module Riddle
  class VersionError < StandardError;  end
  class ResponseError < StandardError; end
  
  # This class was heavily based on the existing Client API by Dmytro Shteflyuk
  # and Alexy Kovyrin. Their code worked fine, I just wanted something a bit
  # more Ruby-ish (ie. lowercase and underscored method names). I also have
  # used a few helper classes, just to neaten things up.
  #
  # Feel free to use it wherever. Send bug reports, patches, comments and
  # suggestions to pat at freelancing-gods dot com.
  #
  # Most properties of the client are accessible through attribute accessors,
  # and where relevant use symboles instead of the long constants common in
  # other clients.
  # Some examples:
  #
  #   client.sort_mode  = :extended
  #   client.sort_by    = "birthday DESC"
  #   client.match_mode = :extended
  #
  # To add a filter, you will need to create a Filter object:
  #
  #   client.filters << Riddle::Client::Filter.new("birthday",
  #     Time.at(1975, 1, 1).to_i..Time.at(1985, 1, 1).to_i, false)
  #
  class Client
    Commands = {
      :search   => 0, # SEARCHD_COMMAND_SEARCH
      :excerpt  => 1, # SEARCHD_COMMAND_EXCERPT
      :update   => 2, # SEARCHD_COMMAND_UPDATE
      :keywords => 3  # SEARCHD_COMMAND_KEYWORDS
    }
    
    Versions = {
      :search   => 0x113, # VER_COMMAND_SEARCH
      :excerpt  => 0x100, # VER_COMMAND_EXCERPT
      :update   => 0x101, # VER_COMMAND_UPDATE
      :keywords => 0x100  # VER_COMMAND_KEYWORDS
    }
    
    Statuses = {
      :ok      => 0, # SEARCHD_OK
      :error   => 1, # SEARCHD_ERROR
      :retry   => 2, # SEARCHD_RETRY
      :warning => 3  # SEARCHD_WARNING
    }
    
    MatchModes = {
      :all        => 0, # SPH_MATCH_ALL
      :any        => 1, # SPH_MATCH_ANY
      :phrase     => 2, # SPH_MATCH_PHRASE
      :boolean    => 3, # SPH_MATCH_BOOLEAN
      :extended   => 4, # SPH_MATCH_EXTENDED
      :fullscan   => 5, # SPH_MATCH_FULLSCAN
      :extended2  => 6  # SPH_MATCH_EXTENDED2
    }
    
    RankModes = {
      :proximity_bm25 => 0, # SPH_RANK_PROXIMITY_BM25
      :bm25           => 1, # SPH_RANK_BM25
      :none           => 2, # SPH_RANK_NONE
      :wordcount      => 3  # SPH_RANK_WORDCOUNT
    }
    
    SortModes = {
      :relevance     => 0, # SPH_SORT_RELEVANCE
      :attr_desc     => 1, # SPH_SORT_ATTR_DESC
      :attr_asc      => 2, # SPH_SORT_ATTR_ASC
      :time_segments => 3, # SPH_SORT_TIME_SEGMENTS
      :extended      => 4, # SPH_SORT_EXTENDED
      :expr          => 5  # SPH_SORT_EXPR
    }
    
    AttributeTypes = {
      :integer    => 1, # SPH_ATTR_INTEGER
      :timestamp  => 2, # SPH_ATTR_TIMESTAMP
      :ordinal    => 3, # SPH_ATTR_ORDINAL
      :bool       => 4, # SPH_ATTR_BOOL
      :float      => 5, # SPH_ATTR_FLOAT
      :multi      => 0x40000000 # SPH_ATTR_MULTI
    }
    
    GroupFunctions = {
      :day      => 0, # SPH_GROUPBY_DAY
      :week     => 1, # SPH_GROUPBY_WEEK
      :month    => 2, # SPH_GROUPBY_MONTH
      :year     => 3, # SPH_GROUPBY_YEAR
      :attr     => 4, # SPH_GROUPBY_ATTR
      :attrpair => 5  # SPH_GROUPBY_ATTRPAIR
    }
    
    FilterTypes = {
      :values       => 0, # SPH_FILTER_VALUES
      :range        => 1, # SPH_FILTER_RANGE
      :float_range  => 2  # SPH_FILTER_FLOATRANGE
    }
    
    attr_accessor :server, :port, :offset, :limit, :max_matches,
      :match_mode, :sort_mode, :sort_by, :weights, :id_range, :filters,
      :group_by, :group_function, :group_clause, :group_distinct, :cut_off,
      :retry_count, :retry_delay, :anchor, :index_weights, :rank_mode,
      :max_query_time, :field_weights, :timeout
    attr_reader :queue
    
    # Can instantiate with a specific server and port - otherwise it assumes
    # defaults of localhost and 3312 respectively. All other settings can be
    # accessed and changed via the attribute accessors.
    def initialize(server=nil, port=nil)
      @server = server || "localhost"
      @port   = port   || 3312
      
      # defaults
      @offset         = 0
      @limit          = 20
      @max_matches    = 1000
      @match_mode     = :all
      @sort_mode      = :relevance
      @sort_by        = ''
      @weights        = []
      @id_range       = 0..0
      @filters        = []
      @group_by       = ''
      @group_function = :day
      @group_clause   = '@group desc'
      @group_distinct = ''
      @cut_off        = 0
      @retry_count    = 0
      @retry_delay    = 0
      @anchor         = {}
      # string keys are index names, integer values are weightings
      @index_weights  = {}
      @rank_mode      = :proximity_bm25
      @max_query_time = 0
      # string keys are field names, integer values are weightings
      @field_weights  = {}
      @timeout        = 0
      
      @queue = []
    end
    
    # Reset attributes and settings to defaults.
    def reset
      # defaults
      @offset         = 0
      @limit          = 20
      @max_matches    = 1000
      @match_mode     = :all
      @sort_mode      = :relevance
      @sort_by        = ''
      @weights        = []
      @id_range       = 0..0
      @filters        = []
      @group_by       = ''
      @group_function = :day
      @group_clause   = '@group desc'
      @group_distinct = ''
      @cut_off        = 0
      @retry_count    = 0
      @retry_delay    = 0
      @anchor         = {}
      # string keys are index names, integer values are weightings
      @index_weights  = {}
      @rank_mode      = :proximity_bm25
      @max_query_time = 0
      # string keys are field names, integer values are weightings
      @field_weights  = {}
      @timeout        = 0
    end
    
    # Set the geo-anchor point - with the names of the attributes that contain
    # the latitude and longitude (in radians), and the reference position.
    # Note that for geocoding to work properly, you must also set
    # match_mode to :extended. To sort results by distance, you will
    # need to set sort_mode to '@geodist asc' for example. Sphinx
    # expects latitude and longitude to be returned from you SQL source
    # in radians.
    #
    # Example:
    #   client.set_anchor('lat', -0.6591741, 'long', 2.530770)
    #
    def set_anchor(lat_attr, lat, long_attr, long)
      @anchor = {
        :latitude_attribute   => lat_attr,
        :latitude             => lat,
        :longitude_attribute  => long_attr,
        :longitude            => long
      }
    end
    
    # Append a query to the queue. This uses the same parameters as the query
    # method.
    def append_query(search, index = '*', comments = '')
      @queue << query_message(search, index, comments)
    end
    
    # Run all the queries currently in the queue. This will return an array of
    # results hashes.
    def run
      response = Response.new request(:search, @queue)
      
      results = @queue.collect do
        result = {
          :matches         => [],
          :fields          => [],
          :attributes      => {},
          :attribute_names => [],
          :words           => {}
        }

        result[:status] = response.next_int
        case result[:status]
        when Statuses[:warning]
          result[:warning] = response.next
        when Statuses[:error]
          result[:error] = response.next
          next result
        end
        
        result[:fields] = response.next_array

        attributes = response.next_int
        for i in 0...attributes
          attribute_name = response.next
          type           = response.next_int

          result[:attributes][attribute_name] = type
          result[:attribute_names] << attribute_name
        end

        matches   = response.next_int
        is_64_bit = response.next_int
        for i in 0...matches
          doc = is_64_bit > 0 ? response.next_64bit_int : response.next_int
          weight = response.next_int

          result[:matches] << {:doc => doc, :weight => weight, :index => i, :attributes => {}}
          result[:attribute_names].each do |attr|
            result[:matches].last[:attributes][attr] = attribute_from_type(
              result[:attributes][attr], response
            )
          end
        end

        result[:total] = response.next_int.to_i || 0
        result[:total_found] = response.next_int.to_i || 0
        result[:time] = ('%.3f' % (response.next_int / 1000.0)).to_f || 0.0

        words = response.next_int
        for i in 0...words
          word = response.next
          docs = response.next_int
          hits = response.next_int
          result[:words][word] = {:docs => docs, :hits => hits}
        end

        result
      end
      
      @queue.clear
      results
    end
    
    # Query the Sphinx daemon - defaulting to all indexes, but you can specify
    # a specific one if you wish. The search parameter should be a string
    # following Sphinx's expectations.
    #
    # The object returned from this method is a hash with the following keys:
    # 
    # * :matches
    # * :fields
    # * :attributes
    # * :attribute_names
    # * :words
    # * :total
    # * :total_found
    # * :time
    # * :status
    # * :warning (if appropriate)
    # * :error (if appropriate)
    #
    # The key <tt>:matches</tt> returns an array of hashes - the actual search
    # results. Each hash has the document id (<tt>:doc</tt>), the result 
    # weighting (<tt>:weight</tt>), and a hash of the attributes for the
    # document (<tt>:attributes</tt>).
    # 
    # The <tt>:fields</tt> and <tt>:attribute_names</tt> keys return list of
    # fields and attributes for the documents. The key <tt>:attributes</tt>
    # will return a hash of attribute name and type pairs, and <tt>:words</tt>
    # returns a hash of hashes representing the words from the search, with the
    # number of documents and hits for each, along the lines of:
    # 
    #   results[:words]["Pat"] #=> {:docs => 12, :hits => 15}
    # 
    # <tt>:total</tt>, <tt>:total_found</tt> and <tt>:time</tt> return the
    # number of matches available, the total number of matches (which may be
    # greater than the maximum available, depending on the number of matches
    # and your sphinx configuration), and the time in milliseconds that the
    # query took to run.
    # 
    # <tt>:status</tt> is the error code for the query - and if there was a
    # related warning, it will be under the <tt>:warning</tt> key. Fatal errors
    # will be described under <tt>:error</tt>.
    #
    def query(search, index = '*', comments = '')
      @queue << query_message(search, index, comments)
      self.run.first
    end
    
    # Build excerpts from search terms (the +words+) and the text of documents. Excerpts are bodies of text that have the +words+ highlighted.
    # They may also be abbreviated to fit within a word limit.
    #
    # As part of the options hash, you will need to
    # define:
    # * :docs
    # * :words
    # * :index
    #
    # Optional settings include:
    # * :before_match (defaults to <span class="match">)
    # * :after_match (defaults to </span>)
    # * :chunk_separator (defaults to ' &#8230; ' - which is an HTML ellipsis)
    # * :limit (defaults to 256)
    # * :around (defaults to 5)
    # * :exact_phrase (defaults to false)
    # * :single_passage (defaults to false)
    #
    # The defaults differ from the official PHP client, as I've opted for
    # semantic HTML markup.
    #
    # Example:
    #
    #   client.excerpts(:docs => ["Pat Allan, Pat Cash"], :words => 'Pat', :index => 'pats')
    #   #=> ["<span class=\"match\">Pat</span> Allan, <span class=\"match\">Pat</span> Cash"]
    #
    #   lorem_lipsum = "Lorem ipsum dolor..."
    #
    #   client.excerpts(:docs => ["Pat Allan, #{lorem_lipsum} Pat Cash"], :words => 'Pat', :index => 'pats')
    #   #=> ["<span class=\"match\">Pat</span> Allan, Lorem ipsum dolor sit amet, consectetur adipisicing
    #          elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua &#8230; . Excepteur 
    #          sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est 
    #          laborum. <span class=\"match\">Pat</span> Cash"]  
    #
    # Workflow:
    #
    # Excerpt creation is completely isolated from searching the index. The nominated index is only used to 
    # discover encoding and charset information.
    #
    # Therefore, the workflow goes:
    #
    # 1. Do the sphinx query.
    # 2. Fetch the documents found by sphinx from their repositories.
    # 3. Pass the documents' text to +excerpts+ for marking up of matched terms.
    #
    def excerpts(options = {})
      options[:index]           ||= '*'
      options[:before_match]    ||= '<span class="match">'
      options[:after_match]     ||= '</span>'
      options[:chunk_separator] ||= ' &#8230; ' # ellipsis
      options[:limit]           ||= 256
      options[:around]          ||= 5
      options[:exact_phrase]    ||= false
      options[:single_passage]  ||= false
      
      response = Response.new request(:excerpt, excerpts_message(options))
      
      options[:docs].collect { response.next }
    end
    
    # Update attributes - first parameter is the relevant index, second is an
    # array of attributes to be updated, and the third is a hash, where the
    # keys are the document ids, and the values are arrays with the attribute
    # values - in the same order as the second parameter.
    #
    # Example:
    # 
    #   client.update('people', ['birthday'], {1 => [Time.at(1982, 20, 8).to_i]})
    # 
    def update(index, attributes, values_by_doc)
      response = Response.new request(
        :update,
        update_message(index, attributes, values_by_doc)
      )
      
      response.next_int
    end
    
    # Generates a keyword list for a given query. Each keyword is represented
    # by a hash, with keys :tokenised and :normalised. If return_hits is set to
    # true it will also report on the number of hits and documents for each
    # keyword (see :hits and :docs keys respectively).
    def keywords(query, index, return_hits = false)
      response = Response.new request(
        :keywords,
        keywords_message(query, index, return_hits)
      )
      
      (0...response.next_int).collect do
        hash = {}
        hash[:tokenised]  = response.next
        hash[:normalised] = response.next
        
        if return_hits
          hash[:docs] = response.next_int
          hash[:hits] = response.next_int
        end
        
        hash
      end
    end
    
    private
    
    # Connects to the Sphinx daemon, and yields a socket to use. The socket is
    # closed at the end of the block.
    def connect(&block)
      socket = nil
      if @timeout == 0
        socket = initialise_connection
      else
        begin
          Timeout.timeout(@timeout) { socket = initialise_connection }
        rescue Timeout::Error
          raise Riddle::ConnectionError,
            "Connection to #{@server} on #{@port} timed out after #{@timeout} seconds"
        end
      end
      
      begin
        yield socket
      ensure
        socket.close
      end
    end
    
    def initialise_connection
      socket = TCPSocket.new @server, @port
      
      # Checking version
      version = socket.recv(4).unpack('N*').first
      if version < 1
        socket.close
        raise VersionError, "Can only connect to searchd version 1.0 or better, not version #{version}"
      end
      
      # Send version
      socket.send [1].pack('N'), 0
      
      socket
    end
    
    # Send a collection of messages, for a command type (eg, search, excerpts,
    # update), to the Sphinx daemon.
    def request(command, messages)
      response = ""
      status   = -1
      version  = 0
      length   = 0
      message  = Array(messages).join("")
          
      connect do |socket|
        case command
        when :search
          # Message length is +4 to account for the following count value for
          # the number of messages (well, that's what I'm assuming).
          socket.send [
            Commands[command], Versions[command],
            4+message.length,  messages.length
          ].pack("nnNN") + message, 0
        else
          socket.send [
            Commands[command], Versions[command], message.length
          ].pack("nnN") + message, 0
        end
        
        header = socket.recv(8)
        status, version, length = header.unpack('n2N')
        
        while response.length < length
          part = socket.recv(length - response.length)
          response << part if part
        end
      end
      
      if response.empty? || response.length != length
        raise ResponseError, "No response from searchd (status: #{status}, version: #{version})"
      end
      
      case status
      when Statuses[:ok]
        if version < Versions[command]
          puts format("searchd command v.%d.%d older than client (v.%d.%d)",
            version >> 8, version & 0xff,
            Versions[command] >> 8, Versions[command] & 0xff)
        end
        response
      when Statuses[:warning]
        length = response[0, 4].unpack('N*').first
        puts response[4, length]
        response[4 + length, response.length - 4 - length]
      when Statuses[:error], Statuses[:retry]
        raise ResponseError, "searchd error (status: #{status}): #{response[4, response.length - 4]}"
      else
        raise ResponseError, "Unknown searchd error (status: #{status})"
      end
    end
    
    # Generation of the message to send to Sphinx for a search.
    def query_message(search, index, comments = '')
      message = Message.new
      
      # Mode, Limits, Sort Mode
      message.append_ints @offset, @limit, MatchModes[@match_mode],
        RankModes[@rank_mode], SortModes[@sort_mode]
      message.append_string @sort_by
      
      # Query
      message.append_string search
      
      # Weights
      message.append_int @weights.length
      message.append_ints *@weights
      
      # Index
      message.append_string index
      
      # ID Range
      message.append_int 1
      message.append_64bit_ints @id_range.first, @id_range.last
      
      # Filters
      message.append_int @filters.length
      @filters.each { |filter| message.append filter.query_message }
      
      # Grouping
      message.append_int GroupFunctions[@group_function]
      message.append_string @group_by
      message.append_int @max_matches
      message.append_string @group_clause
      message.append_ints @cut_off, @retry_count, @retry_delay
      message.append_string @group_distinct
      
      # Anchor Point
      if @anchor.empty?
        message.append_int 0
      else
        message.append_int 1
        message.append_string @anchor[:latitude_attribute]
        message.append_string @anchor[:longitude_attribute]
        message.append_floats @anchor[:latitude], @anchor[:longitude]
      end
      
      # Per Index Weights
      message.append_int @index_weights.length
      @index_weights.each do |key,val|
        message.append_string key.to_s
        message.append_int val
      end
      
      # Max Query Time
      message.append_int @max_query_time
      
      # Per Field Weights
      message.append_int @field_weights.length
      @field_weights.each do |key,val|
        message.append_string key.to_s
        message.append_int val
      end
      
      message.append_string comments
      
      message.to_s
    end
    
    # Generation of the message to send to Sphinx for an excerpts request.
    def excerpts_message(options)
      message = Message.new
      
      flags = 1
      flags |= 2  if options[:exact_phrase]
      flags |= 4  if options[:single_passage]
      flags |= 8  if options[:use_boundaries]
      flags |= 16 if options[:weight_order]
      
      message.append [0, flags].pack('N2') # 0 = mode
      message.append_string options[:index]
      message.append_string options[:words]
      
      # options
      message.append_string options[:before_match]
      message.append_string options[:after_match]
      message.append_string options[:chunk_separator]
      message.append_ints options[:limit], options[:around]
      
      message.append_array options[:docs]
      
      message.to_s
    end
    
    # Generation of the message to send to Sphinx to update attributes of a
    # document.
    def update_message(index, attributes, values_by_doc)
      message = Message.new
      
      message.append_string index
      message.append_array attributes
      
      message.append_int values_by_doc.length
      values_by_doc.each do |key,values|
        message.append_64bit_int key # document ID
        message.append_ints *values # array of new values (integers)
      end
      
      message.to_s
    end
    
    # Generates the simple message to send to the daemon for a keywords request.
    def keywords_message(query, index, return_hits)
      message = Message.new
      
      message.append_string query
      message.append_string index
      message.append_int return_hits ? 1 : 0
      
      message.to_s
    end
    
    def attribute_from_type(type, response)
      type -= AttributeTypes[:multi] if is_multi = type > AttributeTypes[:multi]
      
      case type
      when AttributeTypes[:float]
        is_multi ? response.next_float_array : response.next_float
      else
        is_multi ? response.next_int_array   : response.next_int
      end
    end
  end
end