728 lines
23 KiB
JavaScript
728 lines
23 KiB
JavaScript
//>>built
|
|
define("dojox/data/CsvStore", ["dojo/_base/lang", "dojo/_base/declare", "dojo/_base/xhr", "dojo/_base/window","dojo/data/util/filter", "dojo/data/util/simpleFetch"],
|
|
function(lang, declare, xhr, winUtil, filterUtil, simpleFetch) {
|
|
|
|
var CsvStore = declare("dojox.data.CsvStore", null, {
|
|
// summary:
|
|
// The CsvStore implements the dojo.data.api.Read API and reads
|
|
// data from files in CSV (Comma Separated Values) format.
|
|
// All values are simple string values. References to other items
|
|
// are not supported as attribute values in this datastore.
|
|
//
|
|
// Example data file:
|
|
// name, color, age, tagline
|
|
// Kermit, green, 12, "Hi, I'm Kermit the Frog."
|
|
// Fozzie Bear, orange, 10, "Wakka Wakka Wakka!"
|
|
// Miss Piggy, pink, 11, "Kermie!"
|
|
//
|
|
// Note that values containing a comma must be enclosed with quotes ("")
|
|
// Also note that values containing quotes must be escaped with two consecutive quotes (""quoted"")
|
|
//
|
|
// examples:
|
|
// var csvStore = new dojox.data.CsvStore({url:"movies.csv");
|
|
// var csvStore = new dojox.data.CsvStore({url:"http://example.com/movies.csv");
|
|
|
|
constructor: function(/* Object */ keywordParameters){
|
|
// summary:
|
|
// initializer
|
|
// keywordParameters: {url: String}
|
|
// keywordParameters: {data: String}
|
|
// keywordParameters: {label: String} The column label for the column to use for the label returned by getLabel.
|
|
// keywordParameters: {identifier: String} The column label for the column to use for the identity. Optional. If not set, the identity is the row number.
|
|
|
|
this._attributes = []; // e.g. ["Title", "Year", "Producer"]
|
|
this._attributeIndexes = {}; // e.g. {Title: 0, Year: 1, Producer: 2}
|
|
this._dataArray = []; // e.g. [[<Item0>],[<Item1>],[<Item2>]]
|
|
this._arrayOfAllItems = []; // e.g. [{_csvId:0,_csvStore:store},...]
|
|
this._loadFinished = false;
|
|
if(keywordParameters.url){
|
|
this.url = keywordParameters.url;
|
|
}
|
|
this._csvData = keywordParameters.data;
|
|
if(keywordParameters.label){
|
|
this.label = keywordParameters.label;
|
|
}else if(this.label === ""){
|
|
this.label = undefined;
|
|
}
|
|
this._storeProp = "_csvStore"; // Property name for the store reference on every item.
|
|
this._idProp = "_csvId"; // Property name for the Item Id on every item.
|
|
this._features = {
|
|
'dojo.data.api.Read': true,
|
|
'dojo.data.api.Identity': true
|
|
};
|
|
this._loadInProgress = false; //Got to track the initial load to prevent duelling loads of the dataset.
|
|
this._queuedFetches = [];
|
|
this.identifier = keywordParameters.identifier;
|
|
if(this.identifier === ""){
|
|
delete this.identifier;
|
|
}else{
|
|
this._idMap = {};
|
|
}
|
|
if("separator" in keywordParameters){
|
|
this.separator = keywordParameters.separator;
|
|
}
|
|
if("urlPreventCache" in keywordParameters){
|
|
this.urlPreventCache = keywordParameters.urlPreventCache?true:false;
|
|
}
|
|
},
|
|
|
|
// url: [public] string
|
|
// Declarative hook for setting Csv source url.
|
|
url: "",
|
|
|
|
// label: [public] string
|
|
// Declarative hook for setting the label attribute.
|
|
label: "",
|
|
|
|
// identifier: [public] string
|
|
// Declarative hook for setting the identifier.
|
|
identifier: "",
|
|
|
|
// separator: [public] string
|
|
// Declatative and programmatic hook for defining the separator
|
|
// character used in the Csv style file.
|
|
separator: ",",
|
|
|
|
// separator: [public] string
|
|
// Parameter to allow specifying if preventCache should be passed to
|
|
// the xhrGet call or not when loading data from a url.
|
|
// Note this does not mean the store calls the server on each fetch,
|
|
// only that the data load has preventCache set as an option.
|
|
urlPreventCache: false,
|
|
|
|
_assertIsItem: function(/* item */ item){
|
|
// summary:
|
|
// This function tests whether the item passed in is indeed an item in the store.
|
|
// item:
|
|
// The item to test for being contained by the store.
|
|
if(!this.isItem(item)){
|
|
throw new Error(this.declaredClass + ": a function was passed an item argument that was not an item");
|
|
}
|
|
},
|
|
|
|
_getIndex: function(item){
|
|
// summary:
|
|
// Internal function to get the internal index to the item data from the item handle
|
|
// item:
|
|
// The idem handle to get the index for.
|
|
var idx = this.getIdentity(item);
|
|
if(this.identifier){
|
|
idx = this._idMap[idx];
|
|
}
|
|
return idx;
|
|
},
|
|
|
|
/***************************************
|
|
dojo.data.api.Read API
|
|
***************************************/
|
|
getValue: function( /* item */ item,
|
|
/* attribute || attribute-name-string */ attribute,
|
|
/* value? */ defaultValue){
|
|
// summary:
|
|
// See dojo.data.api.Read.getValue()
|
|
// Note that for the CsvStore, an empty string value is the same as no value,
|
|
// so the defaultValue would be returned instead of an empty string.
|
|
this._assertIsItem(item);
|
|
var itemValue = defaultValue;
|
|
if(typeof attribute === "string"){
|
|
var ai = this._attributeIndexes[attribute];
|
|
if(ai != null){
|
|
var itemData = this._dataArray[this._getIndex(item)];
|
|
itemValue = itemData[ai] || defaultValue;
|
|
}
|
|
}else{
|
|
throw new Error(this.declaredClass + ": a function was passed an attribute argument that was not a string");
|
|
}
|
|
return itemValue; //String
|
|
},
|
|
|
|
getValues: function(/* item */ item,
|
|
/* attribute || attribute-name-string */ attribute){
|
|
// summary:
|
|
// See dojo.data.api.Read.getValues()
|
|
// CSV syntax does not support multi-valued attributes, so this is just a
|
|
// wrapper function for getValue().
|
|
var value = this.getValue(item, attribute);
|
|
return (value ? [value] : []); //Array
|
|
},
|
|
|
|
getAttributes: function(/* item */ item){
|
|
// summary:
|
|
// See dojo.data.api.Read.getAttributes()
|
|
this._assertIsItem(item);
|
|
var attributes = [];
|
|
var itemData = this._dataArray[this._getIndex(item)];
|
|
for(var i=0; i<itemData.length; i++){
|
|
// Check for empty string values. CsvStore treats empty strings as no value.
|
|
if(itemData[i] !== ""){
|
|
attributes.push(this._attributes[i]);
|
|
}
|
|
}
|
|
return attributes; //Array
|
|
},
|
|
|
|
hasAttribute: function( /* item */ item,
|
|
/* attribute-name-string */ attribute){
|
|
// summary:
|
|
// See dojo.data.api.Read.hasAttribute()
|
|
// The hasAttribute test is true if attribute has an index number within the item's array length
|
|
// AND if the item has a value for that attribute. Note that for the CsvStore, an
|
|
// empty string value is the same as no value.
|
|
this._assertIsItem(item);
|
|
if(typeof attribute === "string"){
|
|
var attributeIndex = this._attributeIndexes[attribute];
|
|
var itemData = this._dataArray[this._getIndex(item)];
|
|
return (typeof attributeIndex !== "undefined" && attributeIndex < itemData.length && itemData[attributeIndex] !== ""); //Boolean
|
|
}else{
|
|
throw new Error(this.declaredClass + ": a function was passed an attribute argument that was not a string");
|
|
}
|
|
},
|
|
|
|
containsValue: function(/* item */ item,
|
|
/* attribute || attribute-name-string */ attribute,
|
|
/* anything */ value){
|
|
// summary:
|
|
// See dojo.data.api.Read.containsValue()
|
|
var regexp = undefined;
|
|
if(typeof value === "string"){
|
|
regexp = filterUtil.patternToRegExp(value, false);
|
|
}
|
|
return this._containsValue(item, attribute, value, regexp); //boolean.
|
|
},
|
|
|
|
_containsValue: function( /* item */ item,
|
|
/* attribute || attribute-name-string */ attribute,
|
|
/* anything */ value,
|
|
/* RegExp?*/ regexp){
|
|
// summary:
|
|
// Internal function for looking at the values contained by the item.
|
|
// description:
|
|
// Internal function for looking at the values contained by the item. This
|
|
// function allows for denoting if the comparison should be case sensitive for
|
|
// strings or not (for handling filtering cases where string case should not matter)
|
|
//
|
|
// item:
|
|
// The data item to examine for attribute values.
|
|
// attribute:
|
|
// The attribute to inspect.
|
|
// value:
|
|
// The value to match.
|
|
// regexp:
|
|
// Optional regular expression generated off value if value was of string type to handle wildcarding.
|
|
// If present and attribute values are string, then it can be used for comparison instead of 'value'
|
|
// tags:
|
|
// private
|
|
var values = this.getValues(item, attribute);
|
|
for(var i = 0; i < values.length; ++i){
|
|
var possibleValue = values[i];
|
|
if(typeof possibleValue === "string" && regexp){
|
|
return (possibleValue.match(regexp) !== null);
|
|
}else{
|
|
//Non-string matching.
|
|
if(value === possibleValue){
|
|
return true; // Boolean
|
|
}
|
|
}
|
|
}
|
|
return false; // Boolean
|
|
},
|
|
|
|
isItem: function(/* anything */ something){
|
|
// summary:
|
|
// See dojo.data.api.Read.isItem()
|
|
if(something && something[this._storeProp] === this){
|
|
var identity = something[this._idProp];
|
|
//If an identifier was specified, we have to look it up via that and the mapping,
|
|
//otherwise, just use row number.
|
|
if(this.identifier){
|
|
var data = this._dataArray[this._idMap[identity]];
|
|
if(data){
|
|
return true;
|
|
}
|
|
}else{
|
|
if(identity >= 0 && identity < this._dataArray.length){
|
|
return true; //Boolean
|
|
}
|
|
}
|
|
}
|
|
return false; //Boolean
|
|
},
|
|
|
|
isItemLoaded: function(/* anything */ something){
|
|
// summary:
|
|
// See dojo.data.api.Read.isItemLoaded()
|
|
// The CsvStore always loads all items, so if it's an item, then it's loaded.
|
|
return this.isItem(something); //Boolean
|
|
},
|
|
|
|
loadItem: function(/* item */ item){
|
|
// summary:
|
|
// See dojo.data.api.Read.loadItem()
|
|
// description:
|
|
// The CsvStore always loads all items, so if it's an item, then it's loaded.
|
|
// From the dojo.data.api.Read.loadItem docs:
|
|
// If a call to isItemLoaded() returns true before loadItem() is even called,
|
|
// then loadItem() need not do any work at all and will not even invoke
|
|
// the callback handlers.
|
|
},
|
|
|
|
getFeatures: function(){
|
|
// summary:
|
|
// See dojo.data.api.Read.getFeatures()
|
|
return this._features; //Object
|
|
},
|
|
|
|
getLabel: function(/* item */ item){
|
|
// summary:
|
|
// See dojo.data.api.Read.getLabel()
|
|
if(this.label && this.isItem(item)){
|
|
return this.getValue(item,this.label); //String
|
|
}
|
|
return undefined; //undefined
|
|
},
|
|
|
|
getLabelAttributes: function(/* item */ item){
|
|
// summary:
|
|
// See dojo.data.api.Read.getLabelAttributes()
|
|
if(this.label){
|
|
return [this.label]; //array
|
|
}
|
|
return null; //null
|
|
},
|
|
|
|
|
|
// The dojo.data.api.Read.fetch() function is implemented as
|
|
// a mixin from dojo.data.util.simpleFetch.
|
|
// That mixin requires us to define _fetchItems().
|
|
_fetchItems: function( /* Object */ keywordArgs,
|
|
/* Function */ findCallback,
|
|
/* Function */ errorCallback){
|
|
// summary:
|
|
// See dojo.data.util.simpleFetch.fetch()
|
|
// tags:
|
|
// protected
|
|
var self = this;
|
|
var filter = function(requestArgs, arrayOfAllItems){
|
|
var items = null;
|
|
if(requestArgs.query){
|
|
var key, value;
|
|
items = [];
|
|
var ignoreCase = requestArgs.queryOptions ? requestArgs.queryOptions.ignoreCase : false;
|
|
|
|
//See if there are any string values that can be regexp parsed first to avoid multiple regexp gens on the
|
|
//same value for each item examined. Much more efficient.
|
|
var regexpList = {};
|
|
for(key in requestArgs.query){
|
|
value = requestArgs.query[key];
|
|
if(typeof value === "string"){
|
|
regexpList[key] = filterUtil.patternToRegExp(value, ignoreCase);
|
|
}
|
|
}
|
|
|
|
for(var i = 0; i < arrayOfAllItems.length; ++i){
|
|
var match = true;
|
|
var candidateItem = arrayOfAllItems[i];
|
|
for(key in requestArgs.query){
|
|
value = requestArgs.query[key];
|
|
if(!self._containsValue(candidateItem, key, value, regexpList[key])){
|
|
match = false;
|
|
}
|
|
}
|
|
if(match){
|
|
items.push(candidateItem);
|
|
}
|
|
}
|
|
}else{
|
|
// We want a copy to pass back in case the parent wishes to sort the array. We shouldn't allow resort
|
|
// of the internal list so that multiple callers can get lists and sort without affecting each other.
|
|
items = arrayOfAllItems.slice(0,arrayOfAllItems.length);
|
|
|
|
}
|
|
findCallback(items, requestArgs);
|
|
};
|
|
|
|
if(this._loadFinished){
|
|
filter(keywordArgs, this._arrayOfAllItems);
|
|
}else{
|
|
if(this.url !== ""){
|
|
//If fetches come in before the loading has finished, but while
|
|
//a load is in progress, we have to defer the fetching to be
|
|
//invoked in the callback.
|
|
if(this._loadInProgress){
|
|
this._queuedFetches.push({args: keywordArgs, filter: filter});
|
|
}else{
|
|
this._loadInProgress = true;
|
|
var getArgs = {
|
|
url: self.url,
|
|
handleAs: "text",
|
|
preventCache: self.urlPreventCache
|
|
};
|
|
var getHandler = xhr.get(getArgs);
|
|
getHandler.addCallback(function(data){
|
|
try{
|
|
self._processData(data);
|
|
filter(keywordArgs, self._arrayOfAllItems);
|
|
self._handleQueuedFetches();
|
|
}catch(e){
|
|
errorCallback(e, keywordArgs);
|
|
}
|
|
});
|
|
getHandler.addErrback(function(error){
|
|
self._loadInProgress = false;
|
|
if(errorCallback){
|
|
errorCallback(error, keywordArgs);
|
|
}else{
|
|
throw error;
|
|
}
|
|
});
|
|
//Wire up the cancel to abort of the request
|
|
//This call cancel on the deferred if it hasn't been called
|
|
//yet and then will chain to the simple abort of the
|
|
//simpleFetch keywordArgs
|
|
var oldAbort = null;
|
|
if(keywordArgs.abort){
|
|
oldAbort = keywordArgs.abort;
|
|
}
|
|
keywordArgs.abort = function(){
|
|
var df = getHandler;
|
|
if(df && df.fired === -1){
|
|
df.cancel();
|
|
df = null;
|
|
}
|
|
if(oldAbort){
|
|
oldAbort.call(keywordArgs);
|
|
}
|
|
};
|
|
}
|
|
}else if(this._csvData){
|
|
try{
|
|
this._processData(this._csvData);
|
|
this._csvData = null;
|
|
filter(keywordArgs, this._arrayOfAllItems);
|
|
}catch(e){
|
|
errorCallback(e, keywordArgs);
|
|
}
|
|
}else{
|
|
var error = new Error(this.declaredClass + ": No CSV source data was provided as either URL or String data input.");
|
|
if(errorCallback){
|
|
errorCallback(error, keywordArgs);
|
|
}else{
|
|
throw error;
|
|
}
|
|
}
|
|
}
|
|
},
|
|
|
|
close: function(/*dojo.data.api.Request || keywordArgs || null */ request){
|
|
// summary:
|
|
// See dojo.data.api.Read.close()
|
|
},
|
|
|
|
|
|
// -------------------------------------------------------------------
|
|
// Private methods
|
|
_getArrayOfArraysFromCsvFileContents: function(/* string */ csvFileContents){
|
|
// summary:
|
|
// Parses a string of CSV records into a nested array structure.
|
|
// description:
|
|
// Given a string containing CSV records, this method parses
|
|
// the string and returns a data structure containing the parsed
|
|
// content. The data structure we return is an array of length
|
|
// R, where R is the number of rows (lines) in the CSV data. The
|
|
// return array contains one sub-array for each CSV line, and each
|
|
// sub-array contains C string values, where C is the number of
|
|
// columns in the CSV data.
|
|
// example:
|
|
// For example, given this CSV string as input:
|
|
// "Title, Year, Producer \n Alien, 1979, Ridley Scott \n Blade Runner, 1982, Ridley Scott"
|
|
// this._dataArray will be set to:
|
|
// [["Alien", "1979", "Ridley Scott"],
|
|
// ["Blade Runner", "1982", "Ridley Scott"]]
|
|
// And this._attributes will be set to:
|
|
// ["Title", "Year", "Producer"]
|
|
// And this._attributeIndexes will be set to:
|
|
// { "Title":0, "Year":1, "Producer":2 }
|
|
// tags:
|
|
// private
|
|
if(lang.isString(csvFileContents)){
|
|
var leadingWhiteSpaceCharacters = new RegExp("^\\s+",'g');
|
|
var trailingWhiteSpaceCharacters = new RegExp("\\s+$",'g');
|
|
var doubleQuotes = new RegExp('""','g');
|
|
var arrayOfOutputRecords = [];
|
|
var i;
|
|
|
|
var arrayOfInputLines = this._splitLines(csvFileContents);
|
|
for(i = 0; i < arrayOfInputLines.length; ++i){
|
|
var singleLine = arrayOfInputLines[i];
|
|
if(singleLine.length > 0){
|
|
var listOfFields = singleLine.split(this.separator);
|
|
var j = 0;
|
|
while(j < listOfFields.length){
|
|
var space_field_space = listOfFields[j];
|
|
var field_space = space_field_space.replace(leadingWhiteSpaceCharacters, ''); // trim leading whitespace
|
|
var field = field_space.replace(trailingWhiteSpaceCharacters, ''); // trim trailing whitespace
|
|
var firstChar = field.charAt(0);
|
|
var lastChar = field.charAt(field.length - 1);
|
|
var secondToLastChar = field.charAt(field.length - 2);
|
|
var thirdToLastChar = field.charAt(field.length - 3);
|
|
if(field.length === 2 && field == "\"\""){
|
|
listOfFields[j] = ""; //Special case empty string field.
|
|
}else if((firstChar == '"') &&
|
|
((lastChar != '"') ||
|
|
((lastChar == '"') && (secondToLastChar == '"') && (thirdToLastChar != '"')))){
|
|
if(j+1 === listOfFields.length){
|
|
// alert("The last field in record " + i + " is corrupted:\n" + field);
|
|
return; //null
|
|
}
|
|
var nextField = listOfFields[j+1];
|
|
listOfFields[j] = field_space + this.separator + nextField;
|
|
listOfFields.splice(j+1, 1); // delete element [j+1] from the list
|
|
}else{
|
|
if((firstChar == '"') && (lastChar == '"')){
|
|
field = field.slice(1, (field.length - 1)); // trim the " characters off the ends
|
|
field = field.replace(doubleQuotes, '"'); // replace "" with "
|
|
}
|
|
listOfFields[j] = field;
|
|
j += 1;
|
|
}
|
|
}
|
|
arrayOfOutputRecords.push(listOfFields);
|
|
}
|
|
}
|
|
|
|
// The first item of the array must be the header row with attribute names.
|
|
this._attributes = arrayOfOutputRecords.shift();
|
|
for(i = 0; i<this._attributes.length; i++){
|
|
// Store the index of each attribute
|
|
this._attributeIndexes[this._attributes[i]] = i;
|
|
}
|
|
this._dataArray = arrayOfOutputRecords; //Array
|
|
}
|
|
},
|
|
|
|
_splitLines: function(csvContent){
|
|
// summary:
|
|
// Function to split the CSV file contents into separate lines.
|
|
// Since line breaks can occur inside quotes, a Regexp didn't
|
|
// work as well. A quick passover parse should be just as efficient.
|
|
// tags:
|
|
// private
|
|
var split = [];
|
|
var i;
|
|
var line = "";
|
|
var inQuotes = false;
|
|
for(i = 0; i < csvContent.length; i++){
|
|
var c = csvContent.charAt(i);
|
|
switch(c){
|
|
case '\"':
|
|
inQuotes = !inQuotes;
|
|
line += c;
|
|
break;
|
|
case '\r':
|
|
if(inQuotes){
|
|
line += c;
|
|
}else{
|
|
split.push(line);
|
|
line = "";
|
|
if(i < (csvContent.length - 1) && csvContent.charAt(i + 1) == '\n'){
|
|
i++; //Skip it, it's CRLF
|
|
}
|
|
}
|
|
break;
|
|
case '\n':
|
|
if(inQuotes){
|
|
line += c;
|
|
}else{
|
|
split.push(line);
|
|
line = "";
|
|
}
|
|
break;
|
|
default:
|
|
line +=c;
|
|
}
|
|
}
|
|
if(line !== ""){
|
|
split.push(line);
|
|
}
|
|
return split;
|
|
},
|
|
|
|
_processData: function(/* String */ data){
|
|
// summary:
|
|
// Function for processing the string data from the server.
|
|
// data: String
|
|
// The CSV data.
|
|
// tags:
|
|
// private
|
|
this._getArrayOfArraysFromCsvFileContents(data);
|
|
this._arrayOfAllItems = [];
|
|
|
|
//Check that the specified Identifier is actually a column title, if provided.
|
|
if(this.identifier){
|
|
if(this._attributeIndexes[this.identifier] === undefined){
|
|
throw new Error(this.declaredClass + ": Identity specified is not a column header in the data set.");
|
|
}
|
|
}
|
|
|
|
for(var i=0; i<this._dataArray.length; i++){
|
|
var id = i;
|
|
//Associate the identifier to a row in this case
|
|
//for o(1) lookup.
|
|
if(this.identifier){
|
|
var iData = this._dataArray[i];
|
|
id = iData[this._attributeIndexes[this.identifier]];
|
|
this._idMap[id] = i;
|
|
}
|
|
this._arrayOfAllItems.push(this._createItemFromIdentity(id));
|
|
}
|
|
this._loadFinished = true;
|
|
this._loadInProgress = false;
|
|
},
|
|
|
|
_createItemFromIdentity: function(/* String */ identity){
|
|
// summary:
|
|
// Function for creating a new item from its identifier.
|
|
// identity: String
|
|
// The identity
|
|
// tags:
|
|
// private
|
|
var item = {};
|
|
item[this._storeProp] = this;
|
|
item[this._idProp] = identity;
|
|
return item; //Object
|
|
},
|
|
|
|
|
|
/***************************************
|
|
dojo.data.api.Identity API
|
|
***************************************/
|
|
getIdentity: function(/* item */ item){
|
|
// summary:
|
|
// See dojo.data.api.Identity.getIdentity()
|
|
// tags:
|
|
// public
|
|
if(this.isItem(item)){
|
|
return item[this._idProp]; //String
|
|
}
|
|
return null; //null
|
|
},
|
|
|
|
fetchItemByIdentity: function(/* Object */ keywordArgs){
|
|
// summary:
|
|
// See dojo.data.api.Identity.fetchItemByIdentity()
|
|
// tags:
|
|
// public
|
|
var item;
|
|
var scope = keywordArgs.scope?keywordArgs.scope:winUtil.global;
|
|
//Hasn't loaded yet, we have to trigger the load.
|
|
if(!this._loadFinished){
|
|
var self = this;
|
|
if(this.url !== ""){
|
|
//If fetches come in before the loading has finished, but while
|
|
//a load is in progress, we have to defer the fetching to be
|
|
//invoked in the callback.
|
|
if(this._loadInProgress){
|
|
this._queuedFetches.push({args: keywordArgs});
|
|
}else{
|
|
this._loadInProgress = true;
|
|
var getArgs = {
|
|
url: self.url,
|
|
handleAs: "text"
|
|
};
|
|
var getHandler = xhr.get(getArgs);
|
|
getHandler.addCallback(function(data){
|
|
try{
|
|
self._processData(data);
|
|
var item = self._createItemFromIdentity(keywordArgs.identity);
|
|
if(!self.isItem(item)){
|
|
item = null;
|
|
}
|
|
if(keywordArgs.onItem){
|
|
keywordArgs.onItem.call(scope, item);
|
|
}
|
|
self._handleQueuedFetches();
|
|
}catch(error){
|
|
if(keywordArgs.onError){
|
|
keywordArgs.onError.call(scope, error);
|
|
}
|
|
}
|
|
});
|
|
getHandler.addErrback(function(error){
|
|
this._loadInProgress = false;
|
|
if(keywordArgs.onError){
|
|
keywordArgs.onError.call(scope, error);
|
|
}
|
|
});
|
|
}
|
|
}else if(this._csvData){
|
|
try{
|
|
self._processData(self._csvData);
|
|
self._csvData = null;
|
|
item = self._createItemFromIdentity(keywordArgs.identity);
|
|
if(!self.isItem(item)){
|
|
item = null;
|
|
}
|
|
if(keywordArgs.onItem){
|
|
keywordArgs.onItem.call(scope, item);
|
|
}
|
|
}catch(e){
|
|
if(keywordArgs.onError){
|
|
keywordArgs.onError.call(scope, e);
|
|
}
|
|
}
|
|
}
|
|
}else{
|
|
//Already loaded. We can just look it up and call back.
|
|
item = this._createItemFromIdentity(keywordArgs.identity);
|
|
if(!this.isItem(item)){
|
|
item = null;
|
|
}
|
|
if(keywordArgs.onItem){
|
|
keywordArgs.onItem.call(scope, item);
|
|
}
|
|
}
|
|
},
|
|
|
|
getIdentityAttributes: function(/* item */ item){
|
|
// summary:
|
|
// See dojo.data.api.Identity.getIdentifierAttributes()
|
|
// tags:
|
|
// public
|
|
|
|
//Identity isn't a public attribute in the item, it's the row position index.
|
|
//So, return null.
|
|
if(this.identifier){
|
|
return [this.identifier];
|
|
}else{
|
|
return null;
|
|
}
|
|
},
|
|
|
|
_handleQueuedFetches: function(){
|
|
// summary:
|
|
// Internal function to execute delayed request in the store.
|
|
// tags:
|
|
// private
|
|
|
|
//Execute any deferred fetches now.
|
|
if(this._queuedFetches.length > 0){
|
|
for(var i = 0; i < this._queuedFetches.length; i++){
|
|
var fData = this._queuedFetches[i];
|
|
var delayedFilter = fData.filter;
|
|
var delayedQuery = fData.args;
|
|
if(delayedFilter){
|
|
delayedFilter(delayedQuery, this._arrayOfAllItems);
|
|
}else{
|
|
this.fetchItemByIdentity(fData.args);
|
|
}
|
|
}
|
|
this._queuedFetches = [];
|
|
}
|
|
}
|
|
});
|
|
//Mix in the simple fetch implementation to this class.
|
|
lang.extend(CsvStore, simpleFetch);
|
|
|
|
return CsvStore;
|
|
});
|