# frozen_string_literal: true
require 'rdf'
require 'rdf/vocabulary'
require 'rdf/vocab'

# Monkey-patch RDF::Vocab.each to load all vocabularies

module RDF
  class Vocabulary
    class << self
      ##
      # Enumerates known RDF vocabulary classes.
      #
      # @yield  [klass]
      # @yieldparam [Class] klass
      # @return [Enumerator]
      alias_method :_orig_each, :each
      def each(&block)
        # This is needed since all vocabulary classes are defined using
        # Ruby's autoloading facility, meaning that `@@subclasses` will be
        # empty until each subclass has been touched or require'd.
        RDF::Vocab::VOCABS.each do |n, params|
          begin
            require "rdf/vocab/#{n}"
          rescue LoadError
            # Transient error
            warn "Failed to load #{n}"
          end
        end
        _orig_each(&block)
      end

      begin
        require 'rdf/turtle'
        ##
        # Generate Turtle representation, specific to vocabularies
        #
        # @param [RDF::Queryable] :graph Optional graph, otherwise uses statements from vocabulary.
        # @param [Hash{#to_sym => String}] Prefixes to add to output
        # @return [String]
        def to_ttl(graph: nil, prefixes: nil)
          output = []

          # Find namespaces used in the vocabulary
          graph = RDF::Graph.new {|g| each_statement {|s| g << s}} if graph.nil? || graph.empty?

          prefixes = vocab_prefixes(graph).merge(prefixes || {})
          pfx_width = prefixes.keys.map(&:to_s).map(&:length).max
          prefixes.each do |pfx, uri|
            output << "@prefix %*s: <%s> .\n" % [pfx_width, pfx, uri]
          end

          # Determine the category for each subject in the vocabulary graph
          cats = subject_categories(graph)

          writer = RDF::Turtle::Writer.new(StringIO.new, prefixes: prefixes)

          {
            ont: {
              heading:  "# #{__name__.split('::').last} Vocabulary definition\n"
            },
            classes: {
              heading:  "# Class definitions\n"
            },
            properties: {
              heading:  "# Property definitions\n"
            },
            datatypes: {
              heading:  "# Datatype definitions\n"
            },
            other: {
              heading:  "# Other definitions\n"
            }
          }.each do |key, hash|
            next unless cats[key]

            output << "\n\n#{hash[:heading]}"

            cats[key].each do |subject|
              po = {}

              # Group predicates with their values
              graph.query(subject: subject) do |statement|
                # Sanity check this, as these are set to an empty string if not defined.
                next if [RDF::RDFS.label, RDF::RDFS.comment].include?(statement.predicate) && statement.object.to_s.empty?
                po[statement.predicate] ||= []
                po[statement.predicate] << statement.object
              end

              next if po.empty?

              po_list = []
              unless (types = po.delete(RDF.type)).empty?
                po_list << 'a ' + types.map {|o| writer.format_term(o)}.join(", ")
              end

              # Serialize other predicate/objects
              po.each do |predicate, objects|
                resource = predicate.qname ? predicate.pname : "<#{predicate}>"
                po_list << resource + ' ' + objects.map {|o| writer.format_term(o)}.join(", ")
              end

              # Output statements for this subject
              subj = subject.qname ? subject.pname : "<#{subject}>"
              output << "#{subj} " + po_list.join(";\n  ") + "\n  .\n"
            end
          end

          output.join("")
        end
      rescue LoadError
        # No Turtle serialization unless gem loaded
      end

      begin
        require 'json/ld'

        ##
        # Generate JSON-LD representation, specific to vocabularies
        #
        # @param [RDF::Queryable] :graph Optional graph, otherwise uses statements from vocabulary.
        # @param [Hash{#to_sym => String}] Prefixes to add to output
        # @return [String]
        def to_jsonld(graph: nil, prefixes: nil)
          context = {}
          rdfs_context = ::JSON.parse %({
            "dc:title": {"@container": "@language"},
            "dc:description": {"@container": "@language"},
            "dc:date": {"@type": "xsd:date"},
            "rdfs:comment": {"@container": "@language"},
            "rdfs:domain": {"@type": "@vocab"},
            "rdfs:label": {"@container": "@language"},
            "rdfs:range": {"@type": "@vocab"},
            "rdfs:seeAlso": {"@type": "@id"},
            "rdfs:subClassOf": {"@type": "@vocab"},
            "rdfs:subPropertyOf": {"@type": "@vocab"},
            "schema:domainIncludes": {"@type": "@vocab"},
            "schema:rangeIncludes": {"@type": "@vocab"},
            "owl:equivalentClass": {"@type": "@vocab"},
            "owl:equivalentProperty": {"@type": "@vocab"},
            "owl:oneOf": {"@container": "@list", "@type": "@vocab"},
            "owl:imports": {"@type": "@id"},
            "owl:versionInfo": {"@type": "@id"},
            "owl:inverseOf": {"@type": "@vocab"},
            "owl:unionOf": {"@type": "@vocab", "@container": "@list"},
            "rdfs_classes": {"@reverse": "rdfs:isDefinedBy", "@type": "@id"},
            "rdfs_properties": {"@reverse": "rdfs:isDefinedBy", "@type": "@id"},
            "rdfs_datatypes": {"@reverse": "rdfs:isDefinedBy", "@type": "@id"},
            "rdfs_instances": {"@reverse": "rdfs:isDefinedBy", "@type": "@id"}
          })
          rdfs_classes, rdfs_properties, rdfs_datatypes, rdfs_instances = [], [], [], [], []

          ontology = {
            "@context" => rdfs_context,
            "@id" => to_uri.to_s
          }

          # Find namespaces used in the vocabulary
          graph = RDF::Graph.new {|g| each_statement {|s| g << s}} if graph.nil? || graph.empty?

          prefixes = vocab_prefixes(graph).merge(prefixes || {})
          prefixes.each do |pfx, uri|
            context[pfx.to_s] = uri.to_s unless pfx.to_s.empty?
          end

          # Determine the category for each subject in the vocabulary graph
          cats = subject_categories(graph)

          # Generate term definitions from graph subjects
          cats.values.flatten.each do |term|
            next unless Array(term.qname).length == 2
            context[term.qname.last.to_s] = term.to_uri.to_s
          end

          # Parse the two contexts so we know what terms are in scope
          jld_context = ::JSON::LD::Context.new.parse([context, rdfs_context])

          {
            ont: {
              heading:  "# #{__name__.split('::').last} Vocabulary definition\n",
              bucket:   ontology,
            },
            classes: {
              heading:  "# Class definitions\n",
              bucket:   rdfs_classes,
              rev_prop: "rdfs_classes"
            },
            properties: {
              heading:  "# Property definitions\n",
              bucket:   rdfs_properties,
              rev_prop: "rdfs_properties"
            },
            datatypes: {
              heading:  "# Datatype definitions\n",
              bucket:   rdfs_datatypes,
              rev_prop: "rdfs_datatypes"
            },
            other: {
              heading:  "# Other definitions\n",
              bucket:   rdfs_instances,
              rev_prop: "rdfs_instances"
            }
          }.each do |key, hash|
            next unless cats[key]

            cats[key].each do |subject|
              node = {"@id" => subject.pname}
              po = {}

              # Group predicates with their values
              graph.query(subject: subject) do |statement|
                # Sanity check this, as these are set to an empty string if not defined.
                next if [RDF::RDFS.label, RDF::RDFS.comment].include?(statement.predicate) && statement.object.to_s.empty?
                po[statement.predicate] ||= []
                po[statement.predicate] << statement.object
              end

              next if po.empty?

              node['@type'] = po.delete(RDF.type).map {|t| jld_context.compact_iri(t, vocab: true)}

              po.each do |predicate, objects|
                term = jld_context.compact_iri(predicate, vocab: true)
                node[term] = if jld_context.container(term) == '@language'
                  lang_map = objects.inject({}) do |memo, o|
                    raise "Language-mapped term #{term} with non plain-literal #{o.inspect}" unless o.literal? && o.plain?
                    memo.merge(o.language.to_s => o.value)
                  end
                  # Don't use language map if there's only one entry with no language
                  lang_map = lang_map[""] if lang_map.keys == [""]
                  [lang_map]
                else
                  objects.map do |o|
                    expanded_value = jld_context.expand_value(term, o)
                    jld_context.compact_value(term, expanded_value)
                  end
                end
              end

              node.each do |property, values|
                case values.length
                when 0 then node.delete(property)
                when 1 then node[property] = values.first
                end
              end

              # Either set bucket from node, or append node to bucket
              if hash[:bucket].is_a?(Hash)
                hash[:bucket].merge!(node)
              else
                ontology[hash[:rev_prop]] ||= hash[:bucket]
                hash[:bucket] << node
              end
            end
          end

          # Serialize result
          {
            "@context" => context,
            "@graph" => ontology
          }.to_json(::JSON::LD::JSON_STATE)
        end
      rescue LoadError
        # No JSON-LD serialization unless gem loaded
      end

      ##
      # Generate HTML+RDFa representation, specific to vocabularies. This uses generated JSON-LD and a Haml template.
      #
      # @param [RDF::Queryable] :graph Optional graph, otherwise uses statements from vocabulary.
      # @param [Hash{#to_sym => String}] Prefixes to add to output
      # @param [String, Hash] jsonld
      #   If not provided, the `to_jsonld` method is used to generate it.
      # @param [String] template The path to a Haml or ERB template used to generate the output using the JSON-LD serialization
      # @return [String]
      def to_html(graph: nil, prefixes: nil, jsonld: nil, template: nil)
        # Find namespaces used in the vocabulary
        graph = RDF::Graph.new {|g| each_statement {|s| g << s}} if graph.nil? || graph.empty?

        # Get JSON as an object
        json = case jsonld
        when String then ::JSON.parse(File.read jsonld)
        when Hash   then jsonld
        else
          ::JSON.parse(to_jsonld(graph: graph, prefixes: prefixes))
        end
        raise "Expected JSON-LD data within the '@graph' key" unless json.has_key?('@graph')

        template ||= File.expand_path("../../../../etc/template.erb", __FILE__)

        prefixes = vocab_prefixes(graph).merge(prefixes || {})
        prefixes[:owl] = RDF::OWL.to_uri.to_s

        # Make sure ontology is typed
        json['@graph']['@type'] ||= ['owl:Ontology']

        jld_context = ::JSON::LD::Context.new.parse([json['@context'], json['@graph']['@context']])

        # Expand the JSON-LD to normalize accesses
        expanded = ::JSON::LD::API.expand(json).first
        expanded.delete('@reverse')

        # Re-compact keys
        expanded = expanded.inject({}) do |memo, (k, v)|
          term = RDF::Vocabulary.find_term(k)
          k = term.pname if term
          memo.merge(k => v)
        end

        # Normalize label accessors
        expanded['rdfs:label'] ||= %w(dc:title dc11:title skos:prefLabel).inject(nil) do |memo, key|
          memo || expanded[key]
        end || [{'@value' => json['@graph']['@id']}]
        %w(rdfs_classes rdfs_properties rdfs_datatypes rdfs_instances).each do |section|
          next unless json['@graph'][section]
          json['@graph'][section].each do |node|
            node['rdfs:label'] ||= %w(dc:title dc11:title skos:prefLabel).inject do |memo, key|
              memo || node[key]
            end || [{'@value' => node['@id']}]
          end
        end

        # Expand each part separately, as well.
        %w(rdfs_classes rdfs_properties rdfs_datatypes rdfs_instances).each do |section|
          next unless json['@graph'][section]
          expanded_section = ::JSON::LD::API.expand(json['@graph'][section], expandContext: jld_context)
          # Re-compact keys
          expanded[section] = expanded_section.map do |node|
            node.inject({}) do |memo, (k, v)|
              term = RDF::Vocabulary.find_term(k)
              k = term.pname if term
              memo.merge(k => v)
            end
          end
        end

        # Template invoked with expanded JSON-LD with outer object including `rdfs_classes`, `rdfs_properties`, and `rdf_instances` sections.
        case template
        when /.haml$/
          require 'haml'
          haml = Haml::Engine.new(File.read(template))
          haml.render(self, ont: expanded, context: json['@context'], prefixes: prefixes)
        when /.erb$/
          require 'erubis'
          eruby = Erubis::FastEruby.new(File.read(template))
          result = eruby.evaluate(binding: self, ont: expanded, context: json['@context'], prefixes: prefixes)
        else
          raise "Unknown template type #{template}. Should have '.erb' or '.haml' extension"
        end
      end

      ##
      # Create HTML for values (Helper method, needs to be public)
      def value_to_html(property, value, tag)
        value.map do |v|
          %(<#{tag} property="#{property}") +
          if v['@value']
            (v['@language'] ? %( lang="#{v['@language']}") : "") +
            (v['@type'] ? %( datatype="#{RDF::URI(v['@type']).pname}") : "") +
            %(>#{v['@value']})
          elsif v['@id']
            %( resource="#{RDF::URI(v['@id']).pname}">#{RDF::URI(v['@id']).pname})
          else
            raise "Unknown value type: #{v.inspect}, #{property}"
          end +
          %(</#{tag}>)
        end.join("\n")
      end
    private

      ##
      # Prefixes used in this vocabulary
      #
      # @param [RDF::Graph] graph
      # @return [Hash{Symbol => RDF::URI}]
      def vocab_prefixes(graph)
        vocabs = graph.
          terms.
          select(&:uri?).
          map {|u| RDF::Vocabulary.find(u)}.
          uniq.
          compact.
          sort_by(&:__prefix__)
        vocabs << RDF::XSD  # incase we need it for a literal

        # Generate prefix definitions
        vocabs.inject({}) do |memo, v|
          memo.merge(v.__prefix__ => v.to_uri)
        end
      end

      ##
      # Categorize each subject in the graph
      #
      # @param [RDF::Graph] graph
      # @return [Hash{RDF::URI => Symbol}]
      def subject_categories(graph)
        cats = {}
        categorized = {}
        uncategorized = {}
        graph.query(predicate: RDF.type) do |statement|
          # Only serialize statements that are in the defined vocabulary
          next unless statement.subject.uri? && statement.subject.start_with?(self.to_uri)
          case statement.object
          when RDF.Property,
               RDF::OWL.AnnotationProperty,
               RDF::OWL.DatatypeProperty,
               RDF::OWL.FunctionalProperty,
               RDF::OWL.ObjectProperty,
               RDF::OWL.OntologyProperty
            (cats[:properties] ||= []) << statement.subject unless categorized[statement.subject]
            categorized[statement.subject] = true
          when RDF::RDFS.Class, RDF::OWL.Class
            (cats[:classes] ||= []) << statement.subject unless categorized[statement.subject]
            categorized[statement.subject] = true
          when RDF::RDFS.Datatype, RDF::OWL.DataRange
            (cats[:datatypes] ||= []) << statement.subject unless categorized[statement.subject]
            categorized[statement.subject] = true
          when RDF::OWL.Ontology
            (cats[:ont] ||= []) << statement.subject unless categorized[statement.subject]
            categorized[statement.subject] = true
          else
            if statement.subject == self.to_uri
              (cats[:ont] ||= []) << statement.subject unless categorized[statement.subject]
              categorized[statement.subject] = true
            else
              uncategorized[statement.subject] = true
            end
          end
        end

        # Add all uncategorized subjects as :other
        uncat = (uncategorized.keys - categorized.keys)
        cats[:other] = uncat unless uncat.empty?

        cats
      end
    end

    module VocabFormatExtensions
      ##
      # Hash of CLI commands appropriate for this format
      # @return [Hash{Symbol => Lambda(Array, Hash)}]
      def cli_commands
        super.merge({
          "gen-vocab": {
            description: "Generate a vocabulary using a special serialization.",
            parse: false,  # Only parse if there are input files, otherwise, uses vocabulary
            help: "gen-vocab --uri <vocabulary-URI> [--output format ttl|jsonld|html] [options] [files]\nGenerate a vocabulary from repository using a special serialization.",
            lambda: ->(files, **options) do
              $stdout.puts "Generate Vocabulary"
              raise ArgumentError, "Must specify vocabulary URI" unless options[:base_uri]

              # Parse input graphs, if repository is not already created
              if RDF::CLI.repository.empty? && !files.empty?
                RDF::CLI.parse(files, **options) do |reader|
                  RDF::CLI.repository << reader
                end
              end

              # Lookup vocabulary, or generate a new vocabulary from this URI
              vocab = RDF::Vocabulary.find(options[:base_uri]) || begin
                raise ArgumentError, "Must specify vocabulary prefix if vocabulary not built-in" unless options[:prefix]
                RDF::Vocabulary.from_graph(RDF::CLI.repository, url: options[:base_uri], class_name: options[:prefix].to_s.upcase)
              end

              prefixes = {}
              prefixes[options[:prefix]] = options[:base_uri] if options[:prefix]
              out = options[:output] || $stdout
              case options[:output_format]
              when :ttl, nil then out.write vocab.to_ttl(graph: RDF::CLI.repository, prefixes: prefixes)
              when :jsonld then out.write vocab.to_jsonld(graph: RDF::CLI.repository, prefixes: prefixes)
              when :html then out.write vocab.to_html(graph: RDF::CLI.repository, prefixes: prefixes, template: options[:template])
              else
                # Use whatever writer we find
                writer = RDF::Writer.for(options[:output_format]) || RDF::NTriples::Writer
                writer.new(out, **options) do |w|
                  if RDF::CLI.repository.empty?
                    vocab.each_statement {|s| w << s}
                  else
                    w << RDF::CLI.repository
                  end
                end
              end
            end,
            options: [
              RDF::CLI::Option.new(
                symbol: :prefix,
                datatype: String,
                on: ["--prefix PREFIX"],
                description: "Prefix associated with vocabulary, if not built-in."),
              RDF::CLI::Option.new(
                symbol: :template,
                datatype: String,
                on: ["--template TEMPLATE"],
                description: "Path to local template for generating HTML, either Haml or ERB, depending on file extension.\n" +
                            "See https://github.com/ruby-rdf/rdf-vocab/tree/develop/etc for built-in templates."),
            ]
          }
        })
      end
    end

    # Add cli_commands as class method to RDF::Vocabulary::Format
    # TODO: in Ruby 2.0, `prepend` seems to be a private method of the class singleton; works okay elsewhere.
    Format.singleton_class.send(:prepend, VocabFormatExtensions)
  end
end