troggle-unchained/media/CodeMirror-0.62/js/parsexml.js
2009-06-28 21:26:35 +01:00

293 lines
8.4 KiB
JavaScript

/* This file defines an XML parser, with a few kludges to make it
* useable for HTML. autoSelfClosers defines a set of tag names that
* are expected to not have a closing tag, and doNotIndent specifies
* the tags inside of which no indentation should happen (see Config
* object). These can be disabled by passing the editor an object like
* {useHTMLKludges: false} as parserConfig option.
*/
var XMLParser = Editor.Parser = (function() {
var Kludges = {
autoSelfClosers: {"br": true, "img": true, "hr": true, "link": true, "input": true,
"meta": true, "col": true, "frame": true, "base": true, "area": true},
doNotIndent: {"pre": true, "!cdata": true}
};
var NoKludges = {autoSelfClosers: {}, doNotIndent: {"!cdata": true}};
var UseKludges = Kludges;
var alignCDATA = false;
// Simple stateful tokenizer for XML documents. Returns a
// MochiKit-style iterator, with a state property that contains a
// function encapsulating the current state. See tokenize.js.
var tokenizeXML = (function() {
function inText(source, setState) {
var ch = source.next();
if (ch == "<") {
if (source.equals("!")) {
source.next();
if (source.equals("[")) {
if (source.lookAhead("[CDATA[", true)) {
setState(inBlock("xml-cdata", "]]>"));
return null;
}
else {
return "xml-text";
}
}
else if (source.lookAhead("--", true)) {
setState(inBlock("xml-comment", "-->"));
return null;
}
else {
return "xml-text";
}
}
else if (source.equals("?")) {
source.next();
source.nextWhileMatches(/[\w\._\-]/);
setState(inBlock("xml-processing", "?>"));
return "xml-processing";
}
else {
if (source.equals("/")) source.next();
setState(inTag);
return "xml-punctuation";
}
}
else if (ch == "&") {
while (!source.endOfLine()) {
if (source.next() == ";")
break;
}
return "xml-entity";
}
else {
source.nextWhileMatches(/[^&<\n]/);
return "xml-text";
}
}
function inTag(source, setState) {
var ch = source.next();
if (ch == ">") {
setState(inText);
return "xml-punctuation";
}
else if (/[?\/]/.test(ch) && source.equals(">")) {
source.next();
setState(inText);
return "xml-punctuation";
}
else if (ch == "=") {
return "xml-punctuation";
}
else if (/[\'\"]/.test(ch)) {
setState(inAttribute(ch));
return null;
}
else {
source.nextWhileMatches(/[^\s\u00a0=<>\"\'\/?]/);
return "xml-name";
}
}
function inAttribute(quote) {
return function(source, setState) {
while (!source.endOfLine()) {
if (source.next() == quote) {
setState(inTag);
break;
}
}
return "xml-attribute";
};
}
function inBlock(style, terminator) {
return function(source, setState) {
while (!source.endOfLine()) {
if (source.lookAhead(terminator, true)) {
setState(inText);
break;
}
source.next();
}
return style;
};
}
return function(source, startState) {
return tokenizer(source, startState || inText);
};
})();
// The parser. The structure of this function largely follows that of
// parseJavaScript in parsejavascript.js (there is actually a bit more
// shared code than I'd like), but it is quite a bit simpler.
function parseXML(source) {
var tokens = tokenizeXML(source);
var cc = [base];
var tokenNr = 0, indented = 0;
var currentTag = null, context = null;
var consume, marked;
function push(fs) {
for (var i = fs.length - 1; i >= 0; i--)
cc.push(fs[i]);
}
function cont() {
push(arguments);
consume = true;
}
function pass() {
push(arguments);
consume = false;
}
function mark(style) {
marked = style;
}
function expect(text) {
return function(style, content) {
if (content == text) cont();
else mark("xml-error") || cont(arguments.callee);
};
}
function pushContext(tagname, startOfLine) {
var noIndent = UseKludges.doNotIndent.hasOwnProperty(tagname) || (context && context.noIndent);
context = {prev: context, name: tagname, indent: indented, startOfLine: startOfLine, noIndent: noIndent};
}
function popContext() {
context = context.prev;
}
function computeIndentation(baseContext) {
return function(nextChars, current) {
var context = baseContext;
if (context && context.noIndent)
return current;
if (alignCDATA && /<!\[CDATA\[/.test(nextChars))
return 0;
if (context && /^<\//.test(nextChars))
context = context.prev;
while (context && !context.startOfLine)
context = context.prev;
if (context)
return context.indent + indentUnit;
else
return 0;
};
}
function base() {
return pass(element, base);
}
var harmlessTokens = {"xml-text": true, "xml-entity": true, "xml-comment": true, "xml-processing": true};
function element(style, content) {
if (content == "<") cont(tagname, attributes, endtag(tokenNr == 1));
else if (content == "</") cont(closetagname, expect(">"));
else if (style == "xml-cdata") {
if (!context || context.name != "!cdata") pushContext("!cdata");
if (/\]\]>$/.test(content)) popContext();
cont();
}
else if (harmlessTokens.hasOwnProperty(style)) cont();
else mark("xml-error") || cont();
}
function tagname(style, content) {
if (style == "xml-name") {
currentTag = content.toLowerCase();
mark("xml-tagname");
cont();
}
else {
currentTag = null;
pass();
}
}
function closetagname(style, content) {
if (style == "xml-name" && context && content.toLowerCase() == context.name) {
popContext();
mark("xml-tagname");
}
else {
mark("xml-error");
}
cont();
}
function endtag(startOfLine) {
return function(style, content) {
if (content == "/>" || (content == ">" && UseKludges.autoSelfClosers.hasOwnProperty(currentTag))) cont();
else if (content == ">") pushContext(currentTag, startOfLine) || cont();
else mark("xml-error") || cont(arguments.callee);
};
}
function attributes(style) {
if (style == "xml-name") mark("xml-attname") || cont(attribute, attributes);
else pass();
}
function attribute(style, content) {
if (content == "=") cont(value);
else if (content == ">" || content == "/>") pass(endtag);
else pass();
}
function value(style) {
if (style == "xml-attribute") cont(value);
else pass();
}
return {
indentation: function() {return indented;},
next: function(){
var token = tokens.next();
if (token.style == "whitespace" && tokenNr == 0)
indented = token.value.length;
else
tokenNr++;
if (token.content == "\n") {
indented = tokenNr = 0;
token.indentation = computeIndentation(context);
}
if (token.style == "whitespace" || token.type == "xml-comment")
return token;
while(true){
consume = marked = false;
cc.pop()(token.style, token.content);
if (consume){
if (marked)
token.style = marked;
return token;
}
}
},
copy: function(){
var _cc = cc.concat([]), _tokenState = tokens.state, _context = context;
var parser = this;
return function(input){
cc = _cc.concat([]);
tokenNr = indented = 0;
context = _context;
tokens = tokenizeXML(input, _tokenState);
return parser;
};
}
};
}
return {
make: parseXML,
electricChars: "/",
configure: function(config) {
if (config.useHTMLKludges != null)
UseKludges = config.useHTMLKludges ? Kludges : NoKludges;
if (config.alignCDATA)
alignCDATA = config.alignCDATA;
}
};
})();