dojo.provide("dojo._base.query");
dojo.require("dojo._base.NodeList");

/*
	dojo.query() architectural overview:

		dojo.query is a relatively full-featured CSS3 query library. It is
		designed to take any valid CSS3 selector and return the nodes matching
		the selector. To do this quickly, it processes queries in several
		steps, applying caching where profitable.
		
		The steps (roughly in reverse order of the way they appear in the code):
			1.) check to see if we already have a "query dispatcher"
				- if so, use that with the given parameterization. Skip to step 4.
			2.) attempt to determine which branch to dispatch the query to:
				- JS (optimized DOM iteration)
				- xpath (for browsers that support it and where it's fast)
				- native (not available in any browser yet)
			3.) tokenize and convert to executable "query dispatcher"
				- this is where the lion's share of the complexity in the
				  system lies. In the DOM version, the query dispatcher is
				  assembled as a chain of "yes/no" test functions pertaining to
				  a section of a simple query statement (".blah:nth-child(odd)"
				  but not "div div", which is 2 simple statements). Individual
				  statement dispatchers are cached (to prevent re-definition)
				  as are entire dispatch chains (to make re-execution of the
				  same query fast)
				- in the xpath path, tokenization yeilds a concatenation of
				  parameterized xpath selectors. As with the DOM version, both
				  simple selector blocks and overall evaluators are cached to
				  prevent re-defintion
			4.) the resulting query dispatcher is called in the passed scope (by default the top-level document)
				- for DOM queries, this results in a recursive, top-down
				  evaluation of nodes based on each simple query section
				- xpath queries can, thankfully, be executed in one shot
			5.) matched nodes are pruned to ensure they are unique
*/

;(function(){
	// define everything in a closure for compressability reasons. "d" is an
	// alias to "dojo" since it's so frequently used. This seems a
	// transformation that the build system could perform on a per-file basis.
	var d = dojo;

	////////////////////////////////////////////////////////////////////////
	// Utility code
	////////////////////////////////////////////////////////////////////////

	// for tokenizing simple query sections. The results are often reused
	var _getIndexes = function(q){
		return [ q.indexOf("#"), q.indexOf("."), q.indexOf("["), q.indexOf(":") ];
	}

	var _lowestFromIndex = function(query, index){
		// spammy and verbose, but fast
		var ql = query.length;
		var i = _getIndexes(query);
		var end = ql;
		for(var x=index; x<i.length; x++){
			if(i[x] >= 0){
				if(i[x] < end){
					end = i[x];
				}
			}
		}
		return (end < 0) ? ql : end;
	}

	// returns the "#thinger" section of a longer simple query like "#thinger.blah[selected]"
	var getId = function(query){
		var i = _getIndexes(query);
		if(i[0] != -1){
			return query.substring(i[0]+1, _lowestFromIndex(query, 1));
		}else{
			return "";
		}
	}

	// returns only the tag name component of a simple query
	var getTagName = function(query){
		var tagNameEnd;
		var i = _getIndexes(query);
		if((i[0] == 0)||(i[1] == 0)){
			// hash or dot at the front, no tagname
			tagNameEnd = 0;
		}else{
			tagNameEnd =  _lowestFromIndex(query, 0);
		}
		// FIXME: should this be ">=" to account for tags like <a> ?
		return ((tagNameEnd > 0) ? query.substr(0, tagNameEnd).toLowerCase() : "*");
	}

	// like Math.min
	var smallest = function(arr){
		var ret = -1;
		for(var x=0; x<arr.length; x++){
			var ta = arr[x];
			if(ta >= 0){
				if((ta > ret)||(ret == -1)){
					ret = ta;
				}
			}
		}
		return ret;
	}

	// returns only the class name component (".whatever") of a simple query
	var getClassName = function(query){
		// [ "#", ".", "[", ":" ];
		var i = _getIndexes(query);
		if(-1 == i[1]){ return ""; } // no class component
		var di = i[1]+1;

		var othersStart = smallest(i.slice(2));
		if(di < othersStart){
			return query.substring(di, othersStart);
		}else if(-1 == othersStart){
			return query.substr(di);
		}else{
			return "";
		}
	}

	/*
	var isTagOnly = function(query){
		return (Math.max.apply(this, _getIndexes(query)) == -1);
	}
	*/

	////////////////////////////////////////////////////////////////////////
	// XPath query code
	////////////////////////////////////////////////////////////////////////

	// this array is a lookup used to generate an attribute matching function.
	// There is a similar lookup/generator list for the DOM branch with similar
	// calling semantics.
	var xPathAttrs = [
		// FIXME: need to re-order in order of likelyness to be used in matches
		{
			key: "|=",
			match: function(attr, value){
				return "[contains(concat(' ',@"+attr+",' '), ' "+ value +"-')]";
			}
		},
		{
			key: "~=",
			match: function(attr, value){
				return "[contains(concat(' ',@"+attr+",' '), ' "+ value +" ')]";
			}
		},
		{
			key: "^=",
			match: function(attr, value){
				return "[starts-with(@"+attr+", '"+ value +"')]";
			}
		},
		{
			key: "*=",
			match: function(attr, value){
				return "[contains(@"+attr+", '"+ value +"')]";
			}
		},
		{
			key: "$=",
			match: function(attr, value){
				return "[substring(@"+attr+", string-length(@"+attr+")-"+(value.length-1)+")='"+value+"']";
			}
		},
		{
			key: "!=",
			match: function(attr, value){
				return "[not(@"+attr+"='"+ value +"')]";
			}
		},
		// NOTE: the "=" match MUST come last!
		{
			key: "=",
			match: function(attr, value){
				return "[@"+attr+"='"+ value +"']";
			}
		}
	];

	// takes a list of attribute searches, the overall query, a function to
	// generate a default matcher, and a closure-bound method for providing a
	// matching function that generates whatever type of yes/no distinguisher
	// the query method needs. The method is a bit tortured and hard to read
	// because it needs to be used in both the XPath and DOM branches.
	var handleAttrs = function(	attrList, 
								query, 
								getDefault, 
								handleMatch){
		var matcher;
		var i = _getIndexes(query);
		if(i[2] >= 0){
			var lBktIdx = query.indexOf("]", i[2]);
			var condition = query.substring(i[2]+1, lBktIdx);
			while(condition && condition.length){
				if(condition.charAt(0) == "@"){
					condition = condition.slice(1);
				}

				matcher = null;
				// http://www.w3.org/TR/css3-selectors/#attribute-selectors
				for(var x=0; x<attrList.length; x++){
					var ta = attrList[x];
					var tci = condition.indexOf(ta.key);
					if(tci >= 0){
						var attr = condition.substring(0, tci);
						var value = condition.substring(tci+ta.key.length);
						if(	(value.charAt(0) == "\"")||
							(value.charAt(0) == "\'")){
							value = value.substring(1, value.length-1);
						}
						matcher = ta.match(d.trim(attr), d.trim(value));
						break;
					}
				}
				if((!matcher)&&(condition.length)){
					matcher = getDefault(condition);
				}
				if(matcher){
					handleMatch(matcher);
					// ff = agree(ff, matcher);
				}

				condition = null;
				var nbktIdx = query.indexOf("[", lBktIdx);
				if(0 <= nbktIdx){
					lBktIdx = query.indexOf("]", nbktIdx);
					if(0 <= lBktIdx){
						condition = query.substring(nbktIdx+1, lBktIdx);
					}
				}
			}
		}
	}

	var buildPath = function(query){
		var xpath = ".";
		var qparts = query.split(" "); // FIXME: this break on span[thinger = foo]
		while(qparts.length){
			var tqp = qparts.shift();
			var prefix;
			// FIXME: need to add support for ~ and +
			if(tqp == ">"){
				prefix = "/";
				// prefix = "/child::node()";
				tqp = qparts.shift();
			}else{
				prefix = "//";
				// prefix = "/descendant::node()"
			}

			// get the tag name (if any)
			var tagName = getTagName(tqp);

			xpath += prefix + tagName;
			
			// check to see if it's got an id. Needs to come first in xpath.
			var id = getId(tqp);
			if(id.length){
				xpath += "[@id='"+id+"'][1]";
			}

			var cn = getClassName(tqp);
			// check the class name component
			if(cn.length){
				var padding = " ";
				if(cn.charAt(cn.length-1) == "*"){
					padding = ""; cn = cn.substr(0, cn.length-1);
				}
				xpath += 
					"[contains(concat(' ',@class,' '), ' "+
					cn + padding + "')]";
			}

			handleAttrs(xPathAttrs, tqp, 
				function(condition){
						return "[@"+condition+"]";
				},
				function(matcher){
					xpath += matcher;
				}
			);

			// FIXME: need to implement pseudo-class checks!!
		};
		return xpath;
	};

	var _xpathFuncCache = {};
	var getXPathFunc = function(path){
		if(_xpathFuncCache[path]){
			return _xpathFuncCache[path];
		}

		var doc = d.doc;
		// var parent = d.body(); // FIXME
		// FIXME: don't need to memoize. The closure scope handles it for us.
		var xpath = buildPath(path);
		// console.debug(xpath);

		var tf = function(parent){
			// XPath query strings are memoized.
			var ret = [];
			var xpathResult;
			try{
				xpathResult = doc.evaluate(xpath, parent, null, 
												// XPathResult.UNORDERED_NODE_ITERATOR_TYPE, null);
												XPathResult.ANY_TYPE, null);
			}catch(e){
				console.debug("failure in exprssion:", xpath, "under:", parent);
				console.debug(e);
			}
			var result = xpathResult.iterateNext();
			while(result){
				ret.push(result);
				result = xpathResult.iterateNext();
			}
			return ret;
		}
		return _xpathFuncCache[path] = tf;
	};

	/*
	d.xPathMatch = function(query){
		// XPath based DOM query system. Handles a small subset of CSS
		// selectors, subset is identical to the non-XPath version of this
		// function. 

		// FIXME: need to add support for alternate roots
		return getXPathFunc(query)();
	}
	*/

	////////////////////////////////////////////////////////////////////////
	// DOM query code
	////////////////////////////////////////////////////////////////////////

	var _filtersCache = {};
	var _simpleFiltersCache = {};

	// the basic building block of the yes/no chaining system. agree(f1, f2)
	// generates a new function which returns the boolean results of both of
	// the passed functions to a single logical-anded result.
	var agree = function(first, second){
		if(!first){
			return second;
		}
		if(!second){
			return first;
		}

		return function(){
			return first.apply(window, arguments) && second.apply(window, arguments);
		}
	}

	var _filterDown = function(element, queryParts, matchArr, idx){
		var nidx = idx+1;
		var isFinal = (queryParts.length == nidx);
		var tqp = queryParts[idx];

		// see if we can constrain our next level to direct children
		if(tqp == ">"){
			var ecn = element.childNodes;
			if(!ecn.length){
				return;
			}
			nidx++;
			isFinal = (queryParts.length == nidx);
			// kinda janky, too much array alloc
			var tf = getFilterFunc(queryParts[idx+1]);
			// for(var x=ecn.length-1, te; x>=0, te=ecn[x]; x--){
			for(var x=0, te; x<ecn.length, te=ecn[x]; x++){
				if(tf(te)){
					if(isFinal){
						matchArr.push(te);
					}else{
						_filterDown(te, queryParts, matchArr, nidx);
					}
				}
				/*
				if(x==0){
					break;
				}
				*/
			}
		}

		// otherwise, keep going down, unless we'er at the end
		var candidates = getElementsFunc(tqp)(element);
		if(isFinal){
			while(candidates.length){
				matchArr.push(candidates.shift());
			}
			/*
			candidates.unshift(0, matchArr.length-1);
			matchArr.splice.apply(matchArr, candidates);
			*/
		}else{
			// if we're not yet at the bottom, keep going!
			while(candidates.length){
				_filterDown(candidates.shift(), queryParts, matchArr, nidx);
			}
		}
	}

	var filterDown = function(elements, queryParts){
		var ret = [];

		// for every root, get the elements that match the descendant selector
		// for(var x=elements.length-1, te; x>=0, te=elements[x]; x--){
		var x = elements.length - 1, te;
		while(te = elements[x--]){
			_filterDown(te, queryParts, ret, 0);
		}
		return ret;
	}

	var getFilterFunc = function(query){
		// note: query can't have spaces!
		if(_filtersCache[query]){
			return _filtersCache[query];
		}
		var ff = null;
		var tagName = getTagName(query);

		// does it have a tagName component?
		if(tagName != "*"){
			// tag name match
			ff = agree(ff, 
				function(elem){
					return (
						(elem.nodeType == 1) &&
						(tagName == elem.tagName.toLowerCase())
					);
					// return isTn;
				}
			);
		}

		var idComponent = getId(query);

		// does the node have an ID?
		if(idComponent.length){
			ff = agree(ff, 
				function(elem){
					return (
						(elem.nodeType == 1) &&
						(elem.id == idComponent)
					);
				}
			);
		}

		if(	Math.max.apply(this, _getIndexes(query).slice(1)) >= 0){
			// if we have other query param parts, make sure we add them to the
			// filter chain
			ff = agree(ff, getSimpleFilterFunc(query));
		}

		return _filtersCache[query] = ff;
	}

	var getNodeIndex = function(node){
		// NOTE: we could have a more accurate caching mechanism by
		// invalidating caches after the query has finished, but I think that'd
		// lead to significantly more cache churn than the cache would provide
		// value for in the common case. Generally, we're more conservative
		// (and therefore, more accurate) than jQuery and DomQuery WRT node
		// node indexes, but there may be corner cases in which we fall down.
		// How much we care about them is TBD.

		var pn = node.parentNode;
		var pnc = pn.childNodes;

		// check to see if we can trust the cache. If not, re-key the whole
		// thing and return our node match from that.

		var nidx = -1;
		var child = pn.firstChild;
		if(!child){
			return nidx;
		}

		var ci = node["__cachedIndex"];
		var cl = pn["__cachedLength"];

		// only handle cache building if we've gone out of sync
		if(((typeof cl == "number")&&(cl != pnc.length))||(typeof ci != "number")){
			// rip though the whole set, building cache indexes as we go
			pn["__cachedLength"] = pnc.length;
			var idx = 1;
			do{
				// we only assign indexes for nodes with nodeType == 1, as per:
				//		http://www.w3.org/TR/css3-selectors/#nth-child-pseudo
				// only elements are counted in the search order, and they
				// begin at 1 for the first child's index

				if(child === node){
					nidx = idx;
				}
				if(child.nodeType == 1){
					child["__cachedIndex"] = idx;
					idx++;
				}
				child = child.nextSibling;
			}while(child);
		}else{
			// FIXME: could be incorrect in some cases (node swaps involving
			// the passed node, etc.), but we ignore those for now.
			nidx = ci;
		}
		return nidx;
	}

	var firedCount = 0;

	var _getAttr = function(elem, attr){
		var blank = "";
		if(attr == "class"){
			return elem.className || blank;
		}
		if(attr == "for"){
			return elem.htmlFor || blank;
		}
		return elem.getAttribute(attr, 2) || blank;
	}

	var attrs = [
		// FIXME: need to re-order in order of likelyness to be used in matches
		{
			key: "|=",
			match: function(attr, value){
				// E[hreflang|="en"]
				//		an E element whose "hreflang" attribute has a
				//		hyphen-separated list of values beginning (from the
				//		left) with "en"
				var valueDash = " "+value+"-";
				return function(elem){
					var ea = " "+(elem.getAttribute(attr, 2) || "");
					return (
						(ea == value) ||
						(ea.indexOf(valueDash)==0)
					);
				}
			}
		},
		{
			key: "^=",
			match: function(attr, value){
				return function(elem){
					return (_getAttr(elem, attr).indexOf(value)==0);
				}
			}
		},
		{
			key: "*=",
			match: function(attr, value){
				return function(elem){
					return (_getAttr(elem, attr).indexOf(value)>=0);
				}
			}
		},
		{
			key: "~=",
			match: function(attr, value){
				// return "[contains(concat(' ',@"+attr+",' '), ' "+ value +" ')]";
				var tval = " "+value+" ";
				return function(elem){
					var ea = " "+_getAttr(elem, attr)+" ";
					return (ea.indexOf(tval)>=0);
				}
			}
		},
		{
			key: "$=",
			match: function(attr, value){
				var tval = " "+value;
				return function(elem){
					var ea = " "+_getAttr(elem, attr);
					return (ea.lastIndexOf(value)==(ea.length-value.length));
				}
			}
		},
		{
			key: "!=",
			match: function(attr, value){
				return function(elem){
					return (_getAttr(elem, attr) != value);
				}
			}
		},
		// NOTE: the "=" match MUST come last!
		{
			key: "=",
			match: function(attr, value){
				return function(elem){
					return (_getAttr(elem, attr) == value);
				}
			}
		}
	];

	var pseudos = [
		{
			key: "first-child",
			match: function(name, condition){
				return function(elem){
					if(elem.nodeType != 1){ return false; }
					// check to see if any of the previous siblings are elements
					var fc = elem.previousSibling;
					while(fc && (fc.nodeType != 1)){
						fc = fc.previousSibling;
					}
					return (!fc);
				}
			}
		},
		{
			key: "last-child",
			match: function(name, condition){
				return function(elem){
					if(elem.nodeType != 1){ return false; }
					// check to see if any of the next siblings are elements
					var nc = elem.nextSibling;
					while(nc && (nc.nodeType != 1)){
						nc = nc.nextSibling;
					}
					return (!nc);
				}
			}
		},
		{
			key: "empty",
			match: function(name, condition){
				return function(elem){
					// DomQuery and jQuery get this wrong, oddly enough.
					// The CSS 3 selectors spec is pretty explicit about
					// it, too.
					var cn = elem.childNodes;
					var cnl = elem.childNodes.length;
					// if(!cnl){ return true; }
					for(var x=cnl-1; x >= 0; x--){
						var nt = cn[x].nodeType;
						if((nt == 1)||(nt == 3)){ return false; }
					}
					return true;
				}
			}
		},
		{
			key: "contains",
			match: function(name, condition){
				return function(elem){
					// FIXME: I dislike this version of "contains", as
					// whimsical attribute could set it off. An inner-text
					// based version might be more accurate, but since
					// jQuery and DomQuery also potentially get this wrong,
					// I'm leaving it for now.
					return (elem.innerHTML.indexOf(condition) >= 0);
				}
			}
		},
		{
			key: "not",
			match: function(name, condition){
				var ntf = getFilterFunc(condition);
				return function(elem){
					// FIXME: I dislike this version of "contains", as
					// whimsical attribute could set it off. An inner-text
					// based version might be more accurate, but since
					// jQuery and DomQuery also potentially get this wrong,
					// I'm leaving it for now.
					return (!ntf(elem));
				}
			}
		},
		{
			key: "nth-child",
			match: function(name, condition){
				var pi = parseInt;
				if(condition == "odd"){
					return function(elem){
						return (
							((getNodeIndex(elem)) % 2) == 1
						);
					}
				}else if((condition == "2n")||
					(condition == "even")){
					return function(elem){
						return ((getNodeIndex(elem) % 2) == 0);
					}
				}else if(condition.indexOf("0n+") == 0){
					var ncount = pi(condition.substr(3));
					return function(elem){
						return (elem.parentNode.childNodes[ncount-1] === elem);
					}
				}else if(	(condition.indexOf("n+") > 0) &&
							(condition.length > 3) ){
					var tparts = condition.split("n+", 2);
					var pred = pi(tparts[0]);
					var idx = pi(tparts[1]);
					return function(elem){
						return ((getNodeIndex(elem) % pred) == idx);
					}
				}else if(condition.indexOf("n") == -1){
					var ncount = pi(condition);
					return function(elem){
						return (getNodeIndex(elem) == ncount);
					}
				}
			}
		}
	];

	var getSimpleFilterFunc = function(query){
		// FIXME: this function currently doesn't support chaining of the same
		// sub-selector. E.g., we can't yet search on 
		//		span.thinger.thud 
		// or
		//		div:nth-child(even):last-child

		var fcHit = (_simpleFiltersCache[query]||_filtersCache[query]);
		if(fcHit){ return fcHit; }

		var ff = null;

		// [ q.indexOf("#"), q.indexOf("."), q.indexOf("["), q.indexOf(":") ];
		var i = _getIndexes(query);

		// the only case where we'll need the tag name is if we came from an ID query
		if(i[0] >= 0){ // do we have an ID component?
			var tn = getTagName(query);
			if(tn != "*"){
				ff = agree(ff, function(elem){
					return (elem.tagName.toLowerCase() == tn);
				});
			}
		}

		var matcher;

		// if there's a class in our query, generate a match function for it
		var className = getClassName(query);
		if(className.length){ // do we have a class name component?
			// get the class name
			var isWildcard = className.charAt(className.length-1) == "*";
			if(isWildcard){
				className = className.substr(0, className.length-1);
			}
			// I dislike the regex thing, even if memozied in a cache, but it's VERY short
			var re = new RegExp("(?:^|\\s)" + className + (isWildcard ? ".*" : "") + "(?:\\s|$)");
			ff = agree(ff, function(elem){
				return re.test(elem.className);
			});
		}

		if(i[3]>= 0){ // do we have a pseudo-selector component?
			// NOTE: we count on the pseudo name being at the end
			// FIXME: this is clearly a bug!!!
			var pseudoName = query.substr(i[3]+1);
			var condition = "";
			var obi = pseudoName.indexOf("(");
			var cbi = pseudoName.lastIndexOf(")");
			if(	(0 <= obi)&&
				(0 <= cbi)&&
				(cbi > obi)){
				condition = pseudoName.substring(obi+1, cbi);
				pseudoName = pseudoName.substr(0, obi);
			}

			// NOTE: NOT extensible on purpose until I figure out
			// the portable xpath pseudos extensibility plan.

			// http://www.w3.org/TR/css3-selectors/#structural-pseudos
			matcher = null;
			for(var x=0; x<pseudos.length; x++){
				var ta = pseudos[x];
				if(ta.key == pseudoName){
					matcher = ta.match(pseudoName, condition);
					break;
				}
			}
			if(matcher){
				ff = agree(ff, matcher);
			}
		}

		// [ "#", ".", "[", ":" ];
		var defaultGetter = (d.isIE) ?
			function(cond){
				var clc = cond.toLowerCase();
				return function(elem){
					return elem[cond]||elem[clc];
				}
			} : function(cond){
				return function(elem){
					return (elem && elem.getAttribute && elem.hasAttribute(cond));
				}
			};
		handleAttrs(attrs, query, defaultGetter,
			function(tmatcher){
				ff = agree(ff, tmatcher);
			}
		);
		if(!ff){
			ff = function(){ return true; };
		}
		return _simpleFiltersCache[query] = ff;
	}

	var _getElementsFuncCache = { };

	var getElementsFunc = function(query, root){
		var fHit = _getElementsFuncCache[query];
		if(fHit){ return fHit; }
		// NOTE: this function is in the fast path! not memoized!!!

		// the query doesn't contain any spaces, so there's only so many
		// things it could be
		// [ q.indexOf("#"), q.indexOf("."), q.indexOf("["), q.indexOf(":") ];
		var i = _getIndexes(query);
		var id = getId(query);
		if(i[0] == 0){
			// ID query. Easy.
			return _getElementsFuncCache[query] = function(root){
				return [ d.byId(id) ];
			}
		}

		var filterFunc = getSimpleFilterFunc(query);

		var retFunc;
		if(i[0] >= 0){
			// we got a filtered ID search (e.g., "h4#thinger")
			retFunc = function(root){
				var te = d.byId(id);
				if(filterFunc(te)){
					return [ te ];
				}
			}
		}else{
			// var ret = [];
			var tret;
			var tn = getTagName(query);

			if(Math.max.apply(this, _getIndexes(query)) == -1){
				// if(isTagOnly(query)){
				// it's just a plain-ol elements-by-tag-name query from the root
				retFunc = function(root){
					var ret = [];
					var te, x=0, tret = root.getElementsByTagName(tn);
					while(te=tret[x++]){
						ret.push(te);
					}
					return ret;
				}
			}else{
				retFunc = function(root){
					var ret = [];
					var te, x=0, tret = root.getElementsByTagName(tn);
					while(te=tret[x++]){
						if(filterFunc(te)){
							ret.push(te);
						}
					}
					return ret;
				}
			}
		}
		return _getElementsFuncCache[query] = retFunc;
	}

	var _partsCache = {};

	////////////////////////////////////////////////////////////////////////
	// the query runner
	////////////////////////////////////////////////////////////////////////

	// this is the second level of spliting, from full-length queries (e.g.,
	// "div.foo .bar") into simple query expressions (e.g., ["div.foo",
	// ".bar"])
	var _queryFuncCache = {
		">": function(root){
			var ret = [];
			var te, x=0, tret = root.childNodes;
			while(te=tret[x++]){
				if(te.nodeType == 1){ ret.push(te); }
			}
			return ret;
		}
	};
	var getStepQueryFunc = function(query){
		// if it's trivial, get a fast-path dispatcher
		if(0 > query.indexOf(" ")){
			return getElementsFunc(query);
		}

		// otherwise, break it up and return a runner that iterates over the parts recursively
		var sqf = function(root){
			var qparts = query.split(" "); // FIXME: this is an inaccurate tokenizer!

			/*
			// FIXME: need to make root popping more explicit and cache it somehow

			// see if we can't pop a root off the front
			var partIndex = 0;
			var lastRoot;
			while((partIndex < qparts.length)&&(0 <= qparts[partIndex].indexOf("#"))){
				// FIXME: should we try to cache such that the step function
				// only ever looks for the last ID-based query part, thereby
				// avoiding re-runs and potential array alloc?

				lastRoot = root;
				root = getElementsFunc(qparts[partIndex])()[0];
				if(!root){ root = lastRoot; break; }
				partIndex++;
			}
			// console.debug(qparts[partIndex], root);
			if(qparts.length == partIndex){
				return [ root ];
			}
			// FIXME: need to handle tight-loop "div div div" style queries
			// here so as to avoid huge amounts of array alloc in repeated
			// getElements calls by filterDown
			var candidates;

			// FIXME: this isn't very generic. It lets us run like a bat outta
			// hell on queries like:
			//		div div span
			// and:
			//		#thinger span#blah div span span
			// but we might still fall apart on searches like:
			//		foo.bar span[blah="thonk"] div div span code.example
			// in short, we need to move the look-ahead logic into _filterDown()
			// console.debug(qparts[partIndex]);
			if( isTagOnly(qparts[partIndex]) && 
				(qparts[partIndex+1] != ">")
			){
				// go as far as we can down the chain without any intermediate
				// array allocation
				qparts = qparts.slice(partIndex);
				var searchParts = [];
				var idx = 0;
				while(qparts[idx] && isTagOnly(qparts[idx]) && (qparts[idx+1] != ">" )){
					searchParts.push(qparts[idx]);
					idx++;
				}
				var curLevelItems = [ root ];
				var nextLevelItems;
				for(var x=0; x<searchParts.length; x++){
					nextLevelItems = [];
					var tsp = qparts.shift();
					for(var y=0; y<curLevelItems.length; y++){
						var tze, z=0, tcol = curLevelItems[y].getElementsByTagName(tsp);
						while(tze=tcol[z++]){
							nextLevelItems.push(tze);
						}
					}
					curLevelItems = nextLevelItems;
				}
				candidates = curLevelItems;
				if(!qparts.length){
					return candidates;
				}
			}else{
				// console.debug(qparts);
				candidates = getElementsFunc(qparts.shift())(root);
			}
			*/
			var candidates;
			if(qparts[0] == ">"){
				candidates = [ root ];
				// root = document;
			}else{
				candidates = getElementsFunc(qparts.shift())(root);
			}
			return filterDown(candidates, qparts);
		}
		return sqf;
	}

	// a specialized method that implements our primoridal "query optimizer".
	// This allows us to dispatch queries to the fastest subsystem we can get.
	var _getQueryFunc = (
		// NOTE: 
		//		XPath on the Webkit nighlies is slower than it's DOM iteration
		//		for most test cases
		// FIXME: 
		//		we should try to capture some runtime speed data for each query
		//		function to determine on the fly if we should stick w/ the
		//		potentially optimized variant or if we should try something
		//		new.
		(document["evaluate"] && !d.isSafari) ? 
		function(query){
			// has xpath support that's faster than DOM
			var qparts = query.split(" ");
			// can we handle it?
			if(	(document["evaluate"])&&
				(query.indexOf(":") == -1)&&
				(
					(true) // ||
					// (query.indexOf("[") == -1) ||
					// (query.indexOf("=") == -1)
				)
			){
				// dojo.debug(query);
				// should we handle it?

				// var gtIdx = query.indexOf(">")

				// kind of a lame heuristic, but it works
				if(	
					// a "div div div" style query
					((qparts.length > 2)&&(query.indexOf(">") == -1))||
					// or something else with moderate complexity. kinda janky
					(qparts.length > 3)||
					(query.indexOf("[")>=0)||
					// or if it's a ".thinger" query
					((1 == qparts.length)&&(0 <= query.indexOf(".")))

				){
					// use get and cache a xpath runner for this selector
					return getXPathFunc(query);
				}
			}

			// fallthrough
			return getStepQueryFunc(query);
		} : getStepQueryFunc
	);
	// uncomment to disable XPath for testing and tuning the DOM path
	// _getQueryFunc = getStepQueryFunc;

	// FIXME: we've got problems w/ the NodeList query()/filter() functions if we go XPath for everything

	// uncomment to disable DOM queries for testing and tuning XPath
	// _getQueryFunc = getXPathFunc;

	// this is the primary caching for full-query results. The query dispatcher
	// functions are generated here and then pickled for hash lookup in the
	// future
	var getQueryFunc = function(query){
		// return a cached version if one is available
		if(_queryFuncCache[query]){ return _queryFuncCache[query]; }
		if(0 > query.indexOf(",")){
			// if it's not a compound query (e.g., ".foo, .bar"), cache and return a dispatcher
			return _queryFuncCache[query] = _getQueryFunc(query);
		}else{
			// if it's a complex query, break it up into it's constituent parts
			// and return a dispatcher that will merge the parts when run

			// var parts = query.split(", ");
			var parts = query.split(/\s*,\s*/);
			var tf = function(root){
				var pindex = 0; // avoid array alloc for every invocation
				var ret = [];
				var tp;
				while(tp = parts[pindex++]){
					ret = ret.concat(_getQueryFunc(tp, tp.indexOf(" "))(root));
				}
				return ret;
			}
			// ...cache and return
			return _queryFuncCache[query] = tf;
		}
	}

	/*
	// experimental implementation of _zip for IE
	var _zip = function(arr){
		if(!arr){ return []; }
		var al = arr.length;
		if(al < 2){ return arr; }
		var ret = [arr[0]];
		var lastIdx = arr[0].sourceIndex;
		for(var x=1, te; te = arr[x]; x++){
		// for(var x=1; x<arr.length; x++){
			var nsi = te.sourceIndex;
			if(nsi > lastIdx){
				ret.push(te);
				lastIdx = nsi;
			}
		}
		return ret;
	}
	*/

	// FIXME: 
	//		Dean's new Base2 uses a system whereby queries themselves note if
	//		they'll need duplicate filtering. We need to get on that plan!!

	// attempt to efficiently determine if an item in a list is a dupe,
	// returning a list of "uniques", hopefully in doucment order
	var _zipIdx = 0;
	var _zip = function(arr){
		var ret = new d.NodeList();
		if(!arr){ return ret; }
		if(arr[0]){
			ret.push(arr[0]);
		}
		if(arr.length < 2){ return ret; }
		_zipIdx++;
		arr[0]["_zipIdx"] = _zipIdx;
		for(var x=1, te; te = arr[x]; x++){
			if(arr[x]["_zipIdx"] != _zipIdx){ 
				ret.push(te);
			}
			te["_zipIdx"] = _zipIdx;
		}
		// FIXME: should we consider stripping these properties?
		return ret;
	}

	// the main exectuor
	d.query = function(query, root){
		// return is always an array
		// NOTE: elementsById is not currently supported
		// NOTE: ignores xpath-ish queries for now
		if(typeof query != "string"){
			return new d.NodeList(query);
		}
		if(typeof root == "string"){
			root = d.byId(root);
		}

		// FIXME: should support more methods on the return than the stock array.
		return _zip(getQueryFunc(query)(root||d.doc));
	}

	/*
	// exposing these was a mistake
	d.query.attrs = attrs;
	d.query.pseudos = pseudos;
	*/

	// one-off function for filtering a NodeList based on a simple selector
	d._filterQueryResult = function(nodeList, simpleFilter){
		var tnl = new d.NodeList();
		var ff = (simpleFilter) ? getFilterFunc(simpleFilter) : function(){ return true; };
		// dojo.debug(ff);
		for(var x=0, te; te = nodeList[x]; x++){
			if(ff(te)){ tnl.push(te); }
		}
		return tnl;
	}
})();