require 'digest/sha1' require 'rdf' require 'rdf/ntriples' module RDF ## # Isomorphism for rdf.rb Enumerables # # RDF::Isomorphic provides the functions isomorphic_with and bijection_to for RDF::Enumerable. # # @see http://rdf.rubyforge.org # @see http://www.hpl.hp.com/techreports/2001/HPL-2001-293.pdf module Isomorphic autoload :VERSION, 'rdf/isomorphic/version' # Returns `true` if this RDF::Enumerable is isomorphic with another. # # Takes a canonicalize: true argument. If true, RDF::Literals will be # canonicalized while producing a bijection. This results in broader # matches for isomorphism in the case of equivalent literals with different # representations. # # @param opts [Hash Any>] options # @param other [RDF::Enumerable] # @return [Boolean] # @example # repository_a.isomorphic_with repository_b #=> true def isomorphic_with?(other, opts = {}) !(bijection_to(other, opts).nil?) end alias_method :isomorphic?, :isomorphic_with? # Returns a hash of RDF:Nodes: RDF::Nodes representing an isomorphic # bijection of this RDF::Enumerable's to another RDF::Enumerable's blank # nodes, or nil if a bijection cannot be found. # # Takes a canonicalize: true argument. If true, RDF::Literals will be # canonicalized while producing a bijection. This results in broader # matches for isomorphism in the case of equivalent literals with different # representations. # # @example # repository_a.bijection_to repository_b # @param other [RDF::Enumerable] # @param opts [Hash Any>] options # @return [Hash, nil] def bijection_to(other, opts = {}) grounded_stmts_match = (count == other.count) grounded_stmts_match &&= each_statement.all? do | stmt | stmt.node? || other.has_statement?(stmt) end if grounded_stmts_match # blank_stmts and other_blank_stmts are just a performance # consideration--we could just as well pass in self and other. But we # will be iterating over this list quite a bit during the algorithm, so # we break it down to the parts we're interested in. blank_stmts = find_all { |statement| statement.node? } other_blank_stmts = other.find_all { |statement| statement.node? } nodes = RDF::Isomorphic.blank_nodes_in(blank_stmts) other_nodes = RDF::Isomorphic.blank_nodes_in(other_blank_stmts) build_bijection_to blank_stmts, nodes, other_blank_stmts, other_nodes, {}, {}, opts else nil end end private # The main recursive bijection algorithm. # # This algorithm is very similar to the one explained by Jeremy Carroll in # http://www.hpl.hp.com/techreports/2001/HPL-2001-293.pdf. Page 12 has the # relevant pseudocode. # # Many more comments are in the method itself. # # @param [RDF::Enumerable] anon_stmts # @param [Array] nodes # @param [RDF::Enumerable] other_anon_stmts # @param [Array] other_nodes # @param [Hash] these_grounded_hashes # @param [Hash] other_grounded_hashes # @param [Hash] opts # @return [nil,Hash] # @private def build_bijection_to(anon_stmts, nodes, other_anon_stmts, other_nodes, these_grounded_hashes = {}, other_grounded_hashes = {}, opts = {}) # Create a hash signature of every node, based on the signature of # statements it exists in. # We also save hashes of nodes that cannot be reliably known; we will use # that information to eliminate possible recursion combinations. # # Any mappings given in the method parameters are considered grounded. these_hashes, these_ungrounded_hashes = RDF::Isomorphic.hash_nodes(anon_stmts, nodes, these_grounded_hashes, opts[:canonicalize]) other_hashes, other_ungrounded_hashes = RDF::Isomorphic.hash_nodes(other_anon_stmts, other_nodes, other_grounded_hashes, opts[:canonicalize]) # Grounded hashes are built at the same rate between the two graphs (if # they are isomorphic). If there exists a grounded node in one that is # not in the other, we can just return. Ungrounded nodes might still # conflict, so we don't check them. This is a little bit messy in the # middle of the method, and probably slows down isomorphic checks, but # prevents almost-isomorphic cases from getting nutty. return nil if these_hashes.values.any? { |hash| !(other_hashes.values.member?(hash)) } return nil if other_hashes.values.any? { |hash| !(these_hashes.values.member?(hash)) } # Using the created hashes, map nodes to other_nodes # Ungrounded hashes will also be equal, but we keep the distinction # around for when we recurse later (we only recurse on ungrounded nodes) bijection = {} nodes.each do | node | other_node, other_hash = other_ungrounded_hashes.find do | other_node, other_hash | # we need to use eql?, as coincedentally-named bnode identifiers are == in rdf.rb these_ungrounded_hashes[node].eql? other_hash end next unless other_node bijection[node] = other_node # Deletion is required to keep counts even; two nodes with identical # signatures can biject to each other at random. other_ungrounded_hashes.delete other_node end # bijection is now a mapping of nodes to other_nodes. If all are # accounted for on both sides, we have a bijection. # # If not, we will speculatively mark pairs with matching ungrounded # hashes as bijected and recurse. unless (bijection.keys.sort == nodes.sort) && (bijection.values.sort == other_nodes.sort) bijection = nil nodes.any? do | node | # We don't replace grounded nodes' hashes next if these_hashes.member? node other_nodes.any? do | other_node | # We don't replace grounded other_nodes' hashes next if other_hashes.member? other_node # The ungrounded signature must match for this to potentially work next unless these_ungrounded_hashes[node] == other_ungrounded_hashes[other_node] hash = Digest::SHA1.hexdigest(node.to_s) bijection = build_bijection_to(anon_stmts, nodes, other_anon_stmts, other_nodes, these_hashes.merge( node => hash), other_hashes.merge(other_node => hash)) end bijection end end bijection end # Blank nodes appearing in given list of statements # @private # @param [Array] blank_stmt_list # @return [Array] def self.blank_nodes_in(blank_stmt_list) blank_stmt_list.map {|statement | statement.to_quad.compact.select(&:node?)}.flatten.uniq end # Given a set of statements, create a mapping of node => SHA1 for a given # set of blank nodes. grounded_hashes is a mapping of node => SHA1 pairs # that we will take as a given, and use those to make more specific # signatures of other nodes. # # Returns a tuple of hashes: one of grounded hashes, and one of all # hashes. grounded hashes are based on non-blank nodes and grounded blank # nodes, and can be used to determine if a node's signature matches # another. # # @param [Array] statements # @param [Array] nodes # @param [Hash] grounded_hashes # @private # @return [Hash, Hash] def self.hash_nodes(statements, nodes, grounded_hashes, canonicalize = false) hashes = grounded_hashes.dup ungrounded_hashes = {} hash_needed = true # We may have to go over the list multiple times. If a node is marked as # grounded, other nodes can then use it to decide their own state of # grounded. while hash_needed starting_grounded_nodes = hashes.size nodes.each do | node | unless hashes.member? node grounded, hash = node_hash_for(node, statements, hashes, canonicalize) if grounded hashes[node] = hash end ungrounded_hashes[node] = hash end end # after going over the list, any nodes with a unique hash can be marked # as grounded, even if we have not tied them back to a root yet. uniques = {} ungrounded_hashes.each do |node, hash| uniques[hash] = uniques.has_key?(hash) ? false : node end uniques.each do |hash, node| hashes[node] = hash if node end hash_needed = starting_grounded_nodes != hashes.size end [hashes,ungrounded_hashes] end # Generate a hash for a node based on the signature of the statements it # appears in. Signatures consist of grounded elements in statements # associated with a node, that is, anything but an ungrounded anonymous # node. Creating the hash is simply hashing a sorted list of each # statement's signature, which is itself a concatenation of the string form # of all grounded elements. # # Nodes other than the given node are considered grounded if they are a # member in the given hash. # # Returns a tuple consisting of grounded being true or false and the String # for the hash # @private # @param [RDF::Node] node # @param [Array] statements # @param [Hash] hashes # @param [Boolean] canonicalize # @return [Boolean, String] def self.node_hash_for(node, statements, hashes, canonicalize) statement_signatures = [] grounded = true statements.each do | statement | if statement.to_quad.include?(node) statement_signatures << hash_string_for(statement, hashes, node, canonicalize) statement.to_quad.compact.each do | resource | grounded = false unless grounded?(resource, hashes) || resource == node end end end # Note that we sort the signatures--without a canonical ordering, # we might get different hashes for equivalent nodes. [grounded,Digest::SHA1.hexdigest(statement_signatures.sort.to_s)] end # Provide a string signature for the given statement, collecting # string signatures for grounded node elements. # return [String] # @private def self.hash_string_for(statement, hashes, node, canonicalize) statement.to_quad.map {|r| string_for_node(r, hashes, node, canonicalize)}.join("") end # Returns true if a given node is grounded # A node is groundd if it is not a blank node or it is included # in the given mapping of grounded nodes. # @return [Boolean] # @private def self.grounded?(node, hashes) (!(node.node?)) || (hashes.member? node) end # Provides a string for the given node for use in a string signature # Non-anonymous nodes will return their string form. Grounded anonymous # nodes will return their hashed form. # @return [String] # @private def self.string_for_node(node, hashes,target, canonicalize) case when node.nil? "" when node == target "itself" when node.node? && hashes.member?(node) hashes[node] when node.node? "a blank node" # RDF.rb auto-boxing magic makes some literals the same when they # should not be; the ntriples serializer will take care of us when node.literal? node.class.name + RDF::NTriples.serialize(canonicalize ? node.canonicalize : node) else node.to_s end end end # Extend RDF::Enumerables with these functions. module Enumerable include RDF::Isomorphic end end