1
0
mirror of https://github.com/Jermolene/TiddlyWiki5 synced 2024-12-04 23:39:57 +00:00
TiddlyWiki5/core/modules/deserializers.js
Jeremy Ruston 82c8fe7fa8
Introduce JSON parse utility function with error handling (#6401)
* Introduce JSON parse utility function with error handling

Fixes #6400

* Fix typo
2022-02-21 15:29:25 +00:00

186 lines
5.5 KiB
JavaScript

/*\
title: $:/core/modules/deserializers.js
type: application/javascript
module-type: tiddlerdeserializer
Functions to deserialise tiddlers from a block of text
\*/
(function(){
/*jslint node: true, browser: true */
/*global $tw: false */
"use strict";
exports["application/x-tiddler-html-div"] = function(text,fields) {
return [deserializeTiddlerDiv(text,fields)];
};
exports["application/json"] = function(text,fields) {
var results = [],
incoming = $tw.utils.parseJSONSafe(text,function(err) {
return [{
title: "JSON error: " + err,
text: ""
}];
});
if(!$tw.utils.isArray(incoming)) {
incoming = [incoming];
}
for(var t=0; t<incoming.length; t++) {
var incomingFields = incoming[t],
fields = {};
for(var f in incomingFields) {
if(typeof incomingFields[f] === "string") {
fields[f] = incomingFields[f];
}
}
results.push(fields);
}
return results;
};
/*
Parse an HTML file into tiddlers. There are three possibilities:
# A TiddlyWiki classic HTML file containing `text/x-tiddlywiki` tiddlers
# A TiddlyWiki5 HTML file containing `text/vnd.tiddlywiki` tiddlers
# An ordinary HTML file
*/
exports["text/html"] = function(text,fields) {
var results = [];
// Check if we've got an old-style store area
var storeAreaMarkerRegExp = /<div id=["']?storeArea['"]?( style=["']?display:none;["']?)?>/gi,
storeAreaMatch = storeAreaMarkerRegExp.exec(text);
if(storeAreaMatch) {
// If so, we've got tiddlers in classic TiddlyWiki format or unencrypted old-style TW5 format
results.push.apply(results,deserializeStoreArea(text,storeAreaMarkerRegExp.lastIndex,!!storeAreaMatch[1],fields));
}
// Check for new-style store areas
var newStoreAreaMarkerRegExp = /<script class="tiddlywiki-tiddler-store" type="([^"]*)">/gi,
newStoreAreaMatch = newStoreAreaMarkerRegExp.exec(text),
haveHadNewStoreArea = !!newStoreAreaMatch;
while(newStoreAreaMatch) {
results.push.apply(results,deserializeNewStoreArea(text,newStoreAreaMarkerRegExp.lastIndex,newStoreAreaMatch[1],fields));
newStoreAreaMatch = newStoreAreaMarkerRegExp.exec(text);
}
// Return if we had either an old-style or a new-style store area
if(storeAreaMatch || haveHadNewStoreArea) {
return results;
}
// Otherwise, check whether we've got an encrypted file
var encryptedStoreArea = $tw.utils.extractEncryptedStoreArea(text);
if(encryptedStoreArea) {
// If so, attempt to decrypt it using the current password
return $tw.utils.decryptStoreArea(encryptedStoreArea);
} else {
// It's not a TiddlyWiki so we'll return the entire HTML file as a tiddler
return deserializeHtmlFile(text,fields);
}
};
function deserializeHtmlFile(text,fields) {
var result = {};
$tw.utils.each(fields,function(value,name) {
result[name] = value;
});
result.text = text;
result.type = "text/html";
return [result];
}
function deserializeNewStoreArea(text,storeAreaEnd,type,fields) {
var endOfScriptRegExp = /<\/script>/gi;
endOfScriptRegExp.lastIndex = storeAreaEnd;
var match = endOfScriptRegExp.exec(text);
if(match) {
var scriptContent = text.substring(storeAreaEnd,match.index);
return $tw.wiki.deserializeTiddlers(type,scriptContent);
} else {
return [];
}
}
function deserializeStoreArea(text,storeAreaEnd,isTiddlyWiki5,fields) {
var results = [],
endOfDivRegExp = /(<\/div>\s*)/gi,
startPos = storeAreaEnd,
defaultType = isTiddlyWiki5 ? undefined : "text/x-tiddlywiki";
endOfDivRegExp.lastIndex = startPos;
var match = endOfDivRegExp.exec(text);
while(match) {
var endPos = endOfDivRegExp.lastIndex,
tiddlerFields = deserializeTiddlerDiv(text.substring(startPos,endPos),fields,{type: defaultType});
if(!tiddlerFields) {
break;
}
$tw.utils.each(tiddlerFields,function(value,name) {
if(typeof value === "string") {
tiddlerFields[name] = $tw.utils.htmlDecode(value);
}
});
if(tiddlerFields.text !== null) {
results.push(tiddlerFields);
}
startPos = endPos;
match = endOfDivRegExp.exec(text);
}
return results;
}
/*
Utility function to parse an old-style tiddler DIV in a *.tid file. It looks like this:
<div title="Title" creator="JoeBloggs" modifier="JoeBloggs" created="201102111106" modified="201102111310" tags="myTag [[my long tag]]">
<pre>The text of the tiddler (without the expected HTML encoding).
</pre>
</div>
Note that the field attributes are HTML encoded, but that the body of the <PRE> tag is not encoded.
When these tiddler DIVs are encountered within a TiddlyWiki HTML file then the body is encoded in the usual way.
*/
var deserializeTiddlerDiv = function(text /* [,fields] */) {
// Slot together the default results
var result = {};
if(arguments.length > 1) {
for(var f=1; f<arguments.length; f++) {
var fields = arguments[f];
for(var t in fields) {
result[t] = fields[t];
}
}
}
// Parse the DIV body
var startRegExp = /^\s*<div\s+([^>]*)>(\s*<pre>)?/gi,
endRegExp,
match = startRegExp.exec(text);
if(match) {
// Old-style DIVs don't have the <pre> tag
if(match[2]) {
endRegExp = /<\/pre>\s*<\/div>\s*$/gi;
} else {
endRegExp = /<\/div>\s*$/gi;
}
var endMatch = endRegExp.exec(text);
if(endMatch) {
// Extract the text
result.text = text.substring(match.index + match[0].length,endMatch.index);
// Process the attributes
var attrRegExp = /\s*([^=\s]+)\s*=\s*(?:"([^"]*)"|'([^']*)')/gi,
attrMatch;
do {
attrMatch = attrRegExp.exec(match[1]);
if(attrMatch) {
var name = attrMatch[1];
var value = attrMatch[2] !== undefined ? attrMatch[2] : attrMatch[3];
result[name] = value;
}
} while(attrMatch);
return result;
}
}
return undefined;
};
})();