Subversion-Projekte lars-tiefland.prado

Revision

Blame | Letzte Änderung | Log anzeigen | RSS feed

// 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.
//
//
// TODO(mesch): add jsdoc comments. Use more coherent naming.
//
//
// 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) {
  if (xpathdebug) {
    Log.write('XPath parse ' + expr);
  }
  xpathParseInit();

  var cached = xpathCacheLookup(expr);
  if (cached) {
    if (xpathdebug) {
      Log.write(' ... 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;
    if (xpathdebug) {
      Log.write(' ... simple');
    }
    return ret;
  }

  if (expr.match(/^\w+(\/\w+)*$/i)) {
    var ret = makeSimpleExpr2(expr);
    xpathParseCache[expr] = ret;
    if (xpathdebug) {
      Log.write(' ... simple 2');
    }
    return ret;
  }

  var cachekey = expr; // expr is modified during parse
  if (xpathdebug) {
    Timer.start('XPath parse', cachekey);
  }

  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);
      if (xpathdebug) {
        Log.write('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 {
      if (xpathdebug) {
        Log.write('DONE');
      }
      done = true;
    }

    while (xpathReduce(stack, ahead)) {
      reduce_count++;
      if (xpathdebug) {
        Log.write('stack: ' + stackToString(stack));
      }
    }
  }

  if (xpathdebug) {
    Log.write(stackToString(stack));
  }

  if (stack.length != 1) {
    throw 'XPath parse error ' + cachekey + ':\n' + stackToString(stack);
  }

  var result = stack[0].expr;
  xpathParseCache[cachekey] = result;

  if (xpathdebug) {
    Timer.end('XPath parse', cachekey);
  }

  if (xpathdebug) {
    Log.write('XPath parse: ' + parse_count + ' / ' + 
              lexer_count + ' / ' + reduce_count);
  }

  return result;
}

var xpathParseCache = {};

function xpathCacheLookup(expr) {
  return xpathParseCache[expr];
}

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();
    }

    if (xpathdebug) {
      Log.write('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; });
    cand.expr = cand.rule[3].apply(null, matchexpr);

    stack.push(cand);
    ret = true;

  } else {
    if (ahead) {
      if (xpathdebug) {
        Log.write('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(node, position) -- sets the context to the new node and
//   its corresponding position. Needed to implement scoping rules for
//   variables in XPath. (A variable is visible to all subsequent
//   siblings, not only to its children.)

function ExprContext(node, position, nodelist, parent) {
  this.node = node;
  this.position = position || 0;
  this.nodelist = nodelist || [ node ];
  this.variables = {};
  this.parent = parent || null;
  this.root = parent ? parent.root : node.ownerDocument;
}

ExprContext.prototype.clone = function(node, position, nodelist) {
  return new
  ExprContext(node || this.node,
              typeof position != 'undefined' ? position : this.position,
              nodelist || this.nodelist, this);
};

ExprContext.prototype.setVariable = function(name, value) {
  this.variables[name] = 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(node, position) {
  this.node = node;
  this.position = position;
}


// 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 + ' ' + Error().stack;
}

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 + ' ' + Error().stack;
}

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 + ' ' + Error().stack;
}

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) {
  this.steps.push(s);
}

LocationExpr.prototype.prependStep = function(s) {
  var steps0 = this.steps;
  this.steps = [ s ];
  for (var i = 0; i < steps0.length; ++i) {
    this.steps.push(steps0[i]);
  }
};

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, predicate) {
  this.axis = axis;
  this.nodetest = nodetest;
  this.predicate = predicate || [];
}

StepExpr.prototype.appendPredicate = function(p) {
  this.predicate.push(p);
}

StepExpr.prototype.evaluate = function(ctx) {
  var input = ctx.node;
  var nodelist = [];

  // 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 = input.parentNode) {
      nodelist.push(n);
    }

  } else if (this.axis == xpathAxis.ANCESTOR) {
    for (var n = input.parentNode; n; n = input.parentNode) {
      nodelist.push(n);
    }

  } else if (this.axis == xpathAxis.ATTRIBUTE) {
    copyArray(nodelist, input.attributes);

  } else if (this.axis == xpathAxis.CHILD) {
    copyArray(nodelist, input.childNodes);

  } else if (this.axis == xpathAxis.DESCENDANT_OR_SELF) {
    nodelist.push(input);
    xpathCollectDescendants(nodelist, input);

  } else if (this.axis == xpathAxis.DESCENDANT) {
    xpathCollectDescendants(nodelist, input);

  } else if (this.axis == xpathAxis.FOLLOWING) {
    for (var n = input.parentNode; 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 = input.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.parentNode; 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 = input.previousSibling) {
      nodelist.push(n);
    }

  } else if (this.axis == xpathAxis.SELF) {
    nodelist.push(input);

  } else {
    throw 'ERROR -- NO SUCH AXIS: ' + this.axis;
  }

  // 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 NodeTestElement() {}

NodeTestElement.prototype.evaluate = function(ctx) {
  return new BooleanValue(ctx.node.nodeType == DOM_ELEMENT_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;
}

NodeTestName.prototype.evaluate = function(ctx) {
  var n = ctx.node;
  // NOTE (Patrick Lightbody): this change allows node selection to be case-insensitive
  return new BooleanValue(n.nodeName.toUpperCase() == this.name.toUpperCase());
}

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 {
    Log.write('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.nodelist.length);
  },

  '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.evaluate(ctx);
    var ret = [];
    var ids;
    if (e.type == 'node-set') {
      ids = [];
      for (var i = 0; i < e.length; ++i) {
        var v = xmlValue(e[i]).split(/\s+/);
        for (var ii = 0; ii < v.length; ++ii) {
          ids.push(v[ii]);
        }
      }
    } else {
      ids = e.split(/\s+/);
    }
    var d = ctx.node.ownerDocument;
    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-sprintf': function(ctx) {
    assert(this.args.length >= 1);
    var args = [];
    for (var i = 0; i < this.args.length; ++i) {
      args.push(this.args[i].evaluate(ctx).stringValue());
    }
    return new StringValue(sprintf.apply(null, args));
  },

  // 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) {
    for (var i1 = 0; i1 < I1; ++i1) {
      if (nodes1[i1] == nodes2[i2]) {
        // break inner loop and continue outer loop, labels confuse
        // the js compiler, so we don't use them here.
        i1 = I1;
      }
    }
    nodes1.push(nodes2[i2]);
  }
  return new NodeSetValue(nodes2);
};

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));
  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 NodeTestElement;
}

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 + ' ' + Error().stack;
  }
  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 = expr.split('/');
  var c = new LocationExpr();
  for (var i in steps) {
    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('^[a-z][-\\w]*','i') };

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('^([a-z][-\\w]*:)?[a-z][-\\w]*','i')
};

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.

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 ], 30,
     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 ], 20,
     makePathExpr1 ],
   [ XPathPathExpr,
     [ XPathFilterExpr, TOK_DSLASH, XPathRelativeLocationPath ], 20,
     makePathExpr2 ],

   [ XPathFilterExpr, [ XPathPrimaryExpr, XPathPredicate, Q_MM ], 20,
     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++;
  }

  Log.write('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;
      }
    }
  }

  Log.write('XPath parse INIT: ' + xpathRules.length + ' rule bins');
  
  var sum = 0;
  mapExec(xpathRules, function(i) {
    if (i) {
      sum += i.length;
    }
  });
  
  Log.write('XPath parse INIT: ' + (sum / xpathRules.length) + ' average bin size');
}

// Local utility functions that are used by the lexer or parser.

function xpathCollectDescendants(nodelist, node) {
  for (var n = node.firstChild; n; n = n.nextSibling) {
    nodelist.push(n);
    arguments.callee(nodelist, n);
  }
}

function xpathCollectDescendantsReverse(nodelist, node) {
  for (var n = node.lastChild; n; n = n.previousSibling) {
    nodelist.push(n);
    arguments.callee(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.nodelist.length; ++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(nodes[0], 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;
}