class REXML::XPathParser
You don't want to use this class. Really. Use XPath
, which is a wrapper for this class. Believe me. You don't want to poke around in here. There is strange, dark magic at work in this code. Beware. Go back! Go back while you still can!
Constants
- ALL
Expr takes a stack of path elements and a set of nodes (either a
Parent
or an Array and returns an Array of matching nodes- ELEMENTS
- LITERAL
Public Class Methods
# File lib/rexml/xpath_parser.rb, line 50 def initialize( ) @parser = REXML::Parsers::XPathParser.new @namespaces = nil @variables = {} end
Public Instance Methods
# File lib/rexml/xpath_parser.rb, line 81 def []=( variable_name, value ) @variables[ variable_name ] = value end
Performs a depth-first (document order) XPath
search, and returns the first match. This is the fastest, lightest way to return a single result.
FIXME: This method is incomplete!
# File lib/rexml/xpath_parser.rb, line 90 def first( path_stack, node ) return nil if path.size == 0 case path[0] when :document # do nothing return first( path[1..-1], node ) when :child for c in node.children r = first( path[1..-1], c ) return r if r end when :qname name = path[2] if node.name == name return node if path.size == 3 return first( path[3..-1], node ) else return nil end when :descendant_or_self r = first( path[1..-1], node ) return r if r for c in node.children r = first( path, c ) return r if r end when :node return first( path[1..-1], node ) when :any return first( path[1..-1], node ) end return nil end
# File lib/rexml/xpath_parser.rb, line 71 def get_first path, nodeset path_stack = @parser.parse( path ) first( path_stack, nodeset ) end
# File lib/rexml/xpath_parser.rb, line 126 def match( path_stack, nodeset ) r = expr( path_stack, nodeset ) r end
# File lib/rexml/xpath_parser.rb, line 56 def namespaces=( namespaces={} ) Functions::namespace_context = namespaces @namespaces = namespaces end
# File lib/rexml/xpath_parser.rb, line 66 def parse path, nodeset path_stack = @parser.parse( path ) match( path_stack, nodeset ) end
# File lib/rexml/xpath_parser.rb, line 76 def predicate path, nodeset path_stack = @parser.parse( path ) expr( path_stack, nodeset ) end
# File lib/rexml/xpath_parser.rb, line 61 def variables=( vars={} ) Functions::variables = vars @variables = vars end
Private Instance Methods
# File lib/rexml/xpath_parser.rb, line 681 def compare a, op, b case op when :eq a == b when :neq a != b when :lt a < b when :lteq a <= b when :gt a > b when :gteq a >= b when :and a and b when :or a or b else false end end
# File lib/rexml/xpath_parser.rb, line 477 def d_o_s( p, ns, r ) nt = nil ns.each_index do |i| n = ns[i] x = expr( p.dclone, [ n ] ) nt = n.node_type d_o_s( p, n.children, x ) if nt == :element or nt == :document and n.children.size > 0 r.concat(x) if x.size > 0 end end
FIXME The next two methods are BAD MOJO! This is my achilles heel. If anybody thinks of a better way of doing this, be my guest. This really sucks, but it is a wonder it works at all. ########################################################
# File lib/rexml/xpath_parser.rb, line 470 def descendant_or_self( path_stack, nodeset ) rs = [] d_o_s( path_stack, nodeset, rs ) document_order(rs.flatten.compact) #rs.flatten.compact end
Reorders an array of nodes so that they are in document order It tries to do this efficiently.
FIXME: I need to get rid of this, but the issue is that most of the XPath
interpreter functions as a filter, which means that we lose context going in and out of function calls. If I knew what the index of the nodes was, I wouldn't have to do this. Maybe add a document IDX for each node? Problems with mutable documents. Or, rewrite everything.
# File lib/rexml/xpath_parser.rb, line 497 def document_order( array_of_nodes ) new_arry = [] array_of_nodes.each { |node| node_idx = [] np = node.node_type == :attribute ? node.element : node while np.parent and np.parent.node_type == :element node_idx << np.parent.index( np ) np = np.parent end new_arry << [ node_idx.reverse, node ] } new_arry.sort{ |s1, s2| s1[0] <=> s2[0] }.collect{ |s| s[1] } end
# File lib/rexml/xpath_parser.rb, line 603 def equality_relational_compare( set1, op, set2 ) if set1.kind_of? Array and set2.kind_of? Array if set1.size == 1 and set2.size == 1 set1 = set1[0] set2 = set2[0] elsif set1.size == 0 or set2.size == 0 nd = set1.size==0 ? set2 : set1 rv = nd.collect { |il| compare( il, op, nil ) } return rv else res = [] SyncEnumerator.new( set1, set2 ).each { |i1, i2| i1 = norm( i1 ) i2 = norm( i2 ) res << compare( i1, op, i2 ) } return res end end # If one is nodeset and other is number, compare number to each item # in nodeset s.t. number op number(string(item)) # If one is nodeset and other is string, compare string to each item # in nodeset s.t. string op string(item) # If one is nodeset and other is boolean, compare boolean to each item # in nodeset s.t. boolean op boolean(item) if set1.kind_of? Array or set2.kind_of? Array if set1.kind_of? Array a = set1 b = set2 else a = set2 b = set1 end case b when true, false return a.collect {|v| compare( Functions::boolean(v), op, b ) } when Numeric return a.collect {|v| compare( Functions::number(v), op, b )} when /^\d+(\.\d+)?$/ b = Functions::number( b ) return a.collect {|v| compare( Functions::number(v), op, b )} else b = Functions::string( b ) return a.collect { |v| compare( Functions::string(v), op, b ) } end else # If neither is nodeset, # If op is = or != # If either boolean, convert to boolean # If either number, convert to number # Else, convert to string # Else # Convert both to numbers and compare s1 = set1.to_s s2 = set2.to_s if s1 == 'true' or s1 == 'false' or s2 == 'true' or s2 == 'false' set1 = Functions::boolean( set1 ) set2 = Functions::boolean( set2 ) else if op == :eq or op == :neq if s1 =~ /^\d+(\.\d+)?$/ or s2 =~ /^\d+(\.\d+)?$/ set1 = Functions::number( s1 ) set2 = Functions::number( s2 ) else set1 = Functions::string( set1 ) set2 = Functions::string( set2 ) end else set1 = Functions::number( set1 ) set2 = Functions::number( set2 ) end end return compare( set1, op, set2 ) end return false end
# File lib/rexml/xpath_parser.rb, line 153 def expr( path_stack, nodeset, context=nil ) node_types = ELEMENTS return nodeset if path_stack.length == 0 || nodeset.length == 0 while path_stack.length > 0 if nodeset.length == 0 path_stack.clear return [] end case (op = path_stack.shift) when :document nodeset = [ nodeset[0].root_node ] when :qname prefix = path_stack.shift name = path_stack.shift nodeset.delete_if do |node| # FIXME: This DOUBLES the time XPath searches take ns = get_namespace( node, prefix ) if node.node_type == :element if node.name == name end end !(node.node_type == :element and node.name == name and node.namespace == ns ) end node_types = ELEMENTS when :any nodeset.delete_if { |node| !node_types.include?(node.node_type) } when :self # This space left intentionally blank when :processing_instruction target = path_stack.shift nodeset.delete_if do |node| (node.node_type != :processing_instruction) or ( target!='' and ( node.target != target ) ) end when :text nodeset.delete_if { |node| node.node_type != :text } when :comment nodeset.delete_if { |node| node.node_type != :comment } when :node # This space left intentionally blank node_types = ALL when :child new_nodeset = [] nt = nil nodeset.each do |node| nt = node.node_type new_nodeset += node.children if nt == :element or nt == :document end nodeset = new_nodeset node_types = ELEMENTS when :literal return path_stack.shift when :attribute new_nodeset = [] case path_stack.shift when :qname prefix = path_stack.shift name = path_stack.shift for element in nodeset if element.node_type == :element attrib = element.attribute( name, get_namespace(element, prefix) ) new_nodeset << attrib if attrib end end when :any for element in nodeset if element.node_type == :element new_nodeset += element.attributes.to_a end end end nodeset = new_nodeset when :parent nodeset = nodeset.collect{|n| n.parent}.compact #nodeset = expr(path_stack.dclone, nodeset.collect{|n| n.parent}.compact) node_types = ELEMENTS when :ancestor new_nodeset = [] nodeset.each do |node| while node.parent node = node.parent new_nodeset << node unless new_nodeset.include? node end end nodeset = new_nodeset node_types = ELEMENTS when :ancestor_or_self new_nodeset = [] nodeset.each do |node| if node.node_type == :element new_nodeset << node while ( node.parent ) node = node.parent new_nodeset << node unless new_nodeset.include? node end end end nodeset = new_nodeset node_types = ELEMENTS when :predicate new_nodeset = [] subcontext = { :size => nodeset.size } pred = path_stack.shift nodeset.each_with_index { |node, index| subcontext[ :node ] = node subcontext[ :index ] = index+1 pc = pred.dclone result = expr( pc, [node], subcontext ) result = result[0] if result.kind_of? Array and result.length == 1 if result.kind_of? Numeric new_nodeset << node if result == (index+1) elsif result.instance_of? Array if result.size > 0 and result.inject(false) {|k,s| s or k} new_nodeset << node if result.size > 0 end else new_nodeset << node if result end } nodeset = new_nodeset =begin predicate = path_stack.shift ns = nodeset.clone result = expr( predicate, ns ) if result.kind_of? Array nodeset = result.zip(ns).collect{|m,n| n if m}.compact else nodeset = result ? nodeset : [] end =end when :descendant_or_self rv = descendant_or_self( path_stack, nodeset ) path_stack.clear nodeset = rv node_types = ELEMENTS when :descendant results = [] nt = nil nodeset.each do |node| nt = node.node_type results += expr( path_stack.dclone.unshift( :descendant_or_self ), node.children ) if nt == :element or nt == :document end nodeset = results node_types = ELEMENTS when :following_sibling results = [] nodeset.each do |node| next if node.parent.nil? all_siblings = node.parent.children current_index = all_siblings.index( node ) following_siblings = all_siblings[ current_index+1 .. -1 ] results += expr( path_stack.dclone, following_siblings ) end nodeset = results when :preceding_sibling results = [] nodeset.each do |node| next if node.parent.nil? all_siblings = node.parent.children current_index = all_siblings.index( node ) preceding_siblings = all_siblings[ 0, current_index ].reverse results += preceding_siblings end nodeset = results node_types = ELEMENTS when :preceding new_nodeset = [] nodeset.each do |node| new_nodeset += preceding( node ) end nodeset = new_nodeset node_types = ELEMENTS when :following new_nodeset = [] nodeset.each do |node| new_nodeset += following( node ) end nodeset = new_nodeset node_types = ELEMENTS when :namespace new_nodeset = [] prefix = path_stack.shift nodeset.each do |node| if (node.node_type == :element or node.node_type == :attribute) if @namespaces namespaces = @namespaces elsif (node.node_type == :element) namespaces = node.namespaces else namespaces = node.element.namesapces end if (node.namespace == namespaces[prefix]) new_nodeset << node end end end nodeset = new_nodeset when :variable var_name = path_stack.shift return @variables[ var_name ] # :and, :or, :eq, :neq, :lt, :lteq, :gt, :gteq # TODO: Special case for :or and :and -- not evaluate the right # operand if the left alone determines result (i.e. is true for # :or and false for :and). when :eq, :neq, :lt, :lteq, :gt, :gteq, :or left = expr( path_stack.shift, nodeset.dup, context ) right = expr( path_stack.shift, nodeset.dup, context ) res = equality_relational_compare( left, op, right ) return res when :and left = expr( path_stack.shift, nodeset.dup, context ) return [] unless left if left.respond_to?(:inject) and !left.inject(false) {|a,b| a | b} return [] end right = expr( path_stack.shift, nodeset.dup, context ) res = equality_relational_compare( left, op, right ) return res when :div left = Functions::number(expr(path_stack.shift, nodeset, context)).to_f right = Functions::number(expr(path_stack.shift, nodeset, context)).to_f return (left / right) when :mod left = Functions::number(expr(path_stack.shift, nodeset, context )).to_f right = Functions::number(expr(path_stack.shift, nodeset, context )).to_f return (left % right) when :mult left = Functions::number(expr(path_stack.shift, nodeset, context )).to_f right = Functions::number(expr(path_stack.shift, nodeset, context )).to_f return (left * right) when :plus left = Functions::number(expr(path_stack.shift, nodeset, context )).to_f right = Functions::number(expr(path_stack.shift, nodeset, context )).to_f return (left + right) when :minus left = Functions::number(expr(path_stack.shift, nodeset, context )).to_f right = Functions::number(expr(path_stack.shift, nodeset, context )).to_f return (left - right) when :union left = expr( path_stack.shift, nodeset, context ) right = expr( path_stack.shift, nodeset, context ) return (left | right) when :neg res = expr( path_stack, nodeset, context ) return -(res.to_f) when :not when :function func_name = path_stack.shift.tr('-','_') arguments = path_stack.shift subcontext = context ? nil : { :size => nodeset.size } res = [] cont = context nodeset.each_with_index { |n, i| if subcontext subcontext[:node] = n subcontext[:index] = i cont = subcontext end arg_clone = arguments.dclone args = arg_clone.collect { |arg| expr( arg, [n], cont ) } Functions.context = cont res << Functions.send( func_name, *args ) } return res end end # while return nodeset end
# File lib/rexml/xpath_parser.rb, line 561 def following( node ) acc = [] p = next_sibling_node( node ) while p acc << p p = following_node_of( p ) end acc end
# File lib/rexml/xpath_parser.rb, line 571 def following_node_of( node ) if node.kind_of? Element and node.children.size > 0 return node.children[0] end return next_sibling_node(node) end
Returns a String namespace for a node, given a prefix The rules are:
1. Use the supplied namespace mapping first. 2. If no mapping was supplied, use the context node to look up the namespace
# File lib/rexml/xpath_parser.rb, line 139 def get_namespace( node, prefix ) if @namespaces return @namespaces[prefix] || '' else return node.namespace( prefix ) if node.node_type == :element return '' end end
# File lib/rexml/xpath_parser.rb, line 578 def next_sibling_node(node) psn = node.next_sibling_node while psn.nil? if node.parent.nil? or node.parent.class == Document return nil end node = node.parent psn = node.next_sibling_node end return psn end
# File lib/rexml/xpath_parser.rb, line 590 def norm b case b when true, false return b when 'true', 'false' return Functions::boolean( b ) when /^\d+(\.\d+)?$/ return Functions::number( b ) else return Functions::string( b ) end end
Builds a nodeset of all of the preceding nodes of the supplied node, in reverse document order
- preceding
-
includes every element in the document that precedes this node,
except for ancestors
# File lib/rexml/xpath_parser.rb, line 525 def preceding( node ) ancestors = [] p = node.parent while p ancestors << p p = p.parent end acc = [] p = preceding_node_of( node ) while p if ancestors.include? p ancestors.delete(p) else acc << p end p = preceding_node_of( p ) end acc end
# File lib/rexml/xpath_parser.rb, line 546 def preceding_node_of( node ) psn = node.previous_sibling_node if psn.nil? if node.parent.nil? or node.parent.class == Document return nil end return node.parent #psn = preceding_node_of( node.parent ) end while psn and psn.kind_of? Element and psn.children.size > 0 psn = psn.children[-1] end psn end
# File lib/rexml/xpath_parser.rb, line 512 def recurse( nodeset, &block ) for node in nodeset yield node recurse( node, &block ) if node.node_type == :element end end