// Copyright 2005 Google Inc.
// All Rights Reserved
//
// An XPath parser and evaluator written in JavaScript. The
// implementation is complete except for functions handling
// namespaces.
//
// Reference: [XPATH] XPath Specification
// <http://www.w3.org/TR/1999/REC-xpath-19991116>.
//
//
// The API of the parser has several parts:
//
// 1. The parser function xpathParse() that takes a string and returns
// an expession object.
//
// 2. The expression object that has an evaluate() method to evaluate the
// XPath expression it represents. (It is actually a hierarchy of
// objects that resembles the parse tree, but an application will call
// evaluate() only on the top node of this hierarchy.)
//
// 3. The context object that is passed as an argument to the evaluate()
// method, which represents the DOM context in which the expression is
// evaluated.
//
// 4. The value object that is returned from evaluate() and represents
// values of the different types that are defined by XPath (number,
// string, boolean, and node-set), and allows to convert between them.
//
// These parts are near the top of the file, the functions and data
// that are used internally follow after them.
//
//
// Author: Steffen Meschkat <mesch@google.com>
// The entry point for the parser.
//
// @param expr a string that contains an XPath expression.
// @return an expression object that can be evaluated with an
// expression context.
function xpathParse(expr) {
xpathLog('parse ' + expr);
xpathParseInit();
var cached = xpathCacheLookup(expr);
if (cached) {
xpathLog(' ... cached');
return cached;
}
// Optimize for a few common cases: simple attribute node tests
// (@id), simple element node tests (page), variable references
// ($address), numbers (4), multi-step path expressions where each
// step is a plain element node test
// (page/overlay/locations/location).
if (expr.match(/^(\$|@)?\w+$/i)) {
var ret = makeSimpleExpr(expr);
xpathParseCache[expr] = ret;
xpathLog(' ... simple');
return ret;
}
if (expr.match(/^\w+(\/\w+)*$/i)) {
var ret = makeSimpleExpr2(expr);
xpathParseCache[expr] = ret;
xpathLog(' ... simple 2');
return ret;
}
var cachekey = expr; // expr is modified during parse
var stack = [];
var ahead = null;
var previous = null;
var done = false;
var parse_count = 0;
var lexer_count = 0;
var reduce_count = 0;
while (!done) {
parse_count++;
expr = expr.replace(/^\s*/, '');
previous = ahead;
ahead = null;
var rule = null;
var match = '';
for (var i = 0; i < xpathTokenRules.length; ++i) {
var result = xpathTokenRules[i].re.exec(expr);
lexer_count++;
if (result && result.length > 0 && result[0].length > match.length) {
rule = xpathTokenRules[i];
match = result[0];
break;
}
}
// Special case: allow operator keywords to be element and
// variable names.
// NOTE(mesch): The parser resolves conflicts by looking ahead,
// and this is the only case where we look back to
// disambiguate. So this is indeed something different, and
// looking back is usually done in the lexer (via states in the
// general case, called "start conditions" in flex(1)). Also,the
// conflict resolution in the parser is not as robust as it could
// be, so I'd like to keep as much off the parser as possible (all
// these precedence values should be computed from the grammar
// rules and possibly associativity declarations, as in bison(1),
// and not explicitly set.
if (rule &&
(rule == TOK_DIV ||
rule == TOK_MOD ||
rule == TOK_AND ||
rule == TOK_OR) &&
(!previous ||
previous.tag == TOK_AT ||
previous.tag == TOK_DSLASH ||
previous.tag == TOK_SLASH ||
previous.tag == TOK_AXIS ||
previous.tag == TOK_DOLLAR)) {
rule = TOK_QNAME;
}
if (rule) {
expr = expr.substr(match.length);
xpathLog('token: ' + match + ' -- ' + rule.label);
ahead = {
tag: rule,
match: match,
prec: rule.prec ? rule.prec : 0, // || 0 is removed by the compiler
expr: makeTokenExpr(match)
};
} else {
xpathLog('DONE');
done = true;
}
while (xpathReduce(stack, ahead)) {
reduce_count++;
xpathLog('stack: ' + stackToString(stack));
}
}
xpathLog('stack: ' + stackToString(stack));
// DGF any valid XPath should "reduce" to a single Expr token
if (stack.length != 1) {
throw 'XPath parse error ' + cachekey + ':\n' + stackToString(stack);
}
var result = stack[0].expr;
xpathParseCache[cachekey] = result;
xpathLog('XPath parse: ' + parse_count + ' / ' +
lexer_count + ' / ' + reduce_count);
return result;
}
var xpathParseCache = {};
function xpathCacheLookup(expr) {
return xpathParseCache[expr];
}
/*DGF xpathReduce is where the magic happens in this parser.
Skim down to the bottom of this file to find the table of
grammatical rules and precedence numbers, "The productions of the grammar".
The idea here
is that we want to take a stack of tokens and apply
grammatical rules to them, "reducing" them to higher-level
tokens. Ultimately, any valid XPath should reduce to exactly one
"Expr" token.
Reduce too early or too late and you'll have two tokens that can't reduce
to single Expr. For example, you may hastily reduce a qname that
should name a function, incorrectly treating it as a tag name.
Or you may reduce too late, accidentally reducing the last part of the
XPath into a top-level "Expr" that won't reduce with earlier parts of
the XPath.
A "cand" is a grammatical rule candidate, with a given precedence
number. "ahead" is the upcoming token, which also has a precedence
number. If the token has a higher precedence number than
the rule candidate, we'll "shift" the token onto the token stack,
instead of immediately applying the rule candidate.
Some tokens have left associativity, in which case we shift when they
have LOWER precedence than the candidate.
*/
function xpathReduce(stack, ahead) {
var cand = null;
if (stack.length > 0) {
var top = stack[stack.length-1];
var ruleset = xpathRules[top.tag.key];
if (ruleset) {
for (var i = 0; i < ruleset.length; ++i) {
var rule = ruleset[i];
var match = xpathMatchStack(stack, rule[1]);
if (match.length) {
cand = {
tag: rule[0],
rule: rule,
match: match
};
cand.prec = xpathGrammarPrecedence(cand);
break;
}
}
}
}
var ret;
if (cand && (!ahead || cand.prec > ahead.prec ||
(ahead.tag.left && cand.prec >= ahead.prec))) {
for (var i = 0; i < cand.match.matchlength; ++i) {
stack.pop();
}
xpathLog('reduce ' + cand.tag.label + ' ' + cand.prec +
' ahead ' + (ahead ? ahead.tag.label + ' ' + ahead.prec +
(ahead.tag.left ? ' left' : '')
: ' none '));
var matchexpr = mapExpr(cand.match, function(m) { return m.expr; });
xpathLog('going to apply ' + cand.rule[3].toString());
cand.expr = cand.rule[3].apply(null, matchexpr);
stack.push(cand);
ret = true;
} else {
if (ahead) {
xpathLog('shift ' + ahead.tag.label + ' ' + ahead.prec +
(ahead.tag.left ? ' left' : '') +
' over ' + (cand ? cand.tag.label + ' ' +
cand.prec : ' none'));
stack.push(ahead);
}
ret = false;
}
return ret;
}
function xpathMatchStack(stack, pattern) {
// NOTE(mesch): The stack matches for variable cardinality are
// greedy but don't do backtracking. This would be an issue only
// with rules of the form A* A, i.e. with an element with variable
// cardinality followed by the same element. Since that doesn't
// occur in the grammar at hand, all matches on the stack are
// unambiguous.
var S = stack.length;
var P = pattern.length;
var p, s;
var match = [];
match.matchlength = 0;
var ds = 0;
for (p = P - 1, s = S - 1; p >= 0 && s >= 0; --p, s -= ds) {
ds = 0;
var qmatch = [];
if (pattern[p] == Q_MM) {
p -= 1;
match.push(qmatch);
while (s - ds >= 0 && stack[s - ds].tag == pattern[p]) {
qmatch.push(stack[s - ds]);
ds += 1;
match.matchlength += 1;
}
} else if (pattern[p] == Q_01) {
p -= 1;
match.push(qmatch);
while (s - ds >= 0 && ds < 2 && stack[s - ds].tag == pattern[p]) {
qmatch.push(stack[s - ds]);
ds += 1;
match.matchlength += 1;
}
} else if (pattern[p] == Q_1M) {
p -= 1;
match.push(qmatch);
if (stack[s].tag == pattern[p]) {
while (s - ds >= 0 && stack[s - ds].tag == pattern[p]) {
qmatch.push(stack[s - ds]);
ds += 1;
match.matchlength += 1;
}
} else {
return [];
}
} else if (stack[s].tag == pattern[p]) {
match.push(stack[s]);
ds += 1;
match.matchlength += 1;
} else {
return [];
}
reverseInplace(qmatch);
qmatch.expr = mapExpr(qmatch, function(m) { return m.expr; });
}
reverseInplace(match);
if (p == -1) {
return match;
} else {
return [];
}
}
function xpathTokenPrecedence(tag) {
return tag.prec || 2;
}
function xpathGrammarPrecedence(frame) {
var ret = 0;
if (frame.rule) { /* normal reduce */
if (frame.rule.length >= 3 && frame.rule[2] >= 0) {
ret = frame.rule[2];
} else {
for (var i = 0; i < frame.rule[1].length; ++i) {
var p = xpathTokenPrecedence(frame.rule[1][i]);
ret = Math.max(ret, p);
}
}
} else if (frame.tag) { /* TOKEN match */
ret = xpathTokenPrecedence(frame.tag);
} else if (frame.length) { /* Q_ match */
for (var j = 0; j < frame.length; ++j) {
var p = xpathGrammarPrecedence(frame[j]);
ret = Math.max(ret, p);
}
}
return ret;
}
function stackToString(stack) {
var ret = '';
for (var i = 0; i < stack.length; ++i) {
if (ret) {
ret += '\n';
}
ret += stack[i].tag.label;
}
return ret;
}
// XPath expression evaluation context. An XPath context consists of a
// DOM node, a list of DOM nodes that contains this node, a number
// that represents the position of the single node in the list, and a
// current set of variable bindings. (See XPath spec.)
//
// The interface of the expression context:
//
// Constructor -- gets the node, its position, the node set it
// belongs to, and a parent context as arguments. The parent context
// is used to implement scoping rules for variables: if a variable
// is not found in the current context, it is looked for in the
// parent context, recursively. Except for node, all arguments have
// default values: default position is 0, default node set is the
// set that contains only the node, and the default parent is null.
//
// Notice that position starts at 0 at the outside interface;
// inside XPath expressions this shows up as position()=1.
//
// clone() -- creates a new context with the current context as
// parent. If passed as argument to clone(), the new context has a
// different node, position, or node set. What is not passed is
// inherited from the cloned context.
//
// setVariable(name, expr) -- binds given XPath expression to the
// name.
//
// getVariable(name) -- what the name says.
//
// setNode(position) -- sets the context to the node at the given
// position. Needed to implement scoping rules for variables in
// XPath. (A variable is visible to all subsequent siblings, not
// only to its children.)
//
// set/isCaseInsensitive -- specifies whether node name tests should
// be case sensitive. If you're executing xpaths against a regular
// HTML DOM, you probably don't want case-sensitivity, because
// browsers tend to disagree about whether elements & attributes
// should be upper/lower case. If you're running xpaths in an
// XSLT instance, you probably DO want case sensitivity, as per the
// XSL spec.
function ExprContext(node, opt_position, opt_nodelist, opt_parent, opt_caseInsensitive, opt_ignoreAttributesWithoutValue) {
this.node = node;
this.position = opt_position || 0;
this.nodelist = opt_nodelist || [ node ];
this.variables = {};
this.parent = opt_parent || null;
this.caseInsensitive = opt_caseInsensitive || false;
this.ignoreAttributesWithoutValue = opt_ignoreAttributesWithoutValue || false;
if (opt_parent) {
this.root = opt_parent.root;
} else if (this.node.nodeType == DOM_DOCUMENT_NODE) {
// NOTE(mesch): DOM Spec stipulates that the ownerDocument of a
// document is null. Our root, however is the document that we are
// processing, so the initial context is created from its document
// node, which case we must handle here explcitly.
this.root = node;
} else {
this.root = node.ownerDocument;
}
}
ExprContext.prototype.clone = function(opt_node, opt_position, opt_nodelist) {
return new ExprContext(
opt_node || this.node,
typeof opt_position != 'undefined' ? opt_position : this.position,
opt_nodelist || this.nodelist, this, this.caseInsensitive,
this.ignoreAttributesWithoutValue);
};
ExprContext.prototype.setVariable = function(name, value) {
if (value instanceof StringValue || value instanceof BooleanValue ||
value instanceof NumberValue || value instanceof NodeSetValue) {
this.variables[name] = value;
return;
}
if ('true' === value) {
this.variables[name] = new BooleanValue(true);
} else if ('false' === value) {
this.variables[name] = new BooleanValue(false);
} else if (TOK_NUMBER.re.test(value)) {
this.variables[name] = new NumberValue(value);
} else {
// DGF What if it's null?
this.variables[name] = new StringValue(value);
}
};
ExprContext.prototype.getVariable = function(name) {
if (typeof this.variables[name] != 'undefined') {
return this.variables[name];
} else if (this.parent) {
return this.parent.getVariable(name);
} else {
return null;
}
};
ExprContext.prototype.setNode = function(position) {
this.node = this.nodelist[position];
this.position = position;
};
ExprContext.prototype.contextSize = function() {
return this.nodelist.length;
};
ExprContext.prototype.isCaseInsensitive = function() {
return this.caseInsensitive;
};
ExprContext.prototype.setCaseInsensitive = function(caseInsensitive) {
return this.caseInsensitive = caseInsensitive;
};
ExprContext.prototype.isIgnoreAttributesWithoutValue = function() {
return this.ignoreAttributesWithoutValue;
};
ExprContext.prototype.setIgnoreAttributesWithoutValue = function(ignore) {
return this.ignoreAttributesWithoutValue = ignore;
};
// XPath expression values. They are what XPath expressions evaluate
// to. Strangely, the different value types are not specified in the
// XPath syntax, but only in the semantics, so they don't show up as
// nonterminals in the grammar. Yet, some expressions are required to
// evaluate to particular types, and not every type can be coerced
// into every other type. Although the types of XPath values are
// similar to the types present in JavaScript, the type coercion rules
// are a bit peculiar, so we explicitly model XPath types instead of
// mapping them onto JavaScript types. (See XPath spec.)
//
// The four types are:
//
// StringValue
//
// NumberValue
//
// BooleanValue
//
// NodeSetValue
//
// The common interface of the value classes consists of methods that
// implement the XPath type coercion rules:
//
// stringValue() -- returns the value as a JavaScript String,
//
// numberValue() -- returns the value as a JavaScript Number,
//
// booleanValue() -- returns the value as a JavaScript Boolean,
//
// nodeSetValue() -- returns the value as a JavaScript Array of DOM
// Node objects.
//
function StringValue(value) {
this.value = value;
this.type = 'string';
}
StringValue.prototype.stringValue = function() {
return this.value;
}
StringValue.prototype.booleanValue = function() {
return this.value.length > 0;
}
StringValue.prototype.numberValue = function() {
return this.value - 0;
}
StringValue.prototype.nodeSetValue = function() {
throw this;
}
function BooleanValue(value) {
this.value = value;
this.type = 'boolean';
}
BooleanValue.prototype.stringValue = function() {
return '' + this.value;
}
BooleanValue.prototype.booleanValue = function() {
return this.value;
}
BooleanValue.prototype.numberValue = function() {
return this.value ? 1 : 0;
}
BooleanValue.prototype.nodeSetValue = function() {
throw this;
}
function NumberValue(value) {
this.value = value;
this.type = 'number';
}
NumberValue.prototype.stringValue = function() {
return '' + this.value;
}
NumberValue.prototype.booleanValue = function() {
return !!this.value;
}
NumberValue.prototype.numberValue = function() {
return this.value - 0;
}
NumberValue.prototype.nodeSetValue = function() {
throw this;
}
function NodeSetValue(value) {
this.value = value;
this.type = 'node-set';
}
NodeSetValue.prototype.stringValue = function() {
if (this.value.length == 0) {
return '';
} else {
return xmlValue(this.value[0]);
}
}
NodeSetValue.prototype.booleanValue = function() {
return this.value.length > 0;
}
NodeSetValue.prototype.numberValue = function() {
return this.stringValue() - 0;
}
NodeSetValue.prototype.nodeSetValue = function() {
return this.value;
};
// XPath expressions. They are used as nodes in the parse tree and
// possess an evaluate() method to compute an XPath value given an XPath
// context. Expressions are returned from the parser. Teh set of
// expression classes closely mirrors the set of non terminal symbols
// in the grammar. Every non trivial nonterminal symbol has a
// corresponding expression class.
//
// The common expression interface consists of the following methods:
//
// evaluate(context) -- evaluates the expression, returns a value.
//
// toString() -- returns the XPath text representation of the
// expression (defined in xsltdebug.js).
//
// parseTree(indent) -- returns a parse tree representation of the
// expression (defined in xsltdebug.js).
function TokenExpr(m) {
this.value = m;
}
TokenExpr.prototype.evaluate = function() {
return new StringValue(this.value);
};
function LocationExpr() {
this.absolute = false;
this.steps = [];
}
LocationExpr.prototype.appendStep = function(s) {
var combinedStep = this._combineSteps(this.steps[this.steps.length-1], s);
if (combinedStep) {
this.steps[this.steps.length-1] = combinedStep;
} else {
this.steps.push(s);
}
}
LocationExpr.prototype.prependStep = function(s) {
var combinedStep = this._combineSteps(s, this.steps[0]);
if (combinedStep) {
this.steps[0] = combinedStep;
} else {
this.steps.unshift(s);
}
};
// DGF try to combine two steps into one step (perf enhancement)
LocationExpr.prototype._combineSteps = function(prevStep, nextStep) {
if (!prevStep) return null;
if (!nextStep) return null;
var hasPredicates = (prevStep.predicates && prevStep.predicates.length > 0);
if (prevStep.nodetest instanceof NodeTestAny && !hasPredicates) {
// maybe suitable to be combined
if (prevStep.axis == xpathAxis.DESCENDANT_OR_SELF) {
if (nextStep.axis == xpathAxis.CHILD) {
nextStep.axis = xpathAxis.DESCENDANT;
return nextStep;
} else if (nextStep.axis == xpathAxis.SELF) {
nextStep.axis = xpathAxis.DESCENDANT_OR_SELF;
return nextStep;
}
} else if (prevStep.axis == xpathAxis.DESCENDANT) {
if (nextStep.axis == xpathAxis.SELF) {
nextStep.axis = xpathAxis.DESCENDANT;
return nextStep;
}
}
}
return null;
}
LocationExpr.prototype.evaluate = function(ctx) {
var start;
if (this.absolute) {
start = ctx.root;
} else {
start = ctx.node;
}
var nodes = [];
xPathStep(nodes, this.steps, 0, start, ctx);
return new NodeSetValue(nodes);
};
function xPathStep(nodes, steps, step, input, ctx) {
var s = steps[step];
var ctx2 = ctx.clone(input);
var nodelist = s.evaluate(ctx2).nodeSetValue();
for (var i = 0; i < nodelist.length; ++i) {
if (step == steps.length - 1) {
nodes.push(nodelist[i]);
} else {
xPathStep(nodes, steps, step + 1, nodelist[i], ctx);
}
}
}
function StepExpr(axis, nodetest, opt_predicate) {
this.axis = axis;
this.nodetest = nodetest;
this.predicate = opt_predicate || [];
}
StepExpr.prototype.appendPredicate = function(p) {
this.predicate.push(p);
}
StepExpr.prototype.evaluate = function(ctx) {
var input = ctx.node;
var nodelist = [];
var skipNodeTest = false;
if (this.nodetest instanceof NodeTestAny) {
skipNodeTest = true;
}
// NOTE(mesch): When this was a switch() statement, it didn't work
// in Safari/2.0. Not sure why though; it resulted in the JavaScript
// console output "undefined" (without any line number or so).
if (this.axis == xpathAxis.ANCESTOR_OR_SELF) {
nodelist.push(input);
for (var n = input.parentNode; n; n = n.parentNode) {
nodelist.push(n);
}
} else if (this.axis == xpathAxis.ANCESTOR) {
for (var n = input.parentNode; n; n = n.parentNode) {
nodelist.push(n);
}
} else if (this.axis == xpathAxis.ATTRIBUTE) {
if (ctx.ignoreAttributesWithoutValue) {
copyArrayIgnoringAttributesWithoutValue(nodelist, input.attributes);
}
else {
copyArray(nodelist, input.attributes);
}
} else if (this.axis == xpathAxis.CHILD) {
copyArray(nodelist, input.childNodes);
} else if (this.axis == xpathAxis.DESCENDANT_OR_SELF) {
if (this.nodetest.evaluate(ctx).booleanValue()) {
nodelist.push(input);
}
var tagName = xpathExtractTagNameFromNodeTest(this.nodetest);
xpathCollectDescendants(nodelist, input, tagName);
if (tagName) skipNodeTest = true;
} else if (this.axis == xpathAxis.DESCENDANT) {
var tagName = xpathExtractTagNameFromNodeTest(this.nodetest);
xpathCollectDescendants(nodelist, input, tagName);
if (tagName) skipNodeTest = true;
} else if (this.axis == xpathAxis.FOLLOWING) {
for (var n = input; n; n = n.parentNode) {
for (var nn = n.nextSibling; nn; nn = nn.nextSibling) {
nodelist.push(nn);
xpathCollectDescendants(nodelist, nn);
}
}
} else if (this.axis == xpathAxis.FOLLOWING_SIBLING) {
for (var n = input.nextSibling; n; n = n.nextSibling) {
nodelist.push(n);
}
} else if (this.axis == xpathAxis.NAMESPACE) {
alert('not implemented: axis namespace');
} else if (this.axis == xpathAxis.PARENT) {
if (input.parentNode) {
nodelist.push(input.parentNode);
}
} else if (this.axis == xpathAxis.PRECEDING) {
for (var n = input; n; n = n.parentNode) {
for (var nn = n.previousSibling; nn; nn = nn.previousSibling) {
nodelist.push(nn);
xpathCollectDescendantsReverse(nodelist, nn);
}
}
} else if (this.axis == xpathAxis.PRECEDING_SIBLING) {
for (var n = input.previousSibling; n; n = n.previousSibling) {
nodelist.push(n);
}
} else if (this.axis == xpathAxis.SELF) {
nodelist.push(input);
} else {
throw 'ERROR -- NO SUCH AXIS: ' + this.axis;
}
if (!skipNodeTest) {
// process node test
var nodelist0 = nodelist;
nodelist = [];
for (var i = 0; i < nodelist0.length; ++i) {
var n = nodelist0[i];
if (this.nodetest.evaluate(ctx.clone(n, i, nodelist0)).booleanValue()) {
nodelist.push(n);
}
}
}
// process predicates
for (var i = 0; i < this.predicate.length; ++i) {
var nodelist0 = nodelist;
nodelist = [];
for (var ii = 0; ii < nodelist0.length; ++ii) {
var n = nodelist0[ii];
if (this.predicate[i].evaluate(ctx.clone(n, ii, nodelist0)).booleanValue()) {
nodelist.push(n);
}
}
}
return new NodeSetValue(nodelist);
};
function NodeTestAny() {
this.value = new BooleanValue(true);
}
NodeTestAny.prototype.evaluate = function(ctx) {
return this.value;
};
function NodeTestElementOrAttribute() {}
NodeTestElementOrAttribute.prototype.evaluate = function(ctx) {
return new BooleanValue(
ctx.node.nodeType == DOM_ELEMENT_NODE ||
ctx.node.nodeType == DOM_ATTRIBUTE_NODE);
}
function NodeTestText() {}
NodeTestText.prototype.evaluate = function(ctx) {
return new BooleanValue(ctx.node.nodeType == DOM_TEXT_NODE);
}
function NodeTestComment() {}
NodeTestComment.prototype.evaluate = function(ctx) {
return new BooleanValue(ctx.node.nodeType == DOM_COMMENT_NODE);
}
function NodeTestPI(target) {
this.target = target;
}
NodeTestPI.prototype.evaluate = function(ctx) {
return new
BooleanValue(ctx.node.nodeType == DOM_PROCESSING_INSTRUCTION_NODE &&
(!this.target || ctx.node.nodeName == this.target));
}
function NodeTestNC(nsprefix) {
this.regex = new RegExp("^" + nsprefix + ":");
this.nsprefix = nsprefix;
}
NodeTestNC.prototype.evaluate = function(ctx) {
var n = ctx.node;
return new BooleanValue(this.regex.match(n.nodeName));
}
function NodeTestName(name) {
this.name = name;
this.re = new RegExp('^' + name + '$', "i");
}
NodeTestName.prototype.evaluate = function(ctx) {
var n = ctx.node;
if (ctx.caseInsensitive) {
if (n.nodeName.length != this.name.length) return new BooleanValue(false);
return new BooleanValue(this.re.test(n.nodeName));
} else {
return new BooleanValue(n.nodeName == this.name);
}
}
function PredicateExpr(expr) {
this.expr = expr;
}
PredicateExpr.prototype.evaluate = function(ctx) {
var v = this.expr.evaluate(ctx);
if (v.type == 'number') {
// NOTE(mesch): Internally, position is represented starting with
// 0, however in XPath position starts with 1. See functions
// position() and last().
return new BooleanValue(ctx.position == v.numberValue() - 1);
} else {
return new BooleanValue(v.booleanValue());
}
};
function FunctionCallExpr(name) {
this.name = name;
this.args = [];
}
FunctionCallExpr.prototype.appendArg = function(arg) {
this.args.push(arg);
};
FunctionCallExpr.prototype.evaluate = function(ctx) {
var fn = '' + this.name.value;
var f = this.xpathfunctions[fn];
if (f) {
return f.call(this, ctx);
} else {
xpathLog('XPath NO SUCH FUNCTION ' + fn);
return new BooleanValue(false);
}
};
FunctionCallExpr.prototype.xpathfunctions = {
'last': function(ctx) {
assert(this.args.length == 0);
// NOTE(mesch): XPath position starts at 1.
return new NumberValue(ctx.contextSize());
},
'position': function(ctx) {
assert(this.args.length == 0);
// NOTE(mesch): XPath position starts at 1.
return new NumberValue(ctx.position + 1);
},
'count': function(ctx) {
assert(this.args.length == 1);
var v = this.args[0].evaluate(ctx);
return new NumberValue(v.nodeSetValue().length);
},
'id': function(ctx) {
assert(this.args.length == 1);
var e = this.args[0].evaluate(ctx);
var ret = [];
var ids;
if (e.type == 'node-set') {
ids = [];
var en = e.nodeSetValue();
for (var i = 0; i < en.length; ++i) {
var v = xmlValue(en[i]).split(/\s+/);
for (var ii = 0; ii < v.length; ++ii) {
ids.push(v[ii]);
}
}
} else {
ids = e.stringValue().split(/\s+/);
}
var d = ctx.root;
for (var i = 0; i < ids.length; ++i) {
var n = d.getElementById(ids[i]);
if (n) {
ret.push(n);
}
}
return new NodeSetValue(ret);
},
'local-name': function(ctx) {
alert('not implmented yet: XPath function local-name()');
},
'namespace-uri': function(ctx) {
alert('not implmented yet: XPath function namespace-uri()');
},
'name': function(ctx) {
assert(this.args.length == 1 || this.args.length == 0);
var n;
if (this.args.length == 0) {
n = [ ctx.node ];
} else {
n = this.args[0].evaluate(ctx).nodeSetValue();
}
if (n.length == 0) {
return new StringValue('');
} else {
return new StringValue(n[0].nodeName);
}
},
'string': function(ctx) {
assert(this.args.length == 1 || this.args.length == 0);
if (this.args.length == 0) {
return new StringValue(new NodeSetValue([ ctx.node ]).stringValue());
} else {
return new StringValue(this.args[0].evaluate(ctx).stringValue());
}
},
'concat': function(ctx) {
var ret = '';
for (var i = 0; i < this.args.length; ++i) {
ret += this.args[i].evaluate(ctx).stringValue();
}
return new StringValue(ret);
},
'starts-with': function(ctx) {
assert(this.args.length == 2);
var s0 = this.args[0].evaluate(ctx).stringValue();
var s1 = this.args[1].evaluate(ctx).stringValue();
return new BooleanValue(s0.indexOf(s1) == 0);
},
'contains': function(ctx) {
assert(this.args.length == 2);
var s0 = this.args[0].evaluate(ctx).stringValue();
var s1 = this.args[1].evaluate(ctx).stringValue();
return new BooleanValue(s0.indexOf(s1) != -1);
},
'substring-before': function(ctx) {
assert(this.args.length == 2);
var s0 = this.args[0].evaluate(ctx).stringValue();
var s1 = this.args[1].evaluate(ctx).stringValue();
var i = s0.indexOf(s1);
var ret;
if (i == -1) {
ret = '';
} else {
ret = s0.substr(0,i);
}
return new StringValue(ret);
},
'substring-after': function(ctx) {
assert(this.args.length == 2);
var s0 = this.args[0].evaluate(ctx).stringValue();
var s1 = this.args[1].evaluate(ctx).stringValue();
var i = s0.indexOf(s1);
var ret;
if (i == -1) {
ret = '';
} else {
ret = s0.substr(i + s1.length);
}
return new StringValue(ret);
},
'substring': function(ctx) {
// NOTE: XPath defines the position of the first character in a
// string to be 1, in JavaScript this is 0 ([XPATH] Section 4.2).
assert(this.args.length == 2 || this.args.length == 3);
var s0 = this.args[0].evaluate(ctx).stringValue();
var s1 = this.args[1].evaluate(ctx).numberValue();
var ret;
if (this.args.length == 2) {
var i1 = Math.max(0, Math.round(s1) - 1);
ret = s0.substr(i1);
} else {
var s2 = this.args[2].evaluate(ctx).numberValue();
var i0 = Math.round(s1) - 1;
var i1 = Math.max(0, i0);
var i2 = Math.round(s2) - Math.max(0, -i0);
ret = s0.substr(i1, i2);
}
return new StringValue(ret);
},
'string-length': function(ctx) {
var s;
if (this.args.length > 0) {
s = this.args[0].evaluate(ctx).stringValue();
} else {
s = new NodeSetValue([ ctx.node ]).stringValue();
}
return new NumberValue(s.length);
},
'normalize-space': function(ctx) {
var s;
if (this.args.length > 0) {
s = this.args[0].evaluate(ctx).stringValue();
} else {
s = new NodeSetValue([ ctx.node ]).stringValue();
}
s = s.replace(/^\s*/,'').replace(/\s*$/,'').replace(/\s+/g, ' ');
return new StringValue(s);
},
'translate': function(ctx) {
assert(this.args.length == 3);
var s0 = this.args[0].evaluate(ctx).stringValue();
var s1 = this.args[1].evaluate(ctx).stringValue();
var s2 = this.args[2].evaluate(ctx).stringValue();
for (var i = 0; i < s1.length; ++i) {
s0 = s0.replace(new RegExp(s1.charAt(i), 'g'), s2.charAt(i));
}
return new StringValue(s0);
},
'boolean': function(ctx) {
assert(this.args.length == 1);
return new BooleanValue(this.args[0].evaluate(ctx).booleanValue());
},
'not': function(ctx) {
assert(this.args.length == 1);
var ret = !this.args[0].evaluate(ctx).booleanValue();
return new BooleanValue(ret);
},
'true': function(ctx) {
assert(this.args.length == 0);
return new BooleanValue(true);
},
'false': function(ctx) {
assert(this.args.length == 0);
return new BooleanValue(false);
},
'lang': function(ctx) {
assert(this.args.length == 1);
var lang = this.args[0].evaluate(ctx).stringValue();
var xmllang;
var n = ctx.node;
while (n && n != n.parentNode /* just in case ... */) {
xmllang = n.getAttribute('xml:lang');
if (xmllang) {
break;
}
n = n.parentNode;
}
if (!xmllang) {
return new BooleanValue(false);
} else {
var re = new RegExp('^' + lang + '$', 'i');
return new BooleanValue(xmllang.match(re) ||
xmllang.replace(/_.*$/,'').match(re));
}
},
'number': function(ctx) {
assert(this.args.length == 1 || this.args.length == 0);
if (this.args.length == 1) {
return new NumberValue(this.args[0].evaluate(ctx).numberValue());
} else {
return new NumberValue(new NodeSetValue([ ctx.node ]).numberValue());
}
},
'sum': function(ctx) {
assert(this.args.length == 1);
var n = this.args[0].evaluate(ctx).nodeSetValue();
var sum = 0;
for (var i = 0; i < n.length; ++i) {
sum += xmlValue(n[i]) - 0;
}
return new NumberValue(sum);
},
'floor': function(ctx) {
assert(this.args.length == 1);
var num = this.args[0].evaluate(ctx).numberValue();
return new NumberValue(Math.floor(num));
},
'ceiling': function(ctx) {
assert(this.args.length == 1);
var num = this.args[0].evaluate(ctx).numberValue();
return new NumberValue(Math.ceil(num));
},
'round': function(ctx) {
assert(this.args.length == 1);
var num = this.args[0].evaluate(ctx).numberValue();
return new NumberValue(Math.round(num));
},
// TODO(mesch): The following functions are custom. There is a
// standard that defines how to add functions, which should be
// applied here.
'ext-join': function(ctx) {
assert(this.args.length == 2);
var nodes = this.args[0].evaluate(ctx).nodeSetValue();
var delim = this.args[1].evaluate(ctx).stringValue();
var ret = '';
for (var i = 0; i < nodes.length; ++i) {
if (ret) {
ret += delim;
}
ret += xmlValue(nodes[i]);
}
return new StringValue(ret);
},
// ext-if() evaluates and returns its second argument, if the
// boolean value of its first argument is true, otherwise it
// evaluates and returns its third argument.
'ext-if': function(ctx) {
assert(this.args.length == 3);
if (this.args[0].evaluate(ctx).booleanValue()) {
return this.args[1].evaluate(ctx);
} else {
return this.args[2].evaluate(ctx);
}
},
// ext-cardinal() evaluates its single argument as a number, and
// returns the current node that many times. It can be used in the
// select attribute to iterate over an integer range.
'ext-cardinal': function(ctx) {
assert(this.args.length >= 1);
var c = this.args[0].evaluate(ctx).numberValue();
var ret = [];
for (var i = 0; i < c; ++i) {
ret.push(ctx.node);
}
return new NodeSetValue(ret);
}
};
function UnionExpr(expr1, expr2) {
this.expr1 = expr1;
this.expr2 = expr2;
}
UnionExpr.prototype.evaluate = function(ctx) {
var nodes1 = this.expr1.evaluate(ctx).nodeSetValue();
var nodes2 = this.expr2.evaluate(ctx).nodeSetValue();
var I1 = nodes1.length;
for (var i2 = 0; i2 < nodes2.length; ++i2) {
var n = nodes2[i2];
var inBoth = false;
for (var i1 = 0; i1 < I1; ++i1) {
if (nodes1[i1] == n) {
inBoth = true;
i1 = I1; // break inner loop
}
}
if (!inBoth) {
nodes1.push(n);
}
}
return new NodeSetValue(nodes1);
};
function PathExpr(filter, rel) {
this.filter = filter;
this.rel = rel;
}
PathExpr.prototype.evaluate = function(ctx) {
var nodes = this.filter.evaluate(ctx).nodeSetValue();
var nodes1 = [];
for (var i = 0; i < nodes.length; ++i) {
var nodes0 = this.rel.evaluate(ctx.clone(nodes[i], i, nodes)).nodeSetValue();
for (var ii = 0; ii < nodes0.length; ++ii) {
nodes1.push(nodes0[ii]);
}
}
return new NodeSetValue(nodes1);
};
function FilterExpr(expr, predicate) {
this.expr = expr;
this.predicate = predicate;
}
FilterExpr.prototype.evaluate = function(ctx) {
var nodes = this.expr.evaluate(ctx).nodeSetValue();
for (var i = 0; i < this.predicate.length; ++i) {
var nodes0 = nodes;
nodes = [];
for (var j = 0; j < nodes0.length; ++j) {
var n = nodes0[j];
if (this.predicate[i].evaluate(ctx.clone(n, j, nodes0)).booleanValue()) {
nodes.push(n);
}
}
}
return new NodeSetValue(nodes);
}
function UnaryMinusExpr(expr) {
this.expr = expr;
}
UnaryMinusExpr.prototype.evaluate = function(ctx) {
return new NumberValue(-this.expr.evaluate(ctx).numberValue());
};
function BinaryExpr(expr1, op, expr2) {
this.expr1 = expr1;
this.expr2 = expr2;
this.op = op;
}
BinaryExpr.prototype.evaluate = function(ctx) {
var ret;
switch (this.op.value) {
case 'or':
ret = new BooleanValue(this.expr1.evaluate(ctx).booleanValue() ||
this.expr2.evaluate(ctx).booleanValue());
break;
case 'and':
ret = new BooleanValue(this.expr1.evaluate(ctx).booleanValue() &&
this.expr2.evaluate(ctx).booleanValue());
break;
case '+':
ret = new NumberValue(this.expr1.evaluate(ctx).numberValue() +
this.expr2.evaluate(ctx).numberValue());
break;
case '-':
ret = new NumberValue(this.expr1.evaluate(ctx).numberValue() -
this.expr2.evaluate(ctx).numberValue());
break;
case '*':
ret = new NumberValue(this.expr1.evaluate(ctx).numberValue() *
this.expr2.evaluate(ctx).numberValue());
break;
case 'mod':
ret = new NumberValue(this.expr1.evaluate(ctx).numberValue() %
this.expr2.evaluate(ctx).numberValue());
break;
case 'div':
ret = new NumberValue(this.expr1.evaluate(ctx).numberValue() /
this.expr2.evaluate(ctx).numberValue());
break;
case '=':
ret = this.compare(ctx, function(x1, x2) { return x1 == x2; });
break;
case '!=':
ret = this.compare(ctx, function(x1, x2) { return x1 != x2; });
break;
case '<':
ret = this.compare(ctx, function(x1, x2) { return x1 < x2; });
break;
case '<=':
ret = this.compare(ctx, function(x1, x2) { return x1 <= x2; });
break;
case '>':
ret = this.compare(ctx, function(x1, x2) { return x1 > x2; });
break;
case '>=':
ret = this.compare(ctx, function(x1, x2) { return x1 >= x2; });
break;
default:
alert('BinaryExpr.evaluate: ' + this.op.value);
}
return ret;
};
BinaryExpr.prototype.compare = function(ctx, cmp) {
var v1 = this.expr1.evaluate(ctx);
var v2 = this.expr2.evaluate(ctx);
var ret;
if (v1.type == 'node-set' && v2.type == 'node-set') {
var n1 = v1.nodeSetValue();
var n2 = v2.nodeSetValue();
ret = false;
for (var i1 = 0; i1 < n1.length; ++i1) {
for (var i2 = 0; i2 < n2.length; ++i2) {
if (cmp(xmlValue(n1[i1]), xmlValue(n2[i2]))) {
ret = true;
// Break outer loop. Labels confuse the jscompiler and we
// don't use them.
i2 = n2.length;
i1 = n1.length;
}
}
}
} else if (v1.type == 'node-set' || v2.type == 'node-set') {
if (v1.type == 'number') {
var s = v1.numberValue();
var n = v2.nodeSetValue();
ret = false;
for (var i = 0; i < n.length; ++i) {
var nn = xmlValue(n[i]) - 0;
if (cmp(s, nn)) {
ret = true;
break;
}
}
} else if (v2.type == 'number') {
var n = v1.nodeSetValue();
var s = v2.numberValue();
ret = false;
for (var i = 0; i < n.length; ++i) {
var nn = xmlValue(n[i]) - 0;
if (cmp(nn, s)) {
ret = true;
break;
}
}
} else if (v1.type == 'string') {
var s = v1.stringValue();
var n = v2.nodeSetValue();
ret = false;
for (var i = 0; i < n.length; ++i) {
var nn = xmlValue(n[i]);
if (cmp(s, nn)) {
ret = true;
break;
}
}
} else if (v2.type == 'string') {
var n = v1.nodeSetValue();
var s = v2.stringValue();
ret = false;
for (var i = 0; i < n.length; ++i) {
var nn = xmlValue(n[i]);
if (cmp(nn, s)) {
ret = true;
break;
}
}
} else {
ret = cmp(v1.booleanValue(), v2.booleanValue());
}
} else if (v1.type == 'boolean' || v2.type == 'boolean') {
ret = cmp(v1.booleanValue(), v2.booleanValue());
} else if (v1.type == 'number' || v2.type == 'number') {
ret = cmp(v1.numberValue(), v2.numberValue());
} else {
ret = cmp(v1.stringValue(), v2.stringValue());
}
return new BooleanValue(ret);
}
function LiteralExpr(value) {
this.value = value;
}
LiteralExpr.prototype.evaluate = function(ctx) {
return new StringValue(this.value);
};
function NumberExpr(value) {
this.value = value;
}
NumberExpr.prototype.evaluate = function(ctx) {
return new NumberValue(this.value);
};
function VariableExpr(name) {
this.name = name;
}
VariableExpr.prototype.evaluate = function(ctx) {
return ctx.getVariable(this.name);
}
// Factory functions for semantic values (i.e. Expressions) of the
// productions in the grammar. When a production is matched to reduce
// the current parse state stack, the function is called with the
// semantic values of the matched elements as arguments, and returns
// another semantic value. The semantic value is a node of the parse
// tree, an expression object with an evaluate() method that evaluates the
// expression in an actual context. These factory functions are used
// in the specification of the grammar rules, below.
function makeTokenExpr(m) {
return new TokenExpr(m);
}
function passExpr(e) {
return e;
}
function makeLocationExpr1(slash, rel) {
rel.absolute = true;
return rel;
}
function makeLocationExpr2(dslash, rel) {
rel.absolute = true;
rel.prependStep(makeAbbrevStep(dslash.value));
return rel;
}
function makeLocationExpr3(slash) {
var ret = new LocationExpr();
ret.appendStep(makeAbbrevStep('.'));
ret.absolute = true;
return ret;
}
function makeLocationExpr4(dslash) {
var ret = new LocationExpr();
ret.absolute = true;
ret.appendStep(makeAbbrevStep(dslash.value));
return ret;
}
function makeLocationExpr5(step) {
var ret = new LocationExpr();
ret.appendStep(step);
return ret;
}
function makeLocationExpr6(rel, slash, step) {
rel.appendStep(step);
return rel;
}
function makeLocationExpr7(rel, dslash, step) {
rel.appendStep(makeAbbrevStep(dslash.value));
rel.appendStep(step);
return rel;
}
function makeStepExpr1(dot) {
return makeAbbrevStep(dot.value);
}
function makeStepExpr2(ddot) {
return makeAbbrevStep(ddot.value);
}
function makeStepExpr3(axisname, axis, nodetest) {
return new StepExpr(axisname.value, nodetest);
}
function makeStepExpr4(at, nodetest) {
return new StepExpr('attribute', nodetest);
}
function makeStepExpr5(nodetest) {
return new StepExpr('child', nodetest);
}
function makeStepExpr6(step, predicate) {
step.appendPredicate(predicate);
return step;
}
function makeAbbrevStep(abbrev) {
switch (abbrev) {
case '//':
return new StepExpr('descendant-or-self', new NodeTestAny);
case '.':
return new StepExpr('self', new NodeTestAny);
case '..':
return new StepExpr('parent', new NodeTestAny);
}
}
function makeNodeTestExpr1(asterisk) {
return new NodeTestElementOrAttribute;
}
function makeNodeTestExpr2(ncname, colon, asterisk) {
return new NodeTestNC(ncname.value);
}
function makeNodeTestExpr3(qname) {
return new NodeTestName(qname.value);
}
function makeNodeTestExpr4(typeo, parenc) {
var type = typeo.value.replace(/\s*\($/, '');
switch(type) {
case 'node':
return new NodeTestAny;
case 'text':
return new NodeTestText;
case 'comment':
return new NodeTestComment;
case 'processing-instruction':
return new NodeTestPI('');
}
}
function makeNodeTestExpr5(typeo, target, parenc) {
var type = typeo.replace(/\s*\($/, '');
if (type != 'processing-instruction') {
throw type;
}
return new NodeTestPI(target.value);
}
function makePredicateExpr(pareno, expr, parenc) {
return new PredicateExpr(expr);
}
function makePrimaryExpr(pareno, expr, parenc) {
return expr;
}
function makeFunctionCallExpr1(name, pareno, parenc) {
return new FunctionCallExpr(name);
}
function makeFunctionCallExpr2(name, pareno, arg1, args, parenc) {
var ret = new FunctionCallExpr(name);
ret.appendArg(arg1);
for (var i = 0; i < args.length; ++i) {
ret.appendArg(args[i]);
}
return ret;
}
function makeArgumentExpr(comma, expr) {
return expr;
}
function makeUnionExpr(expr1, pipe, expr2) {
return new UnionExpr(expr1, expr2);
}
function makePathExpr1(filter, slash, rel) {
return new PathExpr(filter, rel);
}
function makePathExpr2(filter, dslash, rel) {
rel.prependStep(makeAbbrevStep(dslash.value));
return new PathExpr(filter, rel);
}
function makeFilterExpr(expr, predicates) {
if (predicates.length > 0) {
return new FilterExpr(expr, predicates);
} else {
return expr;
}
}
function makeUnaryMinusExpr(minus, expr) {
return new UnaryMinusExpr(expr);
}
function makeBinaryExpr(expr1, op, expr2) {
return new BinaryExpr(expr1, op, expr2);
}
function makeLiteralExpr(token) {
// remove quotes from the parsed value:
var value = token.value.substring(1, token.value.length - 1);
return new LiteralExpr(value);
}
function makeNumberExpr(token) {
return new NumberExpr(token.value);
}
function makeVariableReference(dollar, name) {
return new VariableExpr(name.value);
}
// Used before parsing for optimization of common simple cases. See
// the begin of xpathParse() for which they are.
function makeSimpleExpr(expr) {
if (expr.charAt(0) == '$') {
return new VariableExpr(expr.substr(1));
} else if (expr.charAt(0) == '@') {
var a = new NodeTestName(expr.substr(1));
var b = new StepExpr('attribute', a);
var c = new LocationExpr();
c.appendStep(b);
return c;
} else if (expr.match(/^[0-9]+$/)) {
return new NumberExpr(expr);
} else {
var a = new NodeTestName(expr);
var b = new StepExpr('child', a);
var c = new LocationExpr();
c.appendStep(b);
return c;
}
}
function makeSimpleExpr2(expr) {
var steps = stringSplit(expr, '/');
var c = new LocationExpr();
for (var i = 0; i < steps.length; ++i) {
var a = new NodeTestName(steps[i]);
var b = new StepExpr('child', a);
c.appendStep(b);
}
return c;
}
// The axes of XPath expressions.
var xpathAxis = {
ANCESTOR_OR_SELF: 'ancestor-or-self',
ANCESTOR: 'ancestor',
ATTRIBUTE: 'attribute',
CHILD: 'child',
DESCENDANT_OR_SELF: 'descendant-or-self',
DESCENDANT: 'descendant',
FOLLOWING_SIBLING: 'following-sibling',
FOLLOWING: 'following',
NAMESPACE: 'namespace',
PARENT: 'parent',
PRECEDING_SIBLING: 'preceding-sibling',
PRECEDING: 'preceding',
SELF: 'self'
};
var xpathAxesRe = [
xpathAxis.ANCESTOR_OR_SELF,
xpathAxis.ANCESTOR,
xpathAxis.ATTRIBUTE,
xpathAxis.CHILD,
xpathAxis.DESCENDANT_OR_SELF,
xpathAxis.DESCENDANT,
xpathAxis.FOLLOWING_SIBLING,
xpathAxis.FOLLOWING,
xpathAxis.NAMESPACE,
xpathAxis.PARENT,
xpathAxis.PRECEDING_SIBLING,
xpathAxis.PRECEDING,
xpathAxis.SELF
].join('|');
// The tokens of the language. The label property is just used for
// generating debug output. The prec property is the precedence used
// for shift/reduce resolution. Default precedence is 0 as a lookahead
// token and 2 on the stack. TODO(mesch): this is certainly not
// necessary and too complicated. Simplify this!
// NOTE: tabular formatting is the big exception, but here it should
// be OK.
var TOK_PIPE = { label: "|", prec: 17, re: new RegExp("^\\|") };
var TOK_DSLASH = { label: "//", prec: 19, re: new RegExp("^//") };
var TOK_SLASH = { label: "/", prec: 30, re: new RegExp("^/") };
var TOK_AXIS = { label: "::", prec: 20, re: new RegExp("^::") };
var TOK_COLON = { label: ":", prec: 1000, re: new RegExp("^:") };
var TOK_AXISNAME = { label: "[axis]", re: new RegExp('^(' + xpathAxesRe + ')') };
var TOK_PARENO = { label: "(", prec: 34, re: new RegExp("^\\(") };
var TOK_PARENC = { label: ")", re: new RegExp("^\\)") };
var TOK_DDOT = { label: "..", prec: 34, re: new RegExp("^\\.\\.") };
var TOK_DOT = { label: ".", prec: 34, re: new RegExp("^\\.") };
var TOK_AT = { label: "@", prec: 34, re: new RegExp("^@") };
var TOK_COMMA = { label: ",", re: new RegExp("^,") };
var TOK_OR = { label: "or", prec: 10, re: new RegExp("^or\\b") };
var TOK_AND = { label: "and", prec: 11, re: new RegExp("^and\\b") };
var TOK_EQ = { label: "=", prec: 12, re: new RegExp("^=") };
var TOK_NEQ = { label: "!=", prec: 12, re: new RegExp("^!=") };
var TOK_GE = { label: ">=", prec: 13, re: new RegExp("^>=") };
var TOK_GT = { label: ">", prec: 13, re: new RegExp("^>") };
var TOK_LE = { label: "<=", prec: 13, re: new RegExp("^<=") };
var TOK_LT = { label: "<", prec: 13, re: new RegExp("^<") };
var TOK_PLUS = { label: "+", prec: 14, re: new RegExp("^\\+"), left: true };
var TOK_MINUS = { label: "-", prec: 14, re: new RegExp("^\\-"), left: true };
var TOK_DIV = { label: "div", prec: 15, re: new RegExp("^div\\b"), left: true };
var TOK_MOD = { label: "mod", prec: 15, re: new RegExp("^mod\\b"), left: true };
var TOK_BRACKO = { label: "[", prec: 32, re: new RegExp("^\\[") };
var TOK_BRACKC = { label: "]", re: new RegExp("^\\]") };
var TOK_DOLLAR = { label: "$", re: new RegExp("^\\$") };
var TOK_NCNAME = { label: "[ncname]", re: new RegExp('^' + XML_NC_NAME) };
var TOK_ASTERISK = { label: "*", prec: 15, re: new RegExp("^\\*"), left: true };
var TOK_LITERALQ = { label: "[litq]", prec: 20, re: new RegExp("^'[^\\']*'") };
var TOK_LITERALQQ = {
label: "[litqq]",
prec: 20,
re: new RegExp('^"[^\\"]*"')
};
var TOK_NUMBER = {
label: "[number]",
prec: 35,
re: new RegExp('^\\d+(\\.\\d*)?') };
var TOK_QNAME = {
label: "[qname]",
re: new RegExp('^(' + XML_NC_NAME + ':)?' + XML_NC_NAME)
};
var TOK_NODEO = {
label: "[nodetest-start]",
re: new RegExp('^(processing-instruction|comment|text|node)\\(')
};
// The table of the tokens of our grammar, used by the lexer: first
// column the tag, second column a regexp to recognize it in the
// input, third column the precedence of the token, fourth column a
// factory function for the semantic value of the token.
//
// NOTE: order of this list is important, because the first match
// counts. Cf. DDOT and DOT, and AXIS and COLON.
var xpathTokenRules = [
TOK_DSLASH,
TOK_SLASH,
TOK_DDOT,
TOK_DOT,
TOK_AXIS,
TOK_COLON,
TOK_AXISNAME,
TOK_NODEO,
TOK_PARENO,
TOK_PARENC,
TOK_BRACKO,
TOK_BRACKC,
TOK_AT,
TOK_COMMA,
TOK_OR,
TOK_AND,
TOK_NEQ,
TOK_EQ,
TOK_GE,
TOK_GT,
TOK_LE,
TOK_LT,
TOK_PLUS,
TOK_MINUS,
TOK_ASTERISK,
TOK_PIPE,
TOK_MOD,
TOK_DIV,
TOK_LITERALQ,
TOK_LITERALQQ,
TOK_NUMBER,
TOK_QNAME,
TOK_NCNAME,
TOK_DOLLAR
];
// All the nonterminals of the grammar. The nonterminal objects are
// identified by object identity; the labels are used in the debug
// output only.
var XPathLocationPath = { label: "LocationPath" };
var XPathRelativeLocationPath = { label: "RelativeLocationPath" };
var XPathAbsoluteLocationPath = { label: "AbsoluteLocationPath" };
var XPathStep = { label: "Step" };
var XPathNodeTest = { label: "NodeTest" };
var XPathPredicate = { label: "Predicate" };
var XPathLiteral = { label: "Literal" };
var XPathExpr = { label: "Expr" };
var XPathPrimaryExpr = { label: "PrimaryExpr" };
var XPathVariableReference = { label: "Variablereference" };
var XPathNumber = { label: "Number" };
var XPathFunctionCall = { label: "FunctionCall" };
var XPathArgumentRemainder = { label: "ArgumentRemainder" };
var XPathPathExpr = { label: "PathExpr" };
var XPathUnionExpr = { label: "UnionExpr" };
var XPathFilterExpr = { label: "FilterExpr" };
var XPathDigits = { label: "Digits" };
var xpathNonTerminals = [
XPathLocationPath,
XPathRelativeLocationPath,
XPathAbsoluteLocationPath,
XPathStep,
XPathNodeTest,
XPathPredicate,
XPathLiteral,
XPathExpr,
XPathPrimaryExpr,
XPathVariableReference,
XPathNumber,
XPathFunctionCall,
XPathArgumentRemainder,
XPathPathExpr,
XPathUnionExpr,
XPathFilterExpr,
XPathDigits
];
// Quantifiers that are used in the productions of the grammar.
var Q_01 = { label: "?" };
var Q_MM = { label: "*" };
var Q_1M = { label: "+" };
// Tag for left associativity (right assoc is implied by undefined).
var ASSOC_LEFT = true;
// The productions of the grammar. Columns of the table:
//
// - target nonterminal,
// - pattern,
// - precedence,
// - semantic value factory
//
// The semantic value factory is a function that receives parse tree
// nodes from the stack frames of the matched symbols as arguments and
// returns an a node of the parse tree. The node is stored in the top
// stack frame along with the target object of the rule. The node in
// the parse tree is an expression object that has an evaluate() method
// and thus evaluates XPath expressions.
//
// The precedence is used to decide between reducing and shifting by
// comparing the precendence of the rule that is candidate for
// reducing with the precedence of the look ahead token. Precedence of
// -1 means that the precedence of the tokens in the pattern is used
// instead. TODO: It shouldn't be necessary to explicitly assign
// precedences to rules.
// DGF As it stands, these precedences are purely empirical; we're
// not sure they can be made to be consistent at all.
var xpathGrammarRules =
[
[ XPathLocationPath, [ XPathRelativeLocationPath ], 18,
passExpr ],
[ XPathLocationPath, [ XPathAbsoluteLocationPath ], 18,
passExpr ],
[ XPathAbsoluteLocationPath, [ TOK_SLASH, XPathRelativeLocationPath ], 18,
makeLocationExpr1 ],
[ XPathAbsoluteLocationPath, [ TOK_DSLASH, XPathRelativeLocationPath ], 18,
makeLocationExpr2 ],
[ XPathAbsoluteLocationPath, [ TOK_SLASH ], 0,
makeLocationExpr3 ],
[ XPathAbsoluteLocationPath, [ TOK_DSLASH ], 0,
makeLocationExpr4 ],
[ XPathRelativeLocationPath, [ XPathStep ], 31,
makeLocationExpr5 ],
[ XPathRelativeLocationPath,
[ XPathRelativeLocationPath, TOK_SLASH, XPathStep ], 31,
makeLocationExpr6 ],
[ XPathRelativeLocationPath,
[ XPathRelativeLocationPath, TOK_DSLASH, XPathStep ], 31,
makeLocationExpr7 ],
[ XPathStep, [ TOK_DOT ], 33,
makeStepExpr1 ],
[ XPathStep, [ TOK_DDOT ], 33,
makeStepExpr2 ],
[ XPathStep,
[ TOK_AXISNAME, TOK_AXIS, XPathNodeTest ], 33,
makeStepExpr3 ],
[ XPathStep, [ TOK_AT, XPathNodeTest ], 33,
makeStepExpr4 ],
[ XPathStep, [ XPathNodeTest ], 33,
makeStepExpr5 ],
[ XPathStep, [ XPathStep, XPathPredicate ], 33,
makeStepExpr6 ],
[ XPathNodeTest, [ TOK_ASTERISK ], 33,
makeNodeTestExpr1 ],
[ XPathNodeTest, [ TOK_NCNAME, TOK_COLON, TOK_ASTERISK ], 33,
makeNodeTestExpr2 ],
[ XPathNodeTest, [ TOK_QNAME ], 33,
makeNodeTestExpr3 ],
[ XPathNodeTest, [ TOK_NODEO, TOK_PARENC ], 33,
makeNodeTestExpr4 ],
[ XPathNodeTest, [ TOK_NODEO, XPathLiteral, TOK_PARENC ], 33,
makeNodeTestExpr5 ],
[ XPathPredicate, [ TOK_BRACKO, XPathExpr, TOK_BRACKC ], 33,
makePredicateExpr ],
[ XPathPrimaryExpr, [ XPathVariableReference ], 33,
passExpr ],
[ XPathPrimaryExpr, [ TOK_PARENO, XPathExpr, TOK_PARENC ], 33,
makePrimaryExpr ],
[ XPathPrimaryExpr, [ XPathLiteral ], 30,
passExpr ],
[ XPathPrimaryExpr, [ XPathNumber ], 30,
passExpr ],
[ XPathPrimaryExpr, [ XPathFunctionCall ], 31,
passExpr ],
[ XPathFunctionCall, [ TOK_QNAME, TOK_PARENO, TOK_PARENC ], -1,
makeFunctionCallExpr1 ],
[ XPathFunctionCall,
[ TOK_QNAME, TOK_PARENO, XPathExpr, XPathArgumentRemainder, Q_MM,
TOK_PARENC ], -1,
makeFunctionCallExpr2 ],
[ XPathArgumentRemainder, [ TOK_COMMA, XPathExpr ], -1,
makeArgumentExpr ],
[ XPathUnionExpr, [ XPathPathExpr ], 20,
passExpr ],
[ XPathUnionExpr, [ XPathUnionExpr, TOK_PIPE, XPathPathExpr ], 20,
makeUnionExpr ],
[ XPathPathExpr, [ XPathLocationPath ], 20,
passExpr ],
[ XPathPathExpr, [ XPathFilterExpr ], 19,
passExpr ],
[ XPathPathExpr,
[ XPathFilterExpr, TOK_SLASH, XPathRelativeLocationPath ], 19,
makePathExpr1 ],
[ XPathPathExpr,
[ XPathFilterExpr, TOK_DSLASH, XPathRelativeLocationPath ], 19,
makePathExpr2 ],
[ XPathFilterExpr, [ XPathPrimaryExpr, XPathPredicate, Q_MM ], 31,
makeFilterExpr ],
[ XPathExpr, [ XPathPrimaryExpr ], 16,
passExpr ],
[ XPathExpr, [ XPathUnionExpr ], 16,
passExpr ],
[ XPathExpr, [ TOK_MINUS, XPathExpr ], -1,
makeUnaryMinusExpr ],
[ XPathExpr, [ XPathExpr, TOK_OR, XPathExpr ], -1,
makeBinaryExpr ],
[ XPathExpr, [ XPathExpr, TOK_AND, XPathExpr ], -1,
makeBinaryExpr ],
[ XPathExpr, [ XPathExpr, TOK_EQ, XPathExpr ], -1,
makeBinaryExpr ],
[ XPathExpr, [ XPathExpr, TOK_NEQ, XPathExpr ], -1,
makeBinaryExpr ],
[ XPathExpr, [ XPathExpr, TOK_LT, XPathExpr ], -1,
makeBinaryExpr ],
[ XPathExpr, [ XPathExpr, TOK_LE, XPathExpr ], -1,
makeBinaryExpr ],
[ XPathExpr, [ XPathExpr, TOK_GT, XPathExpr ], -1,
makeBinaryExpr ],
[ XPathExpr, [ XPathExpr, TOK_GE, XPathExpr ], -1,
makeBinaryExpr ],
[ XPathExpr, [ XPathExpr, TOK_PLUS, XPathExpr ], -1,
makeBinaryExpr, ASSOC_LEFT ],
[ XPathExpr, [ XPathExpr, TOK_MINUS, XPathExpr ], -1,
makeBinaryExpr, ASSOC_LEFT ],
[ XPathExpr, [ XPathExpr, TOK_ASTERISK, XPathExpr ], -1,
makeBinaryExpr, ASSOC_LEFT ],
[ XPathExpr, [ XPathExpr, TOK_DIV, XPathExpr ], -1,
makeBinaryExpr, ASSOC_LEFT ],
[ XPathExpr, [ XPathExpr, TOK_MOD, XPathExpr ], -1,
makeBinaryExpr, ASSOC_LEFT ],
[ XPathLiteral, [ TOK_LITERALQ ], -1,
makeLiteralExpr ],
[ XPathLiteral, [ TOK_LITERALQQ ], -1,
makeLiteralExpr ],
[ XPathNumber, [ TOK_NUMBER ], -1,
makeNumberExpr ],
[ XPathVariableReference, [ TOK_DOLLAR, TOK_QNAME ], 200,
makeVariableReference ]
];
// That function computes some optimizations of the above data
// structures and will be called right here. It merely takes the
// counter variables out of the global scope.
var xpathRules = [];
function xpathParseInit() {
if (xpathRules.length) {
return;
}
// Some simple optimizations for the xpath expression parser: sort
// grammar rules descending by length, so that the longest match is
// first found.
xpathGrammarRules.sort(function(a,b) {
var la = a[1].length;
var lb = b[1].length;
if (la < lb) {
return 1;
} else if (la > lb) {
return -1;
} else {
return 0;
}
});
var k = 1;
for (var i = 0; i < xpathNonTerminals.length; ++i) {
xpathNonTerminals[i].key = k++;
}
for (i = 0; i < xpathTokenRules.length; ++i) {
xpathTokenRules[i].key = k++;
}
xpathLog('XPath parse INIT: ' + k + ' rules');
// Another slight optimization: sort the rules into bins according
// to the last element (observing quantifiers), so we can restrict
// the match against the stack to the subest of rules that match the
// top of the stack.
//
// TODO(mesch): What we actually want is to compute states as in
// bison, so that we don't have to do any explicit and iterated
// match against the stack.
function push_(array, position, element) {
if (!array[position]) {
array[position] = [];
}
array[position].push(element);
}
for (i = 0; i < xpathGrammarRules.length; ++i) {
var rule = xpathGrammarRules[i];
var pattern = rule[1];
for (var j = pattern.length - 1; j >= 0; --j) {
if (pattern[j] == Q_1M) {
push_(xpathRules, pattern[j-1].key, rule);
break;
} else if (pattern[j] == Q_MM || pattern[j] == Q_01) {
push_(xpathRules, pattern[j-1].key, rule);
--j;
} else {
push_(xpathRules, pattern[j].key, rule);
break;
}
}
}
xpathLog('XPath parse INIT: ' + xpathRules.length + ' rule bins');
var sum = 0;
mapExec(xpathRules, function(i) {
if (i) {
sum += i.length;
}
});
xpathLog('XPath parse INIT: ' + (sum / xpathRules.length) +
' average bin size');
}
// Local utility functions that are used by the lexer or parser.
function xpathCollectDescendants(nodelist, node, opt_tagName) {
if (opt_tagName && node.getElementsByTagName) {
copyArray(nodelist, node.getElementsByTagName(opt_tagName));
return;
}
for (var n = node.firstChild; n; n = n.nextSibling) {
nodelist.push(n);
xpathCollectDescendants(nodelist, n);
}
}
// DGF extract a tag name suitable for getElementsByTagName
function xpathExtractTagNameFromNodeTest(nodetest) {
if (nodetest instanceof NodeTestName) {
return nodetest.name;
} else if (nodetest instanceof NodeTestAny || nodetest instanceof NodeTestElementOrAttribute) {
return "*";
}
}
function xpathCollectDescendantsReverse(nodelist, node) {
for (var n = node.lastChild; n; n = n.previousSibling) {
nodelist.push(n);
xpathCollectDescendantsReverse(nodelist, n);
}
}
// The entry point for the library: match an expression against a DOM
// node. Returns an XPath value.
function xpathDomEval(expr, node) {
var expr1 = xpathParse(expr);
var ret = expr1.evaluate(new ExprContext(node));
return ret;
}
// Utility function to sort a list of nodes. Used by xsltSort() and
// nxslSelect().
function xpathSort(input, sort) {
if (sort.length == 0) {
return;
}
var sortlist = [];
for (var i = 0; i < input.contextSize(); ++i) {
var node = input.nodelist[i];
var sortitem = { node: node, key: [] };
var context = input.clone(node, 0, [ node ]);
for (var j = 0; j < sort.length; ++j) {
var s = sort[j];
var value = s.expr.evaluate(context);
var evalue;
if (s.type == 'text') {
evalue = value.stringValue();
} else if (s.type == 'number') {
evalue = value.numberValue();
}
sortitem.key.push({ value: evalue, order: s.order });
}
// Make the sort stable by adding a lowest priority sort by
// id. This is very convenient and furthermore required by the
// spec ([XSLT] - Section 10 Sorting).
sortitem.key.push({ value: i, order: 'ascending' });
sortlist.push(sortitem);
}
sortlist.sort(xpathSortByKey);
var nodes = [];
for (var i = 0; i < sortlist.length; ++i) {
nodes.push(sortlist[i].node);
}
input.nodelist = nodes;
input.setNode(0);
}
// Sorts by all order criteria defined. According to the JavaScript
// spec ([ECMA] Section 11.8.5), the compare operators compare strings
// as strings and numbers as numbers.
//
// NOTE: In browsers which do not follow the spec, this breaks only in
// the case that numbers should be sorted as strings, which is very
// uncommon.
function xpathSortByKey(v1, v2) {
// NOTE: Sort key vectors of different length never occur in
// xsltSort.
for (var i = 0; i < v1.key.length; ++i) {
var o = v1.key[i].order == 'descending' ? -1 : 1;
if (v1.key[i].value > v2.key[i].value) {
return +1 * o;
} else if (v1.key[i].value < v2.key[i].value) {
return -1 * o;
}
}
return 0;
}
// Parses and then evaluates the given XPath expression in the given
// input context. Notice that parsed xpath expressions are cached.
function xpathEval(select, context) {
var expr = xpathParse(select);
var ret = expr.evaluate(context);
return ret;
}