Created
May 1, 2023 00:49
-
-
Save vitouXY/7ba46e0466df83ab543265bdd1203771 to your computer and use it in GitHub Desktop.
API - Bolsa de Santiago de Chile - AppScript & SpreadSheets
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
/* Google SpreadSheets & AppScript | |
https://script.google.com/home/my | |
https://docs.google.com/spreadsheets/u/0/ | |
https://github.com/bradjasper/ImportJSON/raw/master/ImportJSON.gs | |
*/ | |
function getBCS_PrecioNemo(NEMO) { | |
//const NEMO = 'CFMDIVO'.toUpperCase(); | |
const RESPUESTA = UrlFetchApp.fetch(`https://www.bolsadesantiago.com/api/RV_Instrumentos/getPointIntradayGAT?nemo=${NEMO.toUpperCase()}&frecuencia=1`,{muteHttpExceptions: true}).getContentText(); | |
try {JSON.parse(RESPUESTA);} catch(e) {return 0;} | |
const PRECIO = JSON.parse(RESPUESTA).listaResult[0].pre_cie; | |
//console.log(PRECIO); | |
return PRECIO; | |
} | |
function getBCS_PrecioNemoYF(NEMO) { | |
//const NEMO = 'CFMDIVO'.toUpperCase(); | |
const RESPUESTA = UrlFetchApp.fetch(`https://query2.finance.yahoo.com/v7/finance/quote?symbols=${NEMO.toUpperCase().replace(' ','-')}.SN&fields=regularMarketPrice,regularMarketPreviousClose`,{muteHttpExceptions: true}).getContentText(); | |
try {JSON.parse(RESPUESTA);} catch(e) {return 0;} | |
const PRECIO = JSON.parse(RESPUESTA).quoteResponse.result[0].regularMarketPrice; | |
//console.log(PRECIO); | |
return PRECIO; | |
} | |
function getBCS_PointIntradayGAT(NEMO, query, parseOptions) { | |
var url = `https://www.bolsadesantiago.com/api/RV_Instrumentos/getPointIntradayGAT?nemo=${NEMO.toUpperCase()}&frecuencia=1` | |
return ImportJSONAdvanced(url, null, query, parseOptions, includeXPath_, defaultTransform_); | |
} | |
function getBCS_getPointHistGAT(NEMO, query, parseOptions) { | |
var url = `https://www.bolsadesantiago.com/api/RV_Instrumentos/getPointHistGAT?nemo=${NEMO.toUpperCase()}` | |
return ImportJSONAdvanced(url, null, query, parseOptions, includeXPath_, defaultTransform_); | |
} | |
function getBCS_ResumenInstrumento(NEMO, query, parseOptions) { | |
var url = 'https://www.bolsadesantiago.com/api/RV_Instrumentos/getResumenInstrumento'; | |
var data = {'nemo': NEMO.toUpperCase()}; | |
return getBCS_(url, data, query, parseOptions); | |
} | |
function getBCS_ResumenPrecios(NEMO, query, parseOptions) { | |
var url = 'https://www.bolsadesantiago.com/api/RV_Instrumentos/getResumenPrecios'; | |
var data = {'nemo': NEMO.toUpperCase()}; | |
return getBCS_(url, data, query, parseOptions); | |
} | |
function getBCS_UltimasTransacciones(NEMO, query, parseOptions) { | |
var url = 'https://www.bolsadesantiago.com/api/RV_Instrumentos/getUltimasTransacciones'; | |
var data = {'nemo': NEMO.toUpperCase()}; | |
return getBCS_(url, data, query, parseOptions); | |
} | |
function getBCS_UltimasVariacionesCapital(NEMO, query, parseOptions) { | |
var url = 'https://www.bolsadesantiago.com/api/RV_Instrumentos/getUltimasVariacionesCapital'; | |
var data = {'nemo': NEMO.toUpperCase(), 'tamanopag': 0, 'pagina': 0}; | |
return getBCS_(url, data, query, parseOptions); | |
} | |
function getBCS_Dividendos(NEMO, query, parseOptions) { | |
var url = 'https://www.bolsadesantiago.com/api/RV_ResumenMercado/getDividendos'; | |
let d = new Date; | |
//var data = {'nemo': NEMO.toUpperCase(), 'fec_pagoini': `2023-01-01`, 'fec_pagofin': `2023-12-31`}; | |
var data = {'nemo': NEMO.toUpperCase(), 'fec_pagoini': `${d.getFullYear()}-01-01`, 'fec_pagofin': `${d.getFullYear()}-12-31`}; | |
return getBCS_(url, data, query, parseOptions); | |
} | |
function getBCS_PuntasCompraVenta(NEMO, query, parseOptions) { | |
var url = 'https://www.bolsadesantiago.com/api/RV_Instrumentos/getPuntasCompraVenta'; | |
var data = {'nemo': NEMO.toUpperCase()}; | |
return getBCS_(url, data, query, parseOptions); | |
} | |
function getBCS_Rentabilidad(NEMO, query, parseOptions) { | |
var url = 'https://www.bolsadesantiago.com/api/RV_Instrumentos/getRentabilidad'; | |
var data = {'nemo': NEMO.toUpperCase(), 'idioma': 'es'}; | |
return getBCS_(url, data, query, parseOptions); | |
} | |
function getBCS_MetaIntraday(NEMO, query, parseOptions) { | |
var url = 'https://www.bolsadesantiago.com/api/RV_Instrumentos/getMetaIntraday'; | |
var data = {'symbol': NEMO.toUpperCase(), 'time': ''}; | |
return getBCS_(url, data, query, parseOptions); | |
} | |
function getBCS_OtraInformacion(NEMO, query, parseOptions) { | |
var url = 'https://www.bolsadesantiago.com/api/RV_Instrumentos/getOtraInformacion'; | |
var data = {'nemo': NEMO.toUpperCase(), 'tipo_informacion': 'Reseña Compañía'}; | |
return getBCS_(url, data, query, parseOptions); | |
} | |
function getBCS_ETFs(NEMO, query, parseOptions) { | |
var url = 'https://www.bolsadesantiago.com/api/RV_ResumenMercado/getETFs'; | |
var data = {'Etf': NEMO.toUpperCase(), 'Idioma': 'Español'}; | |
return getBCS_(url, data, query, parseOptions); | |
} | |
function getBCS_ETFCreacionDetalle(NEMO, query, parseOptions) { | |
var url = 'https://www.bolsadesantiago.com/api/RV_ResumenMercado/getETFCreacionDetalle'; | |
var data = {'vetf': NEMO.toUpperCase()}; | |
return getBCS_(url, data, query, parseOptions); | |
} | |
function getBCS_ETFCreacionTotales(NEMO, query, parseOptions) { | |
var url = 'https://www.bolsadesantiago.com/api/RV_ResumenMercado/getETFCreacionTotales'; | |
var data = {'vetf': NEMO.toUpperCase()}; | |
return getBCS_(url, data, query, parseOptions); | |
} | |
function getBCS_IndiceIntraday(NEMO, query, parseOptions) { | |
var url = 'https://www.bolsadesantiago.com/api/RV_Indices/getIndiceIntraday'; | |
var data = {'symbol': NEMO.toUpperCase(), 'condividendo': 'S'}; | |
return getBCS_(url, data, query, parseOptions); | |
} | |
function getBCS_EmpresasIndice(NEMO, query, parseOptions) { | |
var url = 'https://www.bolsadesantiago.com/api/RV_Indices/getEmpresasIndice'; | |
var data = {'indice': NEMO.toUpperCase(), 'dividendo': 'S'}; | |
return getBCS_(url, data, query, parseOptions); | |
} | |
function getBCS_Indice(NEMO, query, parseOptions) { | |
var url = 'https://www.bolsadesantiago.com/api/RV_Indices/getIndice'; | |
var data = {'indice': NEMO.toUpperCase(), 'dividendo': 'S'}; | |
return getBCS_(url, data, query, parseOptions); | |
} | |
function getBCS_DescIndices(NEMO, query, parseOptions) { | |
var url = 'https://www.bolsadesantiago.com/api/RV_Indices/getDescIndices'; | |
var data = {'Indice': NEMO.toUpperCase()}; | |
return getBCS_(url, data, query, parseOptions); | |
} | |
function getBCS_MetodCalculoIndice(NEMO, query, parseOptions) { | |
var url = 'https://www.bolsadesantiago.com/api/RV_Indices/getMetodCalculoIndice'; | |
var data = {'indice': NEMO.toUpperCase()}; | |
return getBCS_(url, data, query, parseOptions); | |
} | |
function getBCS_HoraMercado(query, parseOptions) { | |
var url = 'https://www.bolsadesantiago.com/api/Comunes/getHoraMercado'; | |
var data = {}; | |
return getBCS_(url, data, query, parseOptions); | |
} | |
function getBCS_Indices(query, parseOptions) { | |
var url = 'https://www.bolsadesantiago.com/api/RV_Indices/getIndices'; | |
var data = {'dividendo': 'S'}; | |
return getBCS_(url, data, query, parseOptions); | |
} | |
function getBCS_AccionesPrecios(query, parseOptions) { | |
var url = 'https://www.bolsadesantiago.com/api/RV_ResumenMercado/getAccionesPrecios'; | |
var data = {}; | |
return getBCS_(url, data, query, parseOptions); | |
} | |
function getBCS_(url, data, query, parseOptions) { | |
//(NEMO, query, parseOptions) | |
//(NEMO, query, parseOptions, includeFunc, transformFunc) | |
var headers = { | |
'content-type': 'application/json;charset=UTF-8', | |
'accept': 'application/json, text/plain, */*', | |
'accept-encoding': 'gzip, deflate, br', | |
'accept-language': 'es,es-419;q=0.9,en;q=0.8,ja;q=0.7,zh-CN;q=0.6,zh-TW;q=0.5,zh;q=0.4', | |
'x-csrf-token': 'OIfgunBB-S0uc7xpcHMqks3eZV0AUieKN7nE', | |
'cookie': 'BIGipCookie=000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000; f5avraaaaaaaaaaaaaaaa_session_=OHHPMPKJPCABPNOACCLDCBGHFKAGFBLEMIIMNHEMELBEDGAJHFIACDCICJAIHNLBNEADMKNLIPNPFKODFBPAPHKFIPJBDHJCCEDPMHMADNEGDFBFNEFPHHIKBICIGNPM; BIGipServerPool-Push_HTML5_corporativa=684715681.20480.0000; gb-wbchtbt-uid=1679871467360; _csrf=LifxCrFalOBI4Z-utnGR8ywV; __uzma=3bec6c3c-ce48-49fd-8941-063e080349b4; __uzmb=1679871475; __uzme=3037; access_token=; user_id=; BIGipCookie=000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000; BIGipServerPool-www.bolsadesantiago.com-HTML5_corporativa=735047329.20480.0000; f5avraaaaaaaaaaaaaaaa_session_=HHDINOBHBDLFKCEPIBBOHDGCNIJJNFPDMIBDPFIMJOJLBGHOICEGJDJHEEFCFDLDBFPDOJPPFGGDAOFMCFAABHMOGAKLGHDCBIMBNENGCGBDNJDEKFKEIELNJBAOELFC; __uzmc=2574495267417; __uzmd=1679951558', | |
'user-agent': 'Mozilla/5.0 (X11; CrOS i686 12239.2.0) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/76.0.3809.78 Safari/537.36', | |
'sec-fetch-mode': 'cors', | |
'sec-fetch-site': 'same-origin', | |
'origin':'https://www.bolsadesantiago.com', | |
'authority': 'www.bolsadesantiago.com', | |
'referer': 'https://www.bolsadesantiago.com' | |
}; | |
//'referer': `https://www.bolsadesantiago.com/resumen_instrumento/${NEMO.toUpperCase()}` | |
var fetchOptions = {'method': 'post','contentType': 'application/json;charset=UTF-8','payload': JSON.stringify(data), 'headers': headers, 'muteHttpExceptions': true}; | |
//try {var object = JSON.parse(jsondata.getContentText());} catch(e) {var object = JSON.parse(JSON.stringify({}));} | |
return ImportJSONAdvanced(url, fetchOptions, query, parseOptions, includeXPath_, defaultTransform_); | |
} | |
/*====================================================================================================================================* | |
ImportJSON by Brad Jasper and Trevor Lohrbeer | |
==================================================================================================================================== | |
Version: 1.5.0 | |
Project Page: https://github.com/bradjasper/ImportJSON | |
------------------------------------------------------------------------------------------------------------------------------------ | |
A library for importing JSON feeds into Google spreadsheets. | |
*====================================================================================================================================*/ | |
/** include ImportJSON.gs content here! **/ |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
/*====================================================================================================================================* | |
ImportJSON by Brad Jasper and Trevor Lohrbeer | |
==================================================================================================================================== | |
Version: 1.5.0 | |
Project Page: https://github.com/bradjasper/ImportJSON | |
Copyright: (c) 2017-2019 by Brad Jasper | |
(c) 2012-2017 by Trevor Lohrbeer | |
License: GNU General Public License, version 3 (GPL-3.0) | |
http://www.opensource.org/licenses/gpl-3.0.html | |
------------------------------------------------------------------------------------------------------------------------------------ | |
A library for importing JSON feeds into Google spreadsheets. Functions include: | |
ImportJSON For use by end users to import a JSON feed from a URL | |
ImportJSONFromSheet For use by end users to import JSON from one of the Sheets | |
ImportJSONViaPost For use by end users to import a JSON feed from a URL using POST parameters | |
ImportJSONAdvanced For use by script developers to easily extend the functionality of this library | |
ImportJSONBasicAuth For use by end users to import a JSON feed from a URL with HTTP Basic Auth (added by Karsten Lettow) | |
For future enhancements see https://github.com/bradjasper/ImportJSON/issues?q=is%3Aissue+is%3Aopen+label%3Aenhancement | |
For bug reports see https://github.com/bradjasper/ImportJSON/issues | |
------------------------------------------------------------------------------------------------------------------------------------ | |
Changelog: | |
1.6.0 (June 2, 2019) Fixed null values (thanks @gdesmedt1) | |
1.5.0 (January 11, 2019) Adds ability to include all headers in a fixed order even when no data is present for a given header in some or all rows. | |
1.4.0 (July 23, 2017) Transfer project to Brad Jasper. Fixed off-by-one array bug. Fixed previous value bug. Added custom annotations. Added ImportJSONFromSheet and ImportJSONBasicAuth. | |
1.3.0 Adds ability to import the text from a set of rows containing the text to parse. All cells are concatenated | |
1.2.1 Fixed a bug with how nested arrays are handled. The rowIndex counter wasn't incrementing properly when parsing. | |
1.2.0 Added ImportJSONViaPost and support for fetchOptions to ImportJSONAdvanced | |
1.1.1 Added a version number using Google Scripts Versioning so other developers can use the library | |
1.1.0 Added support for the noHeaders option | |
1.0.0 Initial release | |
*====================================================================================================================================*/ | |
/** | |
* Imports a JSON feed and returns the results to be inserted into a Google Spreadsheet. The JSON feed is flattened to create | |
* a two-dimensional array. The first row contains the headers, with each column header indicating the path to that data in | |
* the JSON feed. The remaining rows contain the data. | |
* | |
* By default, data gets transformed so it looks more like a normal data import. Specifically: | |
* | |
* - Data from parent JSON elements gets inherited to their child elements, so rows representing child elements contain the values | |
* of the rows representing their parent elements. | |
* - Values longer than 256 characters get truncated. | |
* - Headers have slashes converted to spaces, common prefixes removed and the resulting text converted to title case. | |
* | |
* To change this behavior, pass in one of these values in the options parameter: | |
* | |
* noInherit: Don't inherit values from parent elements | |
* noTruncate: Don't truncate values | |
* rawHeaders: Don't prettify headers | |
* noHeaders: Don't include headers, only the data | |
* allHeaders: Include all headers from the query parameter in the order they are listed | |
* debugLocation: Prepend each value with the row & column it belongs in | |
* | |
* For example: | |
* | |
* =ImportJSON("http://gdata.youtube.com/feeds/api/standardfeeds/most_popular?v=2&alt=json", "/feed/entry/title,/feed/entry/content", | |
* "noInherit,noTruncate,rawHeaders") | |
* | |
* @param {url} the URL to a public JSON feed | |
* @param {query} a comma-separated list of paths to import. Any path starting with one of these paths gets imported. | |
* @param {parseOptions} a comma-separated list of options that alter processing of the data | |
* @customfunction | |
* | |
* @return a two-dimensional array containing the data, with the first row containing headers | |
**/ | |
function ImportJSON(url, query, parseOptions) { | |
return ImportJSONAdvanced(url, null, query, parseOptions, includeXPath_, defaultTransform_); | |
} | |
/** | |
* Imports a JSON feed via a POST request and returns the results to be inserted into a Google Spreadsheet. The JSON feed is | |
* flattened to create a two-dimensional array. The first row contains the headers, with each column header indicating the path to | |
* that data in the JSON feed. The remaining rows contain the data. | |
* | |
* To retrieve the JSON, a POST request is sent to the URL and the payload is passed as the content of the request using the content | |
* type "application/x-www-form-urlencoded". If the fetchOptions define a value for "method", "payload" or "contentType", these | |
* values will take precedent. For example, advanced users can use this to make this function pass XML as the payload using a GET | |
* request and a content type of "application/xml; charset=utf-8". For more information on the available fetch options, see | |
* https://developers.google.com/apps-script/reference/url-fetch/url-fetch-app . At this time the "headers" option is not supported. | |
* | |
* By default, the returned data gets transformed so it looks more like a normal data import. Specifically: | |
* | |
* - Data from parent JSON elements gets inherited to their child elements, so rows representing child elements contain the values | |
* of the rows representing their parent elements. | |
* - Values longer than 256 characters get truncated. | |
* - Headers have slashes converted to spaces, common prefixes removed and the resulting text converted to title case. | |
* | |
* To change this behavior, pass in one of these values in the options parameter: | |
* | |
* noInherit: Don't inherit values from parent elements | |
* noTruncate: Don't truncate values | |
* rawHeaders: Don't prettify headers | |
* noHeaders: Don't include headers, only the data | |
* allHeaders: Include all headers from the query parameter in the order they are listed | |
* debugLocation: Prepend each value with the row & column it belongs in | |
* | |
* For example: | |
* | |
* =ImportJSON("http://gdata.youtube.com/feeds/api/standardfeeds/most_popular?v=2&alt=json", "user=bob&apikey=xxxx", | |
* "validateHttpsCertificates=false", "/feed/entry/title,/feed/entry/content", "noInherit,noTruncate,rawHeaders") | |
* | |
* @param {url} the URL to a public JSON feed | |
* @param {payload} the content to pass with the POST request; usually a URL encoded list of parameters separated by ampersands | |
* @param {fetchOptions} a comma-separated list of options used to retrieve the JSON feed from the URL | |
* @param {query} a comma-separated list of paths to import. Any path starting with one of these paths gets imported. | |
* @param {parseOptions} a comma-separated list of options that alter processing of the data | |
* @customfunction | |
* | |
* @return a two-dimensional array containing the data, with the first row containing headers | |
**/ | |
function ImportJSONViaPost(url, payload, fetchOptions, query, parseOptions) { | |
var postOptions = parseToObject_(fetchOptions); | |
if (postOptions["method"] == null) { | |
postOptions["method"] = "POST"; | |
} | |
if (postOptions["payload"] == null) { | |
postOptions["payload"] = payload; | |
} | |
if (postOptions["contentType"] == null) { | |
postOptions["contentType"] = "application/x-www-form-urlencoded"; | |
} | |
convertToBool_(postOptions, "validateHttpsCertificates"); | |
convertToBool_(postOptions, "useIntranet"); | |
convertToBool_(postOptions, "followRedirects"); | |
convertToBool_(postOptions, "muteHttpExceptions"); | |
return ImportJSONAdvanced(url, postOptions, query, parseOptions, includeXPath_, defaultTransform_); | |
} | |
/** | |
* Imports a JSON text from a named Sheet and returns the results to be inserted into a Google Spreadsheet. The JSON feed is flattened to create | |
* a two-dimensional array. The first row contains the headers, with each column header indicating the path to that data in | |
* the JSON feed. The remaining rows contain the data. | |
* | |
* By default, data gets transformed so it looks more like a normal data import. Specifically: | |
* | |
* - Data from parent JSON elements gets inherited to their child elements, so rows representing child elements contain the values | |
* of the rows representing their parent elements. | |
* - Values longer than 256 characters get truncated. | |
* - Headers have slashes converted to spaces, common prefixes removed and the resulting text converted to title case. | |
* | |
* To change this behavior, pass in one of these values in the options parameter: | |
* | |
* noInherit: Don't inherit values from parent elements | |
* noTruncate: Don't truncate values | |
* rawHeaders: Don't prettify headers | |
* noHeaders: Don't include headers, only the data | |
* allHeaders: Include all headers from the query parameter in the order they are listed | |
* debugLocation: Prepend each value with the row & column it belongs in | |
* | |
* For example: | |
* | |
* =ImportJSONFromSheet("Source", "/feed/entry/title,/feed/entry/content", | |
* "noInherit,noTruncate,rawHeaders") | |
* | |
* @param {sheetName} the name of the sheet containg the text for the JSON | |
* @param {query} a comma-separated lists of paths to import. Any path starting with one of these paths gets imported. | |
* @param {options} a comma-separated list of options that alter processing of the data | |
* | |
* @return a two-dimensional array containing the data, with the first row containing headers | |
* @customfunction | |
**/ | |
function ImportJSONFromSheet(sheetName, query, options) { | |
var object = getDataFromNamedSheet_(sheetName); | |
return parseJSONObject_(object, query, options, includeXPath_, defaultTransform_); | |
} | |
/** | |
* An advanced version of ImportJSON designed to be easily extended by a script. This version cannot be called from within a | |
* spreadsheet. | |
* | |
* Imports a JSON feed and returns the results to be inserted into a Google Spreadsheet. The JSON feed is flattened to create | |
* a two-dimensional array. The first row contains the headers, with each column header indicating the path to that data in | |
* the JSON feed. The remaining rows contain the data. | |
* | |
* The fetchOptions can be used to change how the JSON feed is retrieved. For instance, the "method" and "payload" options can be | |
* set to pass a POST request with post parameters. For more information on the available parameters, see | |
* https://developers.google.com/apps-script/reference/url-fetch/url-fetch-app . | |
* | |
* Use the include and transformation functions to determine what to include in the import and how to transform the data after it is | |
* imported. | |
* | |
* For example: | |
* | |
* ImportJSON("http://gdata.youtube.com/feeds/api/standardfeeds/most_popular?v=2&alt=json", | |
* new Object() { "method" : "post", "payload" : "user=bob&apikey=xxxx" }, | |
* "/feed/entry", | |
* "", | |
* function (query, path) { return path.indexOf(query) == 0; }, | |
* function (data, row, column) { data[row][column] = data[row][column].toString().substr(0, 100); } ) | |
* | |
* In this example, the import function checks to see if the path to the data being imported starts with the query. The transform | |
* function takes the data and truncates it. For more robust versions of these functions, see the internal code of this library. | |
* | |
* @param {url} the URL to a public JSON feed | |
* @param {fetchOptions} an object whose properties are options used to retrieve the JSON feed from the URL | |
* @param {query} the query passed to the include function | |
* @param {parseOptions} a comma-separated list of options that may alter processing of the data | |
* @param {includeFunc} a function with the signature func(query, path, options) that returns true if the data element at the given path | |
* should be included or false otherwise. | |
* @param {transformFunc} a function with the signature func(data, row, column, options) where data is a 2-dimensional array of the data | |
* and row & column are the current row and column being processed. Any return value is ignored. Note that row 0 | |
* contains the headers for the data, so test for row==0 to process headers only. | |
* | |
* @return a two-dimensional array containing the data, with the first row containing headers | |
* @customfunction | |
**/ | |
function ImportJSONAdvanced(url, fetchOptions, query, parseOptions, includeFunc, transformFunc) { | |
var jsondata = UrlFetchApp.fetch(url, fetchOptions); | |
var object = JSON.parse(jsondata.getContentText()); | |
return parseJSONObject_(object, query, parseOptions, includeFunc, transformFunc); | |
} | |
/** | |
* Helper function to authenticate with basic auth informations using ImportJSONAdvanced | |
* | |
* Imports a JSON feed and returns the results to be inserted into a Google Spreadsheet. The JSON feed is flattened to create | |
* a two-dimensional array. The first row contains the headers, with each column header indicating the path to that data in | |
* the JSON feed. The remaining rows contain the data. | |
* | |
* The fetchOptions can be used to change how the JSON feed is retrieved. For instance, the "method" and "payload" options can be | |
* set to pass a POST request with post parameters. For more information on the available parameters, see | |
* https://developers.google.com/apps-script/reference/url-fetch/url-fetch-app . | |
* | |
* Use the include and transformation functions to determine what to include in the import and how to transform the data after it is | |
* imported. | |
* | |
* @param {url} the URL to a http basic auth protected JSON feed | |
* @param {username} the Username for authentication | |
* @param {password} the Password for authentication | |
* @param {query} the query passed to the include function (optional) | |
* @param {parseOptions} a comma-separated list of options that may alter processing of the data (optional) | |
* | |
* @return a two-dimensional array containing the data, with the first row containing headers | |
* @customfunction | |
**/ | |
function ImportJSONBasicAuth(url, username, password, query, parseOptions) { | |
var encodedAuthInformation = Utilities.base64Encode(username + ":" + password); | |
var header = {headers: {Authorization: "Basic " + encodedAuthInformation}}; | |
return ImportJSONAdvanced(url, header, query, parseOptions, includeXPath_, defaultTransform_); | |
} | |
/** | |
* Encodes the given value to use within a URL. | |
* | |
* @param {value} the value to be encoded | |
* | |
* @return the value encoded using URL percent-encoding | |
*/ | |
function URLEncode(value) { | |
return encodeURIComponent(value.toString()); | |
} | |
/** | |
* Adds an oAuth service using the given name and the list of properties. | |
* | |
* @note This method is an experiment in trying to figure out how to add an oAuth service without having to specify it on each | |
* ImportJSON call. The idea was to call this method in the first cell of a spreadsheet, and then use ImportJSON in other | |
* cells. This didn't work, but leaving this in here for further experimentation later. | |
* | |
* The test I did was to add the following into the A1: | |
* | |
* =AddOAuthService("twitter", "https://api.twitter.com/oauth/access_token", | |
* "https://api.twitter.com/oauth/request_token", "https://api.twitter.com/oauth/authorize", | |
* "<my consumer key>", "<my consumer secret>", "", "") | |
* | |
* Information on obtaining a consumer key & secret for Twitter can be found at https://dev.twitter.com/docs/auth/using-oauth | |
* | |
* Then I added the following into A2: | |
* | |
* =ImportJSONViaPost("https://api.twitter.com/1.1/statuses/user_timeline.json?screen_name=fastfedora&count=2", "", | |
* "oAuthServiceName=twitter,oAuthUseToken=always", "/", "") | |
* | |
* I received an error that the "oAuthServiceName" was not a valid value. [twl 18.Apr.13] | |
*/ | |
function AddOAuthService__(name, accessTokenUrl, requestTokenUrl, authorizationUrl, consumerKey, consumerSecret, method, paramLocation) { | |
var oAuthConfig = UrlFetchApp.addOAuthService(name); | |
if (accessTokenUrl != null && accessTokenUrl.length > 0) { | |
oAuthConfig.setAccessTokenUrl(accessTokenUrl); | |
} | |
if (requestTokenUrl != null && requestTokenUrl.length > 0) { | |
oAuthConfig.setRequestTokenUrl(requestTokenUrl); | |
} | |
if (authorizationUrl != null && authorizationUrl.length > 0) { | |
oAuthConfig.setAuthorizationUrl(authorizationUrl); | |
} | |
if (consumerKey != null && consumerKey.length > 0) { | |
oAuthConfig.setConsumerKey(consumerKey); | |
} | |
if (consumerSecret != null && consumerSecret.length > 0) { | |
oAuthConfig.setConsumerSecret(consumerSecret); | |
} | |
if (method != null && method.length > 0) { | |
oAuthConfig.setMethod(method); | |
} | |
if (paramLocation != null && paramLocation.length > 0) { | |
oAuthConfig.setParamLocation(paramLocation); | |
} | |
} | |
/** | |
* Parses a JSON object and returns a two-dimensional array containing the data of that object. | |
*/ | |
function parseJSONObject_(object, query, options, includeFunc, transformFunc) { | |
var headers = new Array(); | |
var data = new Array(); | |
if (query && !Array.isArray(query) && query.toString().indexOf(",") != -1) { | |
query = query.toString().split(","); | |
} | |
// Prepopulate the headers to lock in their order | |
if (hasOption_(options, "allHeaders") && Array.isArray(query)) | |
{ | |
for (var i = 0; i < query.length; i++) | |
{ | |
headers[query[i]] = Object.keys(headers).length; | |
} | |
} | |
if (options) { | |
options = options.toString().split(","); | |
} | |
parseData_(headers, data, "", {rowIndex: 1}, object, query, options, includeFunc); | |
parseHeaders_(headers, data); | |
transformData_(data, options, transformFunc); | |
return hasOption_(options, "noHeaders") ? (data.length > 1 ? data.slice(1) : new Array()) : data; | |
} | |
/** | |
* Parses the data contained within the given value and inserts it into the data two-dimensional array starting at the rowIndex. | |
* If the data is to be inserted into a new column, a new header is added to the headers array. The value can be an object, | |
* array or scalar value. | |
* | |
* If the value is an object, it's properties are iterated through and passed back into this function with the name of each | |
* property extending the path. For instance, if the object contains the property "entry" and the path passed in was "/feed", | |
* this function is called with the value of the entry property and the path "/feed/entry". | |
* | |
* If the value is an array containing other arrays or objects, each element in the array is passed into this function with | |
* the rowIndex incremeneted for each element. | |
* | |
* If the value is an array containing only scalar values, those values are joined together and inserted into the data array as | |
* a single value. | |
* | |
* If the value is a scalar, the value is inserted directly into the data array. | |
*/ | |
function parseData_(headers, data, path, state, value, query, options, includeFunc) { | |
var dataInserted = false; | |
if (Array.isArray(value) && isObjectArray_(value)) { | |
for (var i = 0; i < value.length; i++) { | |
if (parseData_(headers, data, path, state, value[i], query, options, includeFunc)) { | |
dataInserted = true; | |
if (data[state.rowIndex]) { | |
state.rowIndex++; | |
} | |
} | |
} | |
} else if (isObject_(value)) { | |
for (key in value) { | |
if (parseData_(headers, data, path + "/" + key, state, value[key], query, options, includeFunc)) { | |
dataInserted = true; | |
} | |
} | |
} else if (!includeFunc || includeFunc(query, path, options)) { | |
// Handle arrays containing only scalar values | |
if (Array.isArray(value)) { | |
value = value.join(); | |
} | |
// Insert new row if one doesn't already exist | |
if (!data[state.rowIndex]) { | |
data[state.rowIndex] = new Array(); | |
} | |
// Add a new header if one doesn't exist | |
if (!headers[path] && headers[path] != 0) { | |
headers[path] = Object.keys(headers).length; | |
} | |
// Insert the data | |
data[state.rowIndex][headers[path]] = value; | |
dataInserted = true; | |
} | |
return dataInserted; | |
} | |
/** | |
* Parses the headers array and inserts it into the first row of the data array. | |
*/ | |
function parseHeaders_(headers, data) { | |
data[0] = new Array(); | |
for (key in headers) { | |
data[0][headers[key]] = key; | |
} | |
} | |
/** | |
* Applies the transform function for each element in the data array, going through each column of each row. | |
*/ | |
function transformData_(data, options, transformFunc) { | |
for (var i = 0; i < data.length; i++) { | |
for (var j = 0; j < data[0].length; j++) { | |
transformFunc(data, i, j, options); | |
} | |
} | |
} | |
/** | |
* Returns true if the given test value is an object; false otherwise. | |
*/ | |
function isObject_(test) { | |
return Object.prototype.toString.call(test) === '[object Object]'; | |
} | |
/** | |
* Returns true if the given test value is an array containing at least one object; false otherwise. | |
*/ | |
function isObjectArray_(test) { | |
for (var i = 0; i < test.length; i++) { | |
if (isObject_(test[i])) { | |
return true; | |
} | |
} | |
return false; | |
} | |
/** | |
* Returns true if the given query applies to the given path. | |
*/ | |
function includeXPath_(query, path, options) { | |
if (!query) { | |
return true; | |
} else if (Array.isArray(query)) { | |
for (var i = 0; i < query.length; i++) { | |
if (applyXPathRule_(query[i], path, options)) { | |
return true; | |
} | |
} | |
} else { | |
return applyXPathRule_(query, path, options); | |
} | |
return false; | |
}; | |
/** | |
* Returns true if the rule applies to the given path. | |
*/ | |
function applyXPathRule_(rule, path, options) { | |
return path.indexOf(rule) == 0; | |
} | |
/** | |
* By default, this function transforms the value at the given row & column so it looks more like a normal data import. Specifically: | |
* | |
* - Data from parent JSON elements gets inherited to their child elements, so rows representing child elements contain the values | |
* of the rows representing their parent elements. | |
* - Values longer than 256 characters get truncated. | |
* - Values in row 0 (headers) have slashes converted to spaces, common prefixes removed and the resulting text converted to title | |
* case. | |
* | |
* To change this behavior, pass in one of these values in the options parameter: | |
* | |
* noInherit: Don't inherit values from parent elements | |
* noTruncate: Don't truncate values | |
* rawHeaders: Don't prettify headers | |
* debugLocation: Prepend each value with the row & column it belongs in | |
*/ | |
function defaultTransform_(data, row, column, options) { | |
if (data[row][column] == null) { | |
if (row < 2 || hasOption_(options, "noInherit")) { | |
data[row][column] = ""; | |
} else { | |
data[row][column] = data[row-1][column]; | |
} | |
} | |
if (!hasOption_(options, "rawHeaders") && row == 0) { | |
if (column == 0 && data[row].length > 1) { | |
removeCommonPrefixes_(data, row); | |
} | |
data[row][column] = toTitleCase_(data[row][column].toString().replace(/[\/\_]/g, " ")); | |
} | |
if (!hasOption_(options, "noTruncate") && data[row][column]) { | |
data[row][column] = data[row][column].toString().substr(0, 256); | |
} | |
if (hasOption_(options, "debugLocation")) { | |
data[row][column] = "[" + row + "," + column + "]" + data[row][column]; | |
} | |
} | |
/** | |
* If all the values in the given row share the same prefix, remove that prefix. | |
*/ | |
function removeCommonPrefixes_(data, row) { | |
var matchIndex = data[row][0].length; | |
for (var i = 1; i < data[row].length; i++) { | |
matchIndex = findEqualityEndpoint_(data[row][i-1], data[row][i], matchIndex); | |
if (matchIndex == 0) { | |
return; | |
} | |
} | |
for (var i = 0; i < data[row].length; i++) { | |
data[row][i] = data[row][i].substring(matchIndex, data[row][i].length); | |
} | |
} | |
/** | |
* Locates the index where the two strings values stop being equal, stopping automatically at the stopAt index. | |
*/ | |
function findEqualityEndpoint_(string1, string2, stopAt) { | |
if (!string1 || !string2) { | |
return -1; | |
} | |
var maxEndpoint = Math.min(stopAt, string1.length, string2.length); | |
for (var i = 0; i < maxEndpoint; i++) { | |
if (string1.charAt(i) != string2.charAt(i)) { | |
return i; | |
} | |
} | |
return maxEndpoint; | |
} | |
/** | |
* Converts the text to title case. | |
*/ | |
function toTitleCase_(text) { | |
if (text == null) { | |
return null; | |
} | |
return text.replace(/\w\S*/g, function(word) { return word.charAt(0).toUpperCase() + word.substr(1).toLowerCase(); }); | |
} | |
/** | |
* Returns true if the given set of options contains the given option. | |
*/ | |
function hasOption_(options, option) { | |
return options && options.indexOf(option) >= 0; | |
} | |
/** | |
* Parses the given string into an object, trimming any leading or trailing spaces from the keys. | |
*/ | |
function parseToObject_(text) { | |
var map = new Object(); | |
var entries = (text != null && text.trim().length > 0) ? text.toString().split(",") : new Array(); | |
for (var i = 0; i < entries.length; i++) { | |
addToMap_(map, entries[i]); | |
} | |
return map; | |
} | |
/** | |
* Parses the given entry and adds it to the given map, trimming any leading or trailing spaces from the key. | |
*/ | |
function addToMap_(map, entry) { | |
var equalsIndex = entry.indexOf("="); | |
var key = (equalsIndex != -1) ? entry.substring(0, equalsIndex) : entry; | |
var value = (key.length + 1 < entry.length) ? entry.substring(key.length + 1) : ""; | |
map[key.trim()] = value; | |
} | |
/** | |
* Returns the given value as a boolean. | |
*/ | |
function toBool_(value) { | |
return value == null ? false : (value.toString().toLowerCase() == "true" ? true : false); | |
} | |
/** | |
* Converts the value for the given key in the given map to a bool. | |
*/ | |
function convertToBool_(map, key) { | |
if (map[key] != null) { | |
map[key] = toBool_(map[key]); | |
} | |
} | |
function getDataFromNamedSheet_(sheetName) { | |
var ss = SpreadsheetApp.getActiveSpreadsheet(); | |
var source = ss.getSheetByName(sheetName); | |
var jsonRange = source.getRange(1,1,source.getLastRow()); | |
var jsonValues = jsonRange.getValues(); | |
var jsonText = ""; | |
for (var row in jsonValues) { | |
for (var col in jsonValues[row]) { | |
jsonText +=jsonValues[row][col]; | |
} | |
} | |
Logger.log(jsonText); | |
return JSON.parse(jsonText); | |
} |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment