trimVal;
|
|
if (row !== 0 &&
|
|
(column.isDatetime ||
|
|
column.isNumeric)) {
|
|
column.mixed = true;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
// If strings are intermixed with numbers or dates in a parsed column,
|
|
// it is an indication that parsing went wrong or the data was not
|
|
// intended to display as numbers or dates and parsing is too
|
|
// aggressive. Fall back to categories. Demonstrated in the
|
|
// highcharts/demo/column-drilldown sample.
|
|
if (isXColumn && column.mixed) {
|
|
columns[col] = rawColumns[col];
|
|
}
|
|
// If the 0 column is date or number and descending, reverse all
|
|
// columns.
|
|
if (isXColumn && descending && this.options.sort) {
|
|
for (col = 0; col < columns.length; col++) {
|
|
columns[col].reverse();
|
|
if (firstRowAsNames) {
|
|
columns[col].unshift(columns[col].pop());
|
|
}
|
|
}
|
|
}
|
|
};
|
|
/**
|
|
* Parse a date and return it as a number. Overridable through
|
|
* `options.parseDate`.
|
|
*
|
|
* @function Highcharts.Data#parseDate
|
|
*
|
|
* @param {string} val
|
|
*
|
|
* @return {number}
|
|
*/
|
|
Data.prototype.parseDate = function (val) {
|
|
var parseDate = this.options.parseDate, ret, key, format, dateFormat = this.options.dateFormat || this.dateFormat, match;
|
|
if (parseDate) {
|
|
ret = parseDate(val);
|
|
}
|
|
else if (typeof val === 'string') {
|
|
// Auto-detect the date format the first time
|
|
if (!dateFormat) {
|
|
for (key in this.dateFormats) { // eslint-disable-line guard-for-in
|
|
format = this.dateFormats[key];
|
|
match = val.match(format.regex);
|
|
if (match) {
|
|
this.dateFormat = dateFormat = key;
|
|
this.alternativeFormat = format.alternative;
|
|
ret = format.parser(match);
|
|
break;
|
|
}
|
|
}
|
|
// Next time, use the one previously found
|
|
}
|
|
else {
|
|
format = this.dateFormats[dateFormat];
|
|
if (!format) {
|
|
// The selected format is invalid
|
|
format = this.dateFormats['YYYY/mm/dd'];
|
|
}
|
|
match = val.match(format.regex);
|
|
if (match) {
|
|
ret = format.parser(match);
|
|
}
|
|
}
|
|
// Fall back to Date.parse
|
|
if (!match) {
|
|
match = Date.parse(val);
|
|
// External tools like Date.js and MooTools extend Date object
|
|
// and returns a date.
|
|
if (typeof match === 'object' &&
|
|
match !== null &&
|
|
match.getTime) {
|
|
ret = (match.getTime() -
|
|
match.getTimezoneOffset() *
|
|
60000);
|
|
// Timestamp
|
|
}
|
|
else if (isNumber(match)) {
|
|
ret = match - (new Date(match)).getTimezoneOffset() * 60000;
|
|
}
|
|
}
|
|
}
|
|
return ret;
|
|
};
|
|
/**
|
|
* Reorganize rows into columns.
|
|
*
|
|
* @function Highcharts.Data#rowsToColumns
|
|
*
|
|
* @param {Array<Array<Highcharts.DataValueType>>} rows
|
|
*
|
|
* @return {Array<Array<Highcharts.DataValueType>>|undefined}
|
|
*/
|
|
Data.prototype.rowsToColumns = function (rows) {
|
|
var row, rowsLength, col, colsLength, columns;
|
|
if (rows) {
|
|
columns = [];
|
|
rowsLength = rows.length;
|
|
for (row = 0; row < rowsLength; row++) {
|
|
colsLength = rows[row].length;
|
|
for (col = 0; col < colsLength; col++) {
|
|
if (!columns[col]) {
|
|
columns[col] = [];
|
|
}
|
|
columns[col][row] = rows[row][col];
|
|
}
|
|
}
|
|
}
|
|
return columns;
|
|
};
|
|
/**
|
|
* Get the parsed data in a form that we can apply directly to the
|
|
* `series.data` config. Array positions can be mapped using the
|
|
* `series.keys` option.
|
|
*
|
|
* @example
|
|
* const data = Highcharts.data({
|
|
* csv: document.getElementById('data').innerHTML
|
|
* }).getData();
|
|
*
|
|
* @function Highcharts.Data#getData
|
|
*
|
|
* @return {Array<Array<(number|string)>>|undefined} Data rows
|
|
*/
|
|
Data.prototype.getData = function () {
|
|
if (this.columns) {
|
|
return this.rowsToColumns(this.columns).slice(1);
|
|
}
|
|
};
|
|
/**
|
|
* A hook for working directly on the parsed columns
|
|
*
|
|
* @function Highcharts.Data#parsed
|
|
*
|
|
* @return {boolean|undefined}
|
|
*/
|
|
Data.prototype.parsed = function () {
|
|
if (this.options.parsed) {
|
|
return this.options.parsed.call(this, this.columns);
|
|
}
|
|
};
|
|
/**
|
|
* @private
|
|
* @function Highcharts.Data#getFreeIndexes
|
|
*/
|
|
Data.prototype.getFreeIndexes = function (numberOfColumns, seriesBuilders) {
|
|
var s, i, freeIndexes = [], freeIndexValues = [], referencedIndexes;
|
|
// Add all columns as free
|
|
for (i = 0; i < numberOfColumns; i = i + 1) {
|
|
freeIndexes.push(true);
|
|
}
|
|
// Loop all defined builders and remove their referenced columns
|
|
for (s = 0; s < seriesBuilders.length; s = s + 1) {
|
|
referencedIndexes = seriesBuilders[s].getReferencedColumnIndexes();
|
|
for (i = 0; i < referencedIndexes.length; i = i + 1) {
|
|
freeIndexes[referencedIndexes[i]] = false;
|
|
}
|
|
}
|
|
// Collect the values for the free indexes
|
|
for (i = 0; i < freeIndexes.length; i = i + 1) {
|
|
if (freeIndexes[i]) {
|
|
freeIndexValues.push(i);
|
|
}
|
|
}
|
|
return freeIndexValues;
|
|
};
|
|
/**
|
|
* If a complete callback function is provided in the options, interpret the
|
|
* columns into a Highcharts options object.
|
|
*
|
|
* @function Highcharts.Data#complete
|
|
*/
|
|
Data.prototype.complete = function () {
|
|
var columns = this.columns, xColumns = [], type, options = this.options, series, data, i, j, r, seriesIndex, chartOptions, allSeriesBuilders = [], builder, freeIndexes, typeCol, index;
|
|
xColumns.length = columns.length;
|
|
if (options.complete || options.afterComplete) {
|
|
// Get the names and shift the top row
|
|
if (this.firstRowAsNames) {
|
|
for (i = 0; i < columns.length; i++) {
|
|
columns[i].name = columns[i].shift();
|
|
}
|
|
}
|
|
// Use the next columns for series
|
|
series = [];
|
|
freeIndexes = this.getFreeIndexes(columns.length, this.valueCount.seriesBuilders);
|
|
// Populate defined series
|
|
for (seriesIndex = 0; seriesIndex < this.valueCount.seriesBuilders.length; seriesIndex++) {
|
|
builder = this.valueCount.seriesBuilders[seriesIndex];
|
|
// If the builder can be populated with remaining columns, then
|
|
// add it to allBuilders
|
|
if (builder.populateColumns(freeIndexes)) {
|
|
allSeriesBuilders.push(builder);
|
|
}
|
|
}
|
|
// Populate dynamic series
|
|
while (freeIndexes.length > 0) {
|
|
builder = new SeriesBuilder();
|
|
builder.addColumnReader(0, 'x');
|
|
// Mark index as used (not free)
|
|
index = freeIndexes.indexOf(0);
|
|
if (index !== -1) {
|
|
freeIndexes.splice(index, 1);
|
|
}
|
|
for (i = 0; i < this.valueCount.global; i++) {
|
|
// Create and add a column reader for the next free column
|
|
// index
|
|
builder.addColumnReader(void 0, this.valueCount.globalPointArrayMap[i]);
|
|
}
|
|
// If the builder can be populated with remaining columns, then
|
|
// add it to allBuilders
|
|
if (builder.populateColumns(freeIndexes)) {
|
|
allSeriesBuilders.push(builder);
|
|
}
|
|
}
|
|
// Get the data-type from the first series x column
|
|
if (allSeriesBuilders.length > 0 &&
|
|
allSeriesBuilders[0].readers.length > 0) {
|
|
typeCol = columns[allSeriesBuilders[0].readers[0].columnIndex];
|
|
if (typeof typeCol !== 'undefined') {
|
|
if (typeCol.isDatetime) {
|
|
type = 'datetime';
|
|
}
|
|
else if (!typeCol.isNumeric) {
|
|
type = 'category';
|
|
}
|
|
}
|
|
}
|
|
// Axis type is category, then the "x" column should be called
|
|
// "name"
|
|
if (type === 'category') {
|
|
for (seriesIndex = 0; seriesIndex < allSeriesBuilders.length; seriesIndex++) {
|
|
builder = allSeriesBuilders[seriesIndex];
|
|
for (r = 0; r < builder.readers.length; r++) {
|
|
if (builder.readers[r].configName === 'x') {
|
|
builder.readers[r].configName = 'name';
|
|
}
|
|
}
|
|
}
|
|
}
|
|
// Read data for all builders
|
|
for (seriesIndex = 0; seriesIndex < allSeriesBuilders.length; seriesIndex++) {
|
|
builder = allSeriesBuilders[seriesIndex];
|
|
// Iterate down the cells of each column and add data to the
|
|
// series
|
|
data = [];
|
|
for (j = 0; j < columns[0].length; j++) {
|
|
data[j] = builder.read(columns, j);
|
|
}
|
|
// Add the series
|
|
series[seriesIndex] = {
|
|
data: data
|
|
};
|
|
if (builder.name) {
|
|
series[seriesIndex].name = builder.name;
|
|
}
|
|
if (type === 'category') {
|
|
series[seriesIndex].turboThreshold = 0;
|
|
}
|
|
}
|
|
// Do the callback
|
|
chartOptions = {
|
|
series: series
|
|
};
|
|
if (type) {
|
|
chartOptions.xAxis = {
|
|
type: type
|
|
};
|
|
if (type === 'category') {
|
|
chartOptions.xAxis.uniqueNames = false;
|
|
}
|
|
}
|
|
if (options.complete) {
|
|
options.complete(chartOptions);
|
|
}
|
|
// The afterComplete hook is used internally to avoid conflict with
|
|
// the externally available complete option.
|
|
if (options.afterComplete) {
|
|
options.afterComplete(chartOptions);
|
|
}
|
|
}
|
|
};
|
|
/**
|
|
* Updates the chart with new data options.
|
|
*
|
|
* @function Highcharts.Data#update
|
|
*
|
|
* @param {Highcharts.DataOptions} options
|
|
*
|
|
* @param {boolean} [redraw=true]
|
|
*/
|
|
Data.prototype.update = function (options, redraw) {
|
|
var chart = this.chart;
|
|
if (options) {
|
|
// Set the complete handler
|
|
options.afterComplete = function (dataOptions) {
|
|
// Avoid setting axis options unless the type changes. Running
|
|
// Axis.update will cause the whole structure to be destroyed
|
|
// and rebuilt, and animation is lost.
|
|
if (dataOptions) {
|
|
if (dataOptions.xAxis &&
|
|
chart.xAxis[0] &&
|
|
dataOptions.xAxis.type ===
|
|
chart.xAxis[0].options.type) {
|
|
delete dataOptions.xAxis;
|
|
}
|
|
// @todo looks not right:
|
|
chart.update(dataOptions, redraw, true);
|
|
}
|
|
};
|
|
// Apply it
|
|
merge(true, chart.options.data, options);
|
|
this.init(chart.options.data);
|
|
}
|
|
};
|
|
return Data;
|
|
}());
|
|
// Register the Data prototype and data function on Highcharts
|
|
// Highcharts.Data = Data as any;
|
|
/**
|
|
* Creates a data object to parse data for a chart.
|
|
*
|
|
* @function Highcharts.data
|
|
*
|
|
* @param {Highcharts.DataOptions} dataOptions
|
|
*
|
|
* @param {Highcharts.Options} [chartOptions]
|
|
*
|
|
* @param {Highcharts.Chart} [chart]
|
|
*
|
|
* @return {Highcharts.Data}
|
|
*/
|
|
H.data = function (dataOptions, chartOptions, chart) {
|
|
return new H.Data(dataOptions, chartOptions, chart);
|
|
};
|
|
// Extend Chart.init so that the Chart constructor accepts a new configuration
|
|
// option group, data.
|
|
addEvent(Chart, 'init', function (e) {
|
|
var chart = this, // eslint-disable-line no-invalid-this
|
|
userOptions = (e.args[0] || {}), callback = e.args[1];
|
|
if (userOptions && userOptions.data && !chart.hasDataDef) {
|
|
chart.hasDataDef = true;
|
|
/**
|
|
* The data parser for this chart.
|
|
*
|
|
* @name Highcharts.Chart#data
|
|
* @type {Highcharts.Data|undefined}
|
|
*/
|
|
chart.data = new H.Data(extend(userOptions.data, {
|
|
afterComplete: function (dataOptions) {
|
|
var i, series;
|
|
// Merge series configs
|
|
if (Object.hasOwnProperty.call(userOptions, 'series')) {
|
|
if (typeof userOptions.series === 'object') {
|
|
i = Math.max(userOptions.series.length, dataOptions && dataOptions.series ?
|
|
dataOptions.series.length :
|
|
0);
|
|
while (i--) {
|
|
series = userOptions.series[i] || {};
|
|
userOptions.series[i] = merge(series, dataOptions && dataOptions.series ?
|
|
dataOptions.series[i] :
|
|
{});
|
|
}
|
|
}
|
|
else { // Allow merging in dataOptions.series (#2856)
|
|
delete userOptions.series;
|
|
}
|
|
}
|
|
// Do the merge
|
|
userOptions = merge(dataOptions, userOptions);
|
|
// Run chart.init again
|
|
chart.init(userOptions, callback);
|
|
}
|
|
}), userOptions, chart);
|
|
e.preventDefault();
|
|
}
|
|
});
|
|
/**
|
|
* Creates a new SeriesBuilder. A SeriesBuilder consists of a number
|
|
* of ColumnReaders that reads columns and give them a name.
|
|
* Ex: A series builder can be constructed to read column 3 as 'x' and
|
|
* column 7 and 8 as 'y1' and 'y2'.
|
|
* The output would then be points/rows of the form {x: 11, y1: 22, y2: 33}
|
|
*
|
|
* The name of the builder is taken from the second column. In the above
|
|
* example it would be the column with index 7.
|
|
*
|
|
* @private
|
|
* @class
|
|
* @name SeriesBuilder
|
|
*/
|
|
var SeriesBuilder = /** @class */ (function () {
|
|
function SeriesBuilder() {
|
|
/* eslint-disable no-invalid-this */
|
|
this.readers = [];
|
|
this.pointIsArray = true;
|
|
/* eslint-enable no-invalid-this */
|
|
this.name = void 0;
|
|
}
|
|
/**
|
|
* Populates readers with column indexes. A reader can be added without
|
|
* a specific index and for those readers the index is taken sequentially
|
|
* from the free columns (this is handled by the ColumnCursor instance).
|
|
*
|
|
* @function SeriesBuilder#populateColumns
|
|
*
|
|
* @param {Array<number>} freeIndexes
|
|
*
|
|
* @returns {boolean}
|
|
*/
|
|
SeriesBuilder.prototype.populateColumns = function (freeIndexes) {
|
|
var builder = this, enoughColumns = true;
|
|
// Loop each reader and give it an index if its missing.
|
|
// The freeIndexes.shift() will return undefined if there
|
|
// are no more columns.
|
|
builder.readers.forEach(function (reader) {
|
|
if (typeof reader.columnIndex === 'undefined') {
|
|
reader.columnIndex = freeIndexes.shift();
|
|
}
|
|
});
|
|
// Now, all readers should have columns mapped. If not
|
|
// then return false to signal that this series should
|
|
// not be added.
|
|
builder.readers.forEach(function (reader) {
|
|
if (typeof reader.columnIndex === 'undefined') {
|
|
enoughColumns = false;
|
|
}
|
|
});
|
|
return enoughColumns;
|
|
};
|
|
/**
|
|
* Reads a row from the dataset and returns a point or array depending
|
|
* on the names of the readers.
|
|
*
|
|
* @function SeriesBuilder#read<T>
|
|
*
|
|
* @param {Array<Array<T>>} columns
|
|
*
|
|
* @param {number} rowIndex
|
|
*
|
|
* @returns {Array<T>|Highcharts.Dictionary<T>}
|
|
*/
|
|
SeriesBuilder.prototype.read = function (columns, rowIndex) {
|
|
var builder = this, pointIsArray = builder.pointIsArray, point = pointIsArray ? [] : {}, columnIndexes;
|
|
// Loop each reader and ask it to read its value.
|
|
// Then, build an array or point based on the readers names.
|
|
builder.readers.forEach(function (reader) {
|
|
var value = columns[reader.columnIndex][rowIndex];
|
|
if (pointIsArray) {
|
|
point.push(value);
|
|
}
|
|
else {
|
|
if (reader.configName.indexOf('.') > 0) {
|
|
// Handle nested property names
|
|
Point.prototype.setNestedProperty(point, value, reader.configName);
|
|
}
|
|
else {
|
|
point[reader.configName] = value;
|
|
}
|
|
}
|
|
});
|
|
// The name comes from the first column (excluding the x column)
|
|
if (typeof this.name === 'undefined' && builder.readers.length >= 2) {
|
|
columnIndexes = builder.getReferencedColumnIndexes();
|
|
if (columnIndexes.length >= 2) {
|
|
// remove the first one (x col)
|
|
columnIndexes.shift();
|
|
// Sort the remaining
|
|
columnIndexes.sort(function (a, b) {
|
|
return a - b;
|
|
});
|
|
// Now use the lowest index as name column
|
|
this.name = columns[columnIndexes.shift()].name;
|
|
}
|
|
}
|
|
return point;
|
|
};
|
|
/**
|
|
* Creates and adds ColumnReader from the given columnIndex and configName.
|
|
* ColumnIndex can be undefined and in that case the reader will be given
|
|
* an index when columns are populated.
|
|
*
|
|
* @function SeriesBuilder#addColumnReader
|
|
*
|
|
* @param {number} columnIndex
|
|
*
|
|
* @param {string} configName
|
|
*/
|
|
SeriesBuilder.prototype.addColumnReader = function (columnIndex, configName) {
|
|
this.readers.push({
|
|
columnIndex: columnIndex,
|
|
configName: configName
|
|
});
|
|
if (!(configName === 'x' ||
|
|
configName === 'y' ||
|
|
typeof configName === 'undefined')) {
|
|
this.pointIsArray = false;
|
|
}
|
|
};
|
|
/**
|
|
* Returns an array of column indexes that the builder will use when
|
|
* reading data.
|
|
*
|
|
* @function SeriesBuilder#getReferencedColumnIndexes
|
|
*
|
|
* @returns {Array<number>}
|
|
*/
|
|
SeriesBuilder.prototype.getReferencedColumnIndexes = function () {
|
|
var i, referencedColumnIndexes = [], columnReader;
|
|
for (i = 0; i < this.readers.length; i = i + 1) {
|
|
columnReader = this.readers[i];
|
|
if (typeof columnReader.columnIndex !== 'undefined') {
|
|
referencedColumnIndexes.push(columnReader.columnIndex);
|
|
}
|
|
}
|
|
return referencedColumnIndexes;
|
|
};
|
|
/**
|
|
* Returns true if the builder has a reader for the given configName.
|
|
*
|
|
* @function SeriesBuider#hasReader
|
|
*
|
|
* @param {string} configName
|
|
*
|
|
* @returns {boolean|undefined}
|
|
*/
|
|
SeriesBuilder.prototype.hasReader = function (configName) {
|
|
var i, columnReader;
|
|
for (i = 0; i < this.readers.length; i = i + 1) {
|
|
columnReader = this.readers[i];
|
|
if (columnReader.configName === configName) {
|
|
return true;
|
|
}
|
|
}
|
|
// Else return undefined
|
|
};
|
|
return SeriesBuilder;
|
|
}());
|
|
H.Data = Data;
|
|
|
|
return H.Data;
|
|
});
|
|
_registerModule(_modules, 'masters/modules/data.src.js', [], function () {
|
|
|
|
|
|
});
|
|
})); |