sequel 2.2.0 → 2.3.0

data/lib/sequel_core/dataset/callback.rb

@@ -0,0 +1,13 @@
+module Sequel
+  class Dataset
+    private
+    # This is run inside .all, after all
+    # of the records have been loaded
+    # via .each, but before any block passed
+    # to all is called.  It is called with
+    # a single argument, an array of all
+    # returned records.
+    def post_load(all_records)
+    end
+  end
+end

data/lib/sequel_core/dataset/convenience.rb

@@ -0,0 +1,245 @@
+module Sequel
+  class Dataset
+    COMMA_SEPARATOR = ', '.freeze
+    COUNT_OF_ALL_AS_COUNT = :count['*'.lit].as(:count)
+
+    # Returns the first record matching the conditions.
+    def [](*conditions)
+      first(*conditions)
+    end
+
+    # Update all records matching the conditions
+    # with the values specified.
+    def []=(conditions, values)
+      filter(conditions).update(values)
+    end
+
+    # Returns the average value for the given column.
+    def avg(column)
+      get(:avg[column])
+    end
+    
+    # Returns true if no records exists in the dataset
+    def empty?
+      get(1).nil?
+    end
+    
+    # Returns the first record in the dataset. If a numeric argument is
+    # given, it is interpreted as a limit, and then returns all 
+    # matching records up to that limit.  If no argument is passed,
+    # it returns the first matching record.  If any other type of
+    # argument(s) is passed, it is given to filter and the
+    # first matching record is returned. If a block is given, it is used
+    # to filter the dataset before returning anything.
+    #
+    # Examples:
+    # 
+    #   ds.first => {:id=>7}
+    #   ds.first(2) => [{:id=>6}, {:id=>4}]
+    #   ds.order(:id).first(2) => [{:id=>1}, {:id=>2}]
+    #   ds.first(:id=>2) => {:id=>2}
+    #   ds.first("id = 3") => {:id=>3}
+    #   ds.first("id = ?", 4) => {:id=>4}
+    #   ds.first{:id > 2} => {:id=>5}
+    #   ds.order(:id).first{:id > 2} => {:id=>3}
+    #   ds.first{:id > 2} => {:id=>5}
+    #   ds.first("id > ?", 4){:id < 6) => {:id=>5}
+    #   ds.order(:id).first(2){:id < 2} => [{:id=>1}]
+    def first(*args, &block)
+      ds = block ? filter(&block) : self
+
+      if args.empty?
+        ds.single_record
+      else
+        args = (args.size == 1) ? args.first : args
+        if Integer === args
+          ds.limit(args).all
+        else
+          ds.filter(args).single_record
+        end
+      end
+    end
+
+    # Return the column value for the first matching record in the dataset.
+    def get(column)
+      select(column).single_value
+    end
+
+    # Returns a dataset grouped by the given column with count by group.
+    def group_and_count(*columns)
+      group(*columns).select(*(columns + [COUNT_OF_ALL_AS_COUNT])).order(:count)
+    end
+    
+    # Returns the interval between minimum and maximum values for the given 
+    # column.
+    def interval(column)
+      get("(max(#{literal(column)}) - min(#{literal(column)}))".lit)
+    end
+
+    # Reverses the order and then runs first.  Note that this
+    # will not necessarily give you the last record in the dataset,
+    # unless you have an unambiguous order.  If there is not
+    # currently an order for this dataset, raises an Error.
+    def last(*args, &block)
+      raise(Error, 'No order specified') unless @opts[:order]
+      reverse.first(*args, &block)
+    end
+    
+    # Maps column values for each record in the dataset (if a column name is
+    # given), or performs the stock mapping functionality of Enumerable.
+    def map(column_name = nil, &block)
+      if column_name
+        super() {|r| r[column_name]}
+      else
+        super(&block)
+      end
+    end
+
+    # Returns the maximum value for the given column.
+    def max(column)
+      get(:max[column])
+    end
+
+    # Returns the minimum value for the given column.
+    def min(column)
+      get(:min[column])
+    end
+
+    # Inserts multiple records into the associated table. This method can be
+    # to efficiently insert a large amounts of records into a table. Inserts
+    # are automatically wrapped in a transaction.
+    # 
+    # This method should be called with a columns array and an array of value arrays:
+    #
+    #   dataset.multi_insert([:x, :y], [[1, 2], [3, 4]])
+    #
+    # This method can also be called with an array of hashes:
+    # 
+    #   dataset.multi_insert({:x => 1}, {:x => 2})
+    #
+    # Be aware that all hashes should have the same keys if you use this calling method,
+    # otherwise some columns could be missed or set to null instead of to default
+    # values.
+    #
+    # The method also accepts a :slice or :commit_every option that specifies
+    # the number of records to insert per transaction. This is useful especially
+    # when inserting a large number of records, e.g.:
+    #
+    #   # this will commit every 50 records
+    #   dataset.multi_insert(lots_of_records, :slice => 50)
+    def multi_insert(*args)
+      if args.empty?
+        return
+      elsif args[0].is_a?(Array) && args[1].is_a?(Array)
+        columns, values, opts = *args
+      elsif args[0].is_a?(Array) && args[1].is_a?(Dataset)
+        table = @opts[:from].first
+        columns, dataset = *args
+        sql = "INSERT INTO #{quote_identifier(table)} #{literal(columns)} VALUES #{literal(dataset)}"
+        return @db.transaction {@db.execute_dui sql}
+      else
+        # we assume that an array of hashes is given
+        hashes, opts = *args
+        return if hashes.empty?
+        columns = hashes.first.keys
+        # convert the hashes into arrays
+        values = hashes.map {|h| columns.map {|c| h[c]}}
+      end
+      # make sure there's work to do
+      return if columns.empty? || values.empty?
+      
+      slice_size = opts && (opts[:commit_every] || opts[:slice])
+      
+      if slice_size
+        values.each_slice(slice_size) do |slice|
+          statements = multi_insert_sql(columns, slice)
+          @db.transaction {statements.each {|st| @db.execute_dui(st)}}
+        end
+      else
+        statements = multi_insert_sql(columns, values)
+        @db.transaction {statements.each {|st| @db.execute_dui(st)}}
+      end
+    end
+    alias_method :import, :multi_insert
+    
+    # Pretty prints the records in the dataset as plain-text table.
+    def print(*cols)
+      Sequel::PrettyTable.print(naked.all, cols.empty? ? columns : cols)
+    end
+    
+    # Returns a Range object made from the minimum and maximum values for the
+    # given column.
+    def range(column)
+      if r = select(:min[column].as(:v1), :max[column].as(:v2)).first
+        (r[:v1]..r[:v2])
+      end
+    end
+    
+    # Returns the first record in the dataset.
+    def single_record(opts = nil)
+      each((opts||{}).merge(:limit=>1)){|r| return r}
+      nil
+    end
+
+    # Returns the first value of the first record in the dataset.
+    # Returns nil if dataset is empty.
+    def single_value(opts = nil)
+      if r = naked.single_record(opts)
+        r.values.first
+      end
+    end
+    
+    # Returns the sum for the given column.
+    def sum(column)
+      get(:sum[column])
+    end
+
+    # Returns true if the table exists.  Will raise an error
+    # if the dataset has fixed SQL or selects from another dataset
+    # or more than one table.
+    def table_exists?
+      if @opts[:sql]
+        raise Sequel::Error, "this dataset has fixed SQL"
+      end
+      
+      if @opts[:from].size != 1
+        raise Sequel::Error, "this dataset selects from multiple sources"
+      end
+      
+      t = @opts[:from].first
+      if t.is_a?(Dataset)
+        raise Sequel::Error, "this dataset selects from a sub query"
+      end
+      
+      @db.table_exists?(t.to_sym)
+    end
+
+    # Returns a string in CSV format containing the dataset records. By 
+    # default the CSV representation includes the column titles in the
+    # first line. You can turn that off by passing false as the 
+    # include_column_titles argument.
+    #
+    # This does not use a CSV library or handle quoting of values in
+    # any way.  If any values in any of the rows could include commas or line
+    # endings, you probably shouldn't use this.
+    def to_csv(include_column_titles = true)
+      n = naked
+      cols = n.columns
+      csv = ''
+      csv << "#{cols.join(COMMA_SEPARATOR)}\r\n" if include_column_titles
+      n.each{|r| csv << "#{cols.collect{|c| r[c]}.join(COMMA_SEPARATOR)}\r\n"}
+      csv
+    end
+    
+    # Returns a hash with one column used as key and another used as value.
+    # If rows have duplicate values for the key column, the latter row(s)
+    # will overwrite the value of the previous row(s). If the value_column
+    # is not given or nil, uses the entire hash as the value.
+    def to_hash(key_column, value_column = nil)
+      inject({}) do |m, r|
+        m[r[key_column]] = value_column ? r[value_column] : r
+        m
+      end
+    end
+  end
+end

data/lib/sequel_core/dataset/pagination.rb

@@ -0,0 +1,96 @@
+module Sequel
+  class Dataset
+    # Returns a paginated dataset. The returned dataset is limited to
+    # the page size at the correct offset, and extended with the Pagination
+    # module.  If a record count is not provided, does a count of total
+    # number of records for this dataset.
+    def paginate(page_no, page_size, record_count=nil)
+      raise(Error, "You cannot paginate a dataset that already has a limit") if @opts[:limit]
+      paginated = limit(page_size, (page_no - 1) * page_size)
+      paginated.extend(Pagination)
+      paginated.set_pagination_info(page_no, page_size, record_count || count)
+    end
+      
+    # Yields a paginated dataset for each page and returns the receiver. Does
+    # a count to find the total number of records for this dataset.
+    def each_page(page_size, &block)
+      raise(Error, "You cannot paginate a dataset that already has a limit") if @opts[:limit]
+      record_count = count
+      total_pages = (record_count / page_size.to_f).ceil
+      (1..total_pages).each{|page_no| yield paginate(page_no, page_size, record_count)}
+      self
+    end
+
+    # Holds methods that only relate to paginated datasets. Paginated dataset
+    # have pages starting at 1 (page 1 is offset 0, page 1 is offset page_size).
+    module Pagination
+      # The number of records per page (the final page may have fewer than
+      # this number of records).
+      attr_accessor :page_size
+
+      # The number of pages in the dataset before pagination, of which
+      # this paginated dataset is one.
+      attr_accessor :page_count
+
+      # The current page of the dataset, starting at 1 and not 0.
+      attr_accessor :current_page
+
+      # The total number of records in the dataset before pagination.
+      attr_accessor :pagination_record_count
+
+      # Returns the record range for the current page
+      def current_page_record_range
+        return (0..0) if @current_page > @page_count
+        
+        a = 1 + (@current_page - 1) * @page_size
+        b = a + @page_size - 1
+        b = @pagination_record_count if b > @pagination_record_count
+        a..b
+      end
+
+      # Returns the number of records in the current page
+      def current_page_record_count
+        return 0 if @current_page > @page_count
+        
+        a = 1 + (@current_page - 1) * @page_size
+        b = a + @page_size - 1
+        b = @pagination_record_count if b > @pagination_record_count
+        b - a + 1
+      end
+
+      # Returns true if the current page is the first page
+      def first_page?
+        @current_page == 1
+      end
+
+      # Returns true if the current page is the last page
+      def last_page?
+        @current_page == @page_count
+      end
+
+      # Returns the next page number or nil if the current page is the last page
+      def next_page
+        current_page < page_count ? (current_page + 1) : nil
+      end
+      
+      # Returns the page range
+      def page_range
+        1..page_count
+      end
+      
+      # Returns the previous page number or nil if the current page is the first
+      def prev_page
+        current_page > 1 ? (current_page - 1) : nil
+      end
+
+      # Sets the pagination info for this paginated dataset, and returns self.
+      def set_pagination_info(page_no, page_size, record_count)
+        @current_page = page_no
+        @page_size = page_size
+        @pagination_record_count = record_count
+        @page_count = (record_count / page_size.to_f).ceil
+        self
+      end
+    end
+  end
+end

data/lib/sequel_core/dataset/query.rb

@@ -0,0 +1,41 @@
+module Sequel
+  class Dataset
+    # Translates a query block into a dataset. Query blocks can be useful
+    # when expressing complex SELECT statements, e.g.:
+    #
+    #   dataset = DB[:items].query do
+    #     select :x, :y, :z
+    #     filter((:x > 1) & (:y > 2))
+    #     order :z.desc
+    #   end
+    #
+    # Which is the same as:
+    #
+    #  dataset = DB[:items].select(:x, :y, :z).filter((:x > 1) & (:y > 2)).order(:z.desc)
+    #
+    # Note that inside a call to query, you cannot call each, insert, update,
+    # or delete (or any method that calls those), or Sequel will raise an
+    # error.
+    def query(&block)
+      copy = clone({})
+      copy.extend(QueryBlockCopy)
+      copy.instance_eval(&block)
+      clone(copy.opts)
+    end
+
+    # Module used by Dataset#query that has the effect of making all
+    # dataset methods into !-style methods that modify the receiver.
+    module QueryBlockCopy
+      %w'each insert update delete'.each do |meth|
+        define_method(meth){|*args| raise Error, "##{meth} cannot be invoked inside a query block."}
+      end
+
+      # Merge the given options into the receiver's options and return the receiver
+      # instead of cloning the receiver.
+      def clone(opts = nil)
+        @opts.merge!(opts)
+        self
+      end
+    end
+  end
+end

data/lib/sequel_core/dataset/schema.rb

@@ -0,0 +1,15 @@
+module Sequel
+  class Dataset
+    # Creates a view in the database with the given named based
+    # on the current dataset.
+    def create_view(name)
+      @db.create_view(name, self)
+    end
+
+    # Creates or replaces a view in the database with the given
+    # named based on the current dataset.
+    def create_or_replace_view(name)
+      @db.create_or_replace_view(name, self)
+    end
+  end
+end

data/lib/sequel_core/dataset/sql.rb

@@ -0,0 +1,889 @@
+module Sequel
+  class Dataset
+    AND_SEPARATOR = " AND ".freeze
+    BOOL_FALSE = "'f'".freeze
+    BOOL_TRUE = "'t'".freeze
+    COLUMN_REF_RE1 = /\A([\w ]+)__([\w ]+)___([\w ]+)\z/.freeze
+    COLUMN_REF_RE2 = /\A([\w ]+)___([\w ]+)\z/.freeze
+    COLUMN_REF_RE3 = /\A([\w ]+)__([\w ]+)\z/.freeze
+    COUNT_FROM_SELF_OPTS = [:distinct, :group, :sql, :limit]
+    DATE_FORMAT = "DATE '%Y-%m-%d'".freeze
+    N_ARITY_OPERATORS = ::Sequel::SQL::ComplexExpression::N_ARITY_OPERATORS
+    NULL = "NULL".freeze
+    QUESTION_MARK = '?'.freeze
+    STOCK_COUNT_OPTS = {:select => ["COUNT(*)".lit], :order => nil}.freeze
+    TIMESTAMP_FORMAT = "TIMESTAMP '%Y-%m-%d %H:%M:%S'".freeze
+    TWO_ARITY_OPERATORS = ::Sequel::SQL::ComplexExpression::TWO_ARITY_OPERATORS
+    WILDCARD = '*'.freeze
+
+    # Adds an further filter to an existing filter using AND. If no filter 
+    # exists an error is raised. This method is identical to #filter except
+    # it expects an existing filter.
+    def and(*cond, &block)
+      raise(Error::NoExistingFilter, "No existing filter found.") unless @opts[:having] || @opts[:where]
+      filter(*cond, &block)
+    end
+
+    # SQL fragment for the aliased expression
+    def aliased_expression_sql(ae)
+      "#{literal(ae.expression)} AS #{quote_identifier(ae.aliaz)}"
+    end
+
+    # SQL fragment for specifying given CaseExpression.
+    def case_expression_sql(ce)
+      "(CASE #{ce.conditions.collect{|c,r| "WHEN #{literal(c)} THEN #{literal(r)} "}.join}ELSE #{literal(ce.default)} END)"
+    end
+
+    # SQL fragment for specifying all columns in a given table.
+    def column_all_sql(ca)
+      "#{quote_identifier(ca.table)}.*"
+    end
+
+    # SQL fragment for complex expressions
+    def complex_expression_sql(op, args)
+      case op
+      when *TWO_ARITY_OPERATORS
+        "(#{literal(args.at(0))} #{op} #{literal(args.at(1))})"
+      when *N_ARITY_OPERATORS
+        "(#{args.collect{|a| literal(a)}.join(" #{op} ")})"
+      when :NOT
+        "NOT #{literal(args.at(0))}"
+      when :NOOP
+        literal(args.at(0))
+      when :'B~'
+        "~#{literal(args.at(0))}"
+      else
+        raise(Sequel::Error, "invalid operator #{op}")
+      end
+    end
+
+    # Returns the number of records in the dataset.
+    def count
+      options_overlap(COUNT_FROM_SELF_OPTS) ? from_self.count : single_value(STOCK_COUNT_OPTS).to_i
+    end
+    alias_method :size, :count
+
+    # Formats a DELETE statement using the given options and dataset options.
+    # 
+    #   dataset.filter(:price >= 100).delete_sql #=>
+    #     "DELETE FROM items WHERE (price >= 100)"
+    def delete_sql(opts = nil)
+      opts = opts ? @opts.merge(opts) : @opts
+
+      if opts[:group]
+        raise Error::InvalidOperation, "Grouped datasets cannot be deleted from"
+      elsif opts[:from].is_a?(Array) && opts[:from].size > 1
+        raise Error::InvalidOperation, "Joined datasets cannot be deleted from"
+      end
+
+      sql = "DELETE FROM #{source_list(opts[:from])}"
+
+      if where = opts[:where]
+        sql << " WHERE #{literal(where)}"
+      end
+
+      sql
+    end
+
+    # Adds an EXCEPT clause using a second dataset object. If all is true the
+    # clause used is EXCEPT ALL, which may return duplicate rows.
+    #
+    #   DB[:items].except(DB[:other_items]).sql
+    #   #=> "SELECT * FROM items EXCEPT SELECT * FROM other_items"
+    def except(dataset, all = false)
+      clone(:except => dataset, :except_all => all)
+    end
+
+    # Performs the inverse of Dataset#filter.
+    #
+    #   dataset.exclude(:category => 'software').sql #=>
+    #     "SELECT * FROM items WHERE (category != 'software')"
+    def exclude(*cond, &block)
+      clause = (@opts[:having] ? :having : :where)
+      cond = cond.first if cond.size == 1
+      cond = cond.sql_or if (Hash === cond) || ((Array === cond) && (cond.all_two_pairs?))
+      cond = filter_expr(cond, &block)
+      cond = SQL::BooleanExpression.invert(cond)
+      cond = SQL::BooleanExpression.new(:AND, @opts[clause], cond) if @opts[clause]
+      clone(clause => cond)
+    end
+
+    # Returns an EXISTS clause for the dataset.
+    #
+    #   DB.select(1).where(DB[:items].exists).sql
+    #   #=> "SELECT 1 WHERE EXISTS (SELECT * FROM items)"
+    def exists(opts = nil)
+      "EXISTS (#{select_sql(opts)})"
+    end
+
+    # Returns a copy of the dataset with the given conditions imposed upon it.  
+    # If the query has been grouped, then the conditions are imposed in the 
+    # HAVING clause. If not, then they are imposed in the WHERE clause. Filter
+    # 
+    # filter accepts the following argument types:
+    #
+    # * Hash - list of equality expressions
+    # * Array - depends:
+    #   * If first member is a string, assumes the rest of the arguments
+    #     are parameters and interpolates them into the string.
+    #   * If all members are arrays of length two, treats the same way
+    #     as a hash, except it allows for duplicate keys to be
+    #     specified.
+    # * String - taken literally
+    # * Symbol - taken as a boolean column argument (e.g. WHERE active)
+    # * Sequel::SQL::BooleanExpression - an existing condition expression,
+    #   probably created using the Sequel blockless filter DSL.
+    #
+    # filter also takes a block, which should return one of the above argument
+    # types, and is treated the same way. If both a block and regular argument
+    # are provided, they get ANDed together.
+    #
+    # Examples:
+    #
+    #   dataset.filter(:id => 3).sql #=>
+    #     "SELECT * FROM items WHERE (id = 3)"
+    #   dataset.filter('price < ?', 100).sql #=>
+    #     "SELECT * FROM items WHERE price < 100"
+    #   dataset.filter([[:id, (1,2,3)], [:id, 0..10]]).sql #=>
+    #     "SELECT * FROM items WHERE ((id IN (1, 2, 3)) AND ((id >= 0) AND (id <= 10)))"
+    #   dataset.filter('price < 100').sql #=>
+    #     "SELECT * FROM items WHERE price < 100"
+    #   dataset.filter(:active).sql #=>
+    #     "SELECT * FROM items WHERE :active
+    #   dataset.filter(:price < 100).sql #=>
+    #     "SELECT * FROM items WHERE (price < 100)"
+    # 
+    # Multiple filter calls can be chained for scoping:
+    #
+    #   software = dataset.filter(:category => 'software')
+    #   software.filter(price < 100).sql #=>
+    #     "SELECT * FROM items WHERE ((category = 'software') AND (price < 100))"
+    #
+    # See doc/dataset_filters.rdoc for more examples and details.
+    def filter(*cond, &block)
+      clause = (@opts[:having] ? :having : :where)
+      cond = cond.first if cond.size == 1
+      cond = transform_save(cond) if @transform if cond.is_a?(Hash)
+      cond = filter_expr(cond, &block)
+      cond = SQL::BooleanExpression.new(:AND, @opts[clause], cond) if @opts[clause] && !@opts[clause].blank?
+      clone(clause => cond)
+    end
+    alias_method :where, :filter
+
+    # The first source (primary table) for this dataset.  If the dataset doesn't
+    # have a table, raises an error.  If the table is aliased, returns the actual
+    # table name, not the alias.
+    def first_source
+      source = @opts[:from]
+      if source.nil? || source.empty?
+        raise Error, 'No source specified for query'
+      end
+      case s = source.first
+      when Hash
+        s.values.first
+      when Symbol
+        sch, table, aliaz = split_symbol(s)
+        aliaz ? aliaz.to_sym : s
+      else
+        s
+      end
+    end
+
+    # Returns a copy of the dataset with the source changed.
+    def from(*source)
+      clone(:from => source)
+    end
+    
+    # Returns a dataset selecting from the current dataset.
+    #
+    #   ds = DB[:items].order(:name)
+    #   ds.sql #=> "SELECT * FROM items ORDER BY name"
+    #   ds.from_self.sql #=> "SELECT * FROM (SELECT * FROM items ORDER BY name)"
+    def from_self
+      fs = {}
+      @opts.keys.each{|k| fs[k] = nil} 
+      fs[:from] = [self]
+      clone(fs)
+    end
+
+    # SQL fragment specifying an SQL function call
+    def function_sql(f)
+      args = f.args
+      "#{f.f}#{args.empty? ? '()' : literal(args)}"
+    end
+
+    # Pattern match any of the columns to any of the terms.  The terms can be
+    # strings (which use LIKE) or regular expressions (which are only supported
+    # in some databases).  See Sequel::SQL::StringExpression.like.  Note that the
+    # total number of pattern matches will be cols.length * terms.length,
+    # which could cause performance issues.
+    def grep(cols, terms)
+      filter(SQL::BooleanExpression.new(:OR, *Array(cols).collect{|c| SQL::StringExpression.like(c, *terms)}))
+    end
+
+    # Returns a copy of the dataset with the results grouped by the value of 
+    # the given columns
+    def group(*columns)
+      clone(:group => columns)
+    end
+    alias_method :group_by, :group
+
+    # Returns a copy of the dataset with the having conditions changed. Raises 
+    # an error if the dataset has not been grouped. See also #filter.
+    def having(*cond, &block)
+      raise(Error::InvalidOperation, "Can only specify a HAVING clause on a grouped dataset") unless @opts[:group]
+      clone(:having=>{}).filter(*cond, &block)
+    end
+    
+    # Inserts multiple values. If a block is given it is invoked for each
+    # item in the given array before inserting it.  See #multi_insert as
+    # a possible faster version that inserts multiple records in one
+    # SQL statement.
+    def insert_multiple(array, &block)
+      if block
+        array.each {|i| insert(block[i])}
+      else
+        array.each {|i| insert(i)}
+      end
+    end
+
+    # Formats an INSERT statement using the given values. If a hash is given,
+    # the resulting statement includes column names. If no values are given, 
+    # the resulting statement includes a DEFAULT VALUES clause.
+    #
+    #   dataset.insert_sql() #=> 'INSERT INTO items DEFAULT VALUES'
+    #   dataset.insert_sql(1,2,3) #=> 'INSERT INTO items VALUES (1, 2, 3)'
+    #   dataset.insert_sql(:a => 1, :b => 2) #=>
+    #     'INSERT INTO items (a, b) VALUES (1, 2)'
+    def insert_sql(*values)
+      if values.empty?
+        insert_default_values_sql
+      else
+        values = values[0] if values.size == 1
+        
+        # if hash or array with keys we need to transform the values
+        if @transform && (values.is_a?(Hash) || (values.is_a?(Array) && values.keys))
+          values = transform_save(values)
+        end
+        from = source_list(@opts[:from])
+
+        case values
+        when Array
+          if values.empty?
+            insert_default_values_sql
+          else
+            "INSERT INTO #{from} VALUES #{literal(values)}"
+          end
+        when Hash
+          if values.empty?
+            insert_default_values_sql
+          else
+            fl, vl = [], []
+            values.each {|k, v| fl << literal(k.is_a?(String) ? k.to_sym : k); vl << literal(v)}
+            "INSERT INTO #{from} (#{fl.join(COMMA_SEPARATOR)}) VALUES (#{vl.join(COMMA_SEPARATOR)})"
+          end
+        when Dataset
+          "INSERT INTO #{from} #{literal(values)}"
+        else
+          if values.respond_to?(:values)
+            insert_sql(values.values)
+          else
+            "INSERT INTO #{from} VALUES (#{literal(values)})"
+          end
+        end
+      end
+    end
+    
+    # Adds an INTERSECT clause using a second dataset object. If all is true 
+    # the clause used is INTERSECT ALL, which may return duplicate rows.
+    #
+    #   DB[:items].intersect(DB[:other_items]).sql
+    #   #=> "SELECT * FROM items INTERSECT SELECT * FROM other_items"
+    def intersect(dataset, all = false)
+      clone(:intersect => dataset, :intersect_all => all)
+    end
+
+    # Inverts the current filter
+    #
+    #   dataset.filter(:category => 'software').invert.sql #=>
+    #     "SELECT * FROM items WHERE (category != 'software')"
+    def invert
+      having, where = @opts[:having], @opts[:where]
+      raise(Error, "No current filter") unless having || where
+      o = {}
+      o[:having] = SQL::BooleanExpression.invert(having) if having
+      o[:where] = SQL::BooleanExpression.invert(where) if where
+      clone(o)
+    end
+
+    # SQL fragment specifying an Irregular (cast/extract) SQL function call
+    def irregular_function_sql(f)
+      "#{f.f}(#{literal(f.arg1)} #{f.joiner} #{literal(f.arg2)})"
+    end
+
+    # SQL fragment specifying a JOIN clause without ON or USING.
+    def join_clause_sql(jc)
+      table = jc.table
+      table_alias = jc.table_alias
+      table_alias = nil if table == table_alias
+      " #{join_type_sql(jc.join_type)} #{table_ref(table)}" \
+        "#{" AS #{quote_identifier(jc.table_alias)}" if table_alias}"
+    end
+
+    # SQL fragment specifying a JOIN clause with ON.
+    def join_on_clause_sql(jc)
+      "#{join_clause_sql(jc)} ON #{literal(filter_expr(jc.on))}"
+    end
+
+    # SQL fragment specifying a JOIN clause with USING.
+    def join_using_clause_sql(jc)
+      "#{join_clause_sql(jc)} USING (#{column_list(jc.using)})"
+    end
+
+    # Returns a joined dataset.  Uses the following arguments:
+    #
+    # * type - The type of join to do (:inner, :left_outer, :right_outer, :full)
+    # * table - Depends on type:
+    #   * Dataset - a subselect is performed with an alias of tN for some value of N
+    #   * Model (or anything responding to :table_name) - table.table_name
+    #   * String, Symbol: table
+    # * expr - specifies conditions, depends on type:
+    #   * Hash, Array with all two pairs - Assumes key (1st arg) is column of joined table (unless already
+    #     qualified), and value (2nd arg) is column of the last joined or primary table.
+    #     To specify multiple conditions on a single joined table column, you must use an array.
+    #     Uses a JOIN with an ON clause.
+    #   * Array - If all members of the array are symbols, considers them as columns and 
+    #     uses a JOIN with a USING clause.  Most databases will remove duplicate columns from
+    #     the result set if this is used.
+    #   * nil - If a block is not given, doesn't use ON or USING, so the JOIN should be a NATURAL
+    #     or CROSS join. If a block is given, uses a ON clause based on the block, see below.
+    #   * Everything else - pretty much the same as a using the argument in a call to filter,
+    #     so strings are considered literal, symbols specify boolean columns, and blockless
+    #     filter expressions can be used. Uses a JOIN with an ON clause.
+    # * table_alias - the name of the table's alias when joining, necessary for joining
+    #   to the same table more than once.  No alias is used by default.
+    # * block - The block argument should only be given if a JOIN with an ON clause is used,
+    #   in which case it yields the table alias/name for the table currently being joined,
+    #   the table alias/name for the last joined (or first table), and an array of previous
+    #   SQL::JoinClause.
+    def join_table(type, table, expr=nil, table_alias=nil, &block)
+      if Dataset === table
+        if table_alias.nil?
+          table_alias_num = (@opts[:num_dataset_sources] || 0) + 1
+          table_alias = "t#{table_alias_num}"
+        end
+        table_name = table_alias
+      else
+        table = table.table_name if table.respond_to?(:table_name)
+        table_name = table_alias || table
+      end
+
+      join = if expr.nil? and !block_given?
+        SQL::JoinClause.new(type, table, table_alias)
+      elsif Array === expr and !expr.empty? and expr.all?{|x| Symbol === x}
+        raise(Sequel::Error, "can't use a block if providing an array of symbols as expr") if block_given?
+        SQL::JoinUsingClause.new(expr, type, table, table_alias)
+      else
+        last_alias = @opts[:last_joined_table] || first_source
+        if Hash === expr or (Array === expr and expr.all_two_pairs?)
+          expr = expr.collect do |k, v|
+            k = qualified_column_name(k, table_name) if k.is_a?(Symbol)
+            v = qualified_column_name(v, last_alias) if v.is_a?(Symbol)
+            [k,v]
+          end
+        end
+        if block_given?
+          expr2 = yield(table_name, last_alias, @opts[:join] || [])
+          expr = expr ? SQL::BooleanExpression.new(:AND, expr, expr2) : expr2
+        end
+        SQL::JoinOnClause.new(expr, type, table, table_alias)
+      end
+
+      opts = {:join => (@opts[:join] || []) + [join], :last_joined_table => table_name}
+      opts[:num_dataset_sources] = table_alias_num if table_alias_num
+      clone(opts)
+    end
+
+    # If given an integer, the dataset will contain only the first l results.
+    # If given a range, it will contain only those at offsets within that
+    # range. If a second argument is given, it is used as an offset.
+    def limit(l, o = nil)
+      return from_self.limit(l, o) if @opts[:sql]
+
+      if Range === l
+        o = l.first
+        l = l.interval + 1
+      end
+      l = l.to_i
+      raise(Error, 'Limits must be greater than or equal to 1') unless l >= 1
+      opts = {:limit => l}
+      if o
+        o = o.to_i
+        raise(Error, 'Offsets must be greater than or equal to 0') unless o >= 0
+        opts[:offset] = o
+      end
+      clone(opts)
+    end
+    
+    # Returns a literal representation of a value to be used as part
+    # of an SQL expression. 
+    # 
+    #   dataset.literal("abc'def\\") #=> "'abc''def\\\\'"
+    #   dataset.literal(:items__id) #=> "items.id"
+    #   dataset.literal([1, 2, 3]) => "(1, 2, 3)"
+    #   dataset.literal(DB[:items]) => "(SELECT * FROM items)"
+    #   dataset.literal(:x + 1 > :y) => "((x + 1) > y)"
+    #
+    # If an unsupported object is given, an exception is raised.
+    def literal(v)
+      case v
+      when LiteralString
+        v
+      when String
+        "'#{v.gsub(/\\/, "\\\\\\\\").gsub(/'/, "''")}'"
+      when Integer, Float
+        v.to_s
+      when BigDecimal
+        v.to_s("F")
+      when NilClass
+        NULL
+      when TrueClass
+        BOOL_TRUE
+      when FalseClass
+        BOOL_FALSE
+      when Symbol
+        symbol_to_column_ref(v)
+      when ::Sequel::SQL::Expression
+        v.to_s(self)
+      when Array
+        v.all_two_pairs? ? literal(v.sql_expr) : (v.empty? ? '(NULL)' : "(#{v.collect{|i| literal(i)}.join(COMMA_SEPARATOR)})")
+      when Hash
+        literal(v.sql_expr)
+      when Time, DateTime
+        v.strftime(TIMESTAMP_FORMAT)
+      when Date
+        v.strftime(DATE_FORMAT)
+      when Dataset
+        "(#{v.sql})"
+      else
+        raise Error, "can't express #{v.inspect} as a SQL literal"
+      end
+    end
+
+    # Returns an array of insert statements for inserting multiple records.
+    # This method is used by #multi_insert to format insert statements and
+    # expects a keys array and and an array of value arrays.
+    #
+    # This method should be overridden by descendants if the support
+    # inserting multiple records in a single SQL statement.
+    def multi_insert_sql(columns, values)
+      table = quote_identifier(@opts[:from].first)
+      columns = literal(columns)
+      values.map do |r|
+        "INSERT INTO #{table} #{columns} VALUES #{literal(r)}"
+      end
+    end
+    
+    # Adds an alternate filter to an existing filter using OR. If no filter 
+    # exists an error is raised.
+    def or(*cond, &block)
+      clause = (@opts[:having] ? :having : :where)
+      cond = cond.first if cond.size == 1
+      if @opts[clause]
+        clone(clause => SQL::BooleanExpression.new(:OR, @opts[clause], filter_expr(cond, &block)))
+      else
+        raise Error::NoExistingFilter, "No existing filter found."
+      end
+    end
+
+    # Returns a copy of the dataset with the order changed. If a nil is given
+    # the returned dataset has no order. This can accept multiple arguments
+    # of varying kinds, and even SQL functions.
+    #
+    #   ds.order(:name).sql #=> 'SELECT * FROM items ORDER BY name'
+    #   ds.order(:a, :b).sql #=> 'SELECT * FROM items ORDER BY a, b'
+    #   ds.order('a + b'.lit).sql #=> 'SELECT * FROM items ORDER BY a + b'
+    #   ds.order(:a + :b).sql #=> 'SELECT * FROM items ORDER BY (a + b)'
+    #   ds.order(:name.desc).sql #=> 'SELECT * FROM items ORDER BY name DESC'
+    #   ds.order(:name.asc).sql #=> 'SELECT * FROM items ORDER BY name ASC'
+    #   ds.order(:arr|1).sql #=> 'SELECT * FROM items ORDER BY arr[1]'
+    #   ds.order(nil).sql #=> 'SELECT * FROM items'
+    def order(*order)
+      clone(:order => (order.compact.empty?) ? nil : order)
+    end
+    alias_method :order_by, :order
+    
+    # Returns a copy of the dataset with the order columns added
+    # to the existing order.
+    def order_more(*order)
+      order(*((@opts[:order] || []) + order))
+    end
+    
+    # SQL fragment for the ordered expression, used in the ORDER BY
+    # clause.
+    def ordered_expression_sql(oe)
+      "#{literal(oe.expression)} #{oe.descending ? 'DESC' : 'ASC'}"
+    end
+
+    # SQL fragment for the qualifed identifier, specifying
+    # a table and a column (or schema and table).
+    def qualified_identifier_sql(qcr)
+      [qcr.table, qcr.column].map{|x| x.is_one_of?(SQL::QualifiedIdentifier, SQL::Identifier) ? literal(x) : quote_identifier(x)}.join('.')
+    end
+
+    # Adds quoting to identifiers (columns and tables). If identifiers are not
+    # being quoted, returns name as a string.  If identifiers are being quoted
+    # quote the name with quoted_identifier.
+    def quote_identifier(name)
+      quote_identifiers? ? quoted_identifier(name) : name.to_s
+    end
+    alias_method :quote_column_ref, :quote_identifier
+
+    # This method quotes the given name with the SQL standard double quote. 
+    # It uppercases the name given to conform with the SQL standard. This
+    # should be overridden by subclasses to provide quoting not matching the
+    # SQL standard, such as backtick (used by MySQL and SQLite), or where
+    # lowercase is the default for unquoted identifiers (PostgreSQL).
+    #
+    # If you are using a database such as Oracle that defaults to uppercase
+    # but you are using lower case identifiers, you should override this
+    # method to not upcase the name for those identifiers.
+    def quoted_identifier(name)
+      "\"#{name.to_s.upcase}\""
+    end
+
+    # Returns a copy of the dataset with the order reversed. If no order is
+    # given, the existing order is inverted.
+    def reverse_order(*order)
+      order(*invert_order(order.empty? ? @opts[:order] : order))
+    end
+    alias_method :reverse, :reverse_order
+
+    # Returns a copy of the dataset with the columns selected changed
+    # to the given columns.
+    def select(*columns)
+      clone(:select => columns)
+    end
+    
+    # Returns a copy of the dataset selecting the wildcard.
+    def select_all
+      clone(:select => nil)
+    end
+
+    # Returns a copy of the dataset with the given columns added
+    # to the existing selected columns.
+    def select_more(*columns)
+      select(*((@opts[:select] || []) + columns))
+    end
+    
+    # Formats a SELECT statement using the given options and the dataset
+    # options.
+    def select_sql(opts = nil)
+      opts = opts ? @opts.merge(opts) : @opts
+      
+      if sql = opts[:sql]
+        return sql
+      end
+
+      columns = opts[:select]
+      select_columns = columns ? column_list(columns) : WILDCARD
+
+      if distinct = opts[:distinct]
+        distinct_clause = distinct.empty? ? "DISTINCT" : "DISTINCT ON (#{expression_list(distinct)})"
+        sql = "SELECT #{distinct_clause} #{select_columns}"
+      else
+        sql = "SELECT #{select_columns}"
+      end
+      
+      if opts[:from]
+        sql << " FROM #{source_list(opts[:from])}"
+      end
+      
+      if join = opts[:join]
+        join.each{|j| sql << literal(j)}
+      end
+
+      if where = opts[:where]
+        sql << " WHERE #{literal(where)}"
+      end
+
+      if group = opts[:group]
+        sql << " GROUP BY #{expression_list(group)}"
+      end
+
+      if having = opts[:having]
+        sql << " HAVING #{literal(having)}"
+      end
+
+      if order = opts[:order]
+        sql << " ORDER BY #{expression_list(order)}"
+      end
+
+      if limit = opts[:limit]
+        sql << " LIMIT #{limit}"
+        if offset = opts[:offset]
+          sql << " OFFSET #{offset}"
+        end
+      end
+
+      if union = opts[:union]
+        sql << (opts[:union_all] ? \
+          " UNION ALL #{union.sql}" : " UNION #{union.sql}")
+      elsif intersect = opts[:intersect]
+        sql << (opts[:intersect_all] ? \
+          " INTERSECT ALL #{intersect.sql}" : " INTERSECT #{intersect.sql}")
+      elsif except = opts[:except]
+        sql << (opts[:except_all] ? \
+          " EXCEPT ALL #{except.sql}" : " EXCEPT #{except.sql}")
+      end
+
+      sql
+    end
+    alias_method :sql, :select_sql
+
+    # SQL fragment for specifying subscripts (SQL arrays)
+    def subscript_sql(s)
+      "#{s.f}[#{s.sub.join(COMMA_SEPARATOR)}]"
+    end
+
+    # Converts a symbol into a column name. This method supports underscore
+    # notation in order to express qualified (two underscores) and aliased
+    # (three underscores) columns:
+    #
+    #   ds = DB[:items]
+    #   :abc.to_column_ref(ds) #=> "abc"
+    #   :abc___a.to_column_ref(ds) #=> "abc AS a"
+    #   :items__abc.to_column_ref(ds) #=> "items.abc"
+    #   :items__abc___a.to_column_ref(ds) #=> "items.abc AS a"
+    #
+    def symbol_to_column_ref(sym)
+      c_table, column, c_alias = split_symbol(sym)
+      "#{"#{quote_identifier(c_table)}." if c_table}#{quote_identifier(column)}#{" AS #{quote_identifier(c_alias)}" if c_alias}"
+    end
+
+    # Returns a copy of the dataset with no filters (HAVING or WHERE clause) applied.
+    def unfiltered
+      clone(:where => nil, :having => nil)
+    end
+
+    # Adds a UNION clause using a second dataset object. If all is true the
+    # clause used is UNION ALL, which may return duplicate rows.
+    #
+    #   DB[:items].union(DB[:other_items]).sql
+    #   #=> "SELECT * FROM items UNION SELECT * FROM other_items"
+    def union(dataset, all = false)
+      clone(:union => dataset, :union_all => all)
+    end
+
+    # Returns a copy of the dataset with the distinct option.
+    def uniq(*args)
+      clone(:distinct => args)
+    end
+    alias_method :distinct, :uniq
+
+    # Returns a copy of the dataset with no order.
+    def unordered
+      order(nil)
+    end
+
+    # Formats an UPDATE statement using the given values.
+    #
+    #   dataset.update_sql(:price => 100, :category => 'software') #=>
+    #     "UPDATE items SET price = 100, category = 'software'"
+    #
+    # Accepts a block, but such usage is discouraged.
+    #
+    # Raises an error if the dataset is grouped or includes more
+    # than one table.
+    def update_sql(values = {}, opts = nil)
+      opts = opts ? @opts.merge(opts) : @opts
+
+      if opts[:group]
+        raise Error::InvalidOperation, "A grouped dataset cannot be updated"
+      elsif (opts[:from].size > 1) or opts[:join]
+        raise Error::InvalidOperation, "A joined dataset cannot be updated"
+      end
+      
+      sql = "UPDATE #{source_list(@opts[:from])} SET "
+      set = if values.is_a?(Hash)
+        # get values from hash
+        values = transform_save(values) if @transform
+        values.map do |k, v|
+          # convert string key into symbol
+          k = k.to_sym if String === k
+          "#{literal(k)} = #{literal(v)}"
+        end.join(COMMA_SEPARATOR)
+      else
+        # copy values verbatim
+        values
+      end
+      sql << set
+      if where = opts[:where]
+        sql << " WHERE #{literal(where)}"
+      end
+
+      sql
+    end
+
+    [:inner, :full_outer, :right_outer, :left_outer].each do |jtype|
+      class_eval("def #{jtype}_join(*args, &block); join_table(:#{jtype}, *args, &block) end")
+    end
+    alias_method :join, :inner_join
+
+    protected
+
+    # Returns a table reference for use in the FROM clause.  Returns an SQL subquery
+    # frgament with an optional table alias.
+    def to_table_reference(table_alias=nil)
+      "(#{sql})#{" #{quote_identifier(table_alias)}" if table_alias}"
+    end
+
+    private
+
+    # Converts an array of column names into a comma seperated string of 
+    # column names. If the array is empty, a wildcard (*) is returned.
+    def column_list(columns)
+      if columns.empty?
+        WILDCARD
+      else
+        m = columns.map do |i|
+          i.is_a?(Hash) ? i.map{|k, v| "#{literal(k)} AS #{quote_identifier(v)}"} : literal(i)
+        end
+        m.join(COMMA_SEPARATOR)
+      end
+    end
+    
+    # Converts an array of expressions into a comma separated string of
+    # expressions.
+    def expression_list(columns)
+      columns.map{|i| literal(i)}.join(COMMA_SEPARATOR)
+    end
+    
+    # SQL fragment based on the expr type.  See #filter.
+    def filter_expr(expr = nil, &block)
+      expr = nil if expr == []
+      if expr && block
+        return SQL::BooleanExpression.new(:AND, filter_expr(expr), filter_expr(block))
+      elsif block
+        expr = block
+      end
+      case expr
+      when Hash
+        SQL::BooleanExpression.from_value_pairs(expr)
+      when Array
+        if String === expr[0]
+          filter_expr(expr.shift.gsub(QUESTION_MARK){literal(expr.shift)}.lit)
+        else
+          SQL::BooleanExpression.from_value_pairs(expr)
+        end
+      when Proc
+        filter_expr(expr.call(SQL::VirtualRow.new))
+      when SQL::NumericExpression, SQL::StringExpression
+        raise(Error, "Invalid SQL Expression type: #{expr.inspect}") 
+      when Symbol, SQL::Expression
+        expr
+      when TrueClass, FalseClass
+        SQL::BooleanExpression.new(:NOOP, expr)
+      when String
+        "(#{expr})".lit
+      else
+        raise(Error, 'Invalid filter argument')
+      end
+    end
+
+    # SQL statement for formatting an insert statement with default values
+    def insert_default_values_sql
+      "INSERT INTO #{source_list(@opts[:from])} DEFAULT VALUES"
+    end
+
+    # Inverts the given order by breaking it into a list of column references
+    # and inverting them.
+    #
+    #   dataset.invert_order([:id.desc]]) #=> [:id]
+    #   dataset.invert_order(:category, :price.desc]) #=>
+    #     [:category.desc, :price]
+    def invert_order(order)
+      return nil unless order
+      new_order = []
+      order.map do |f|
+        case f
+        when SQL::OrderedExpression
+          SQL::OrderedExpression.new(f.expression, !f.descending)
+        else
+          SQL::OrderedExpression.new(f)
+        end
+      end
+    end
+    
+    # SQL fragment specifying a JOIN type, converts underscores to
+    # spaces and upcases.
+    def join_type_sql(join_type)
+      "#{join_type.to_s.gsub('_', ' ').upcase} JOIN"
+    end
+
+    # Returns a qualified column name (including a table name) if the column
+    # name isn't already qualified.
+    def qualified_column_name(column, table)
+      if Symbol === column 
+        c_table, column, c_alias = split_symbol(column)
+        schema, table, t_alias = split_symbol(table) if Symbol === table
+        c_table ||= t_alias || table
+        ::Sequel::SQL::QualifiedIdentifier.new(c_table, column)
+      else
+        column
+      end
+    end
+
+    # Converts an array of source names into into a comma separated list.
+    def source_list(source)
+      if source.nil? || source.empty?
+        raise Error, 'No source specified for query'
+      end
+      auto_alias_count = @opts[:num_dataset_sources] || 0
+      m = source.map do |s|
+        case s
+        when Dataset
+          auto_alias_count += 1
+          s.to_table_reference("t#{auto_alias_count}")
+        else
+          table_ref(s)
+        end
+      end
+      m.join(COMMA_SEPARATOR)
+    end
+    
+    # Splits the symbol into three parts.  Each part will
+    # either be a string or nil.
+    #
+    # For columns, these parts are the table, column, and alias.
+    # For tables, these parts are the schema, table, and alias.
+    def split_symbol(sym)
+      s = sym.to_s
+      if m = COLUMN_REF_RE1.match(s)
+        m[1..3]
+      elsif m = COLUMN_REF_RE2.match(s)
+        [nil, m[1], m[2]]
+      elsif m = COLUMN_REF_RE3.match(s)
+        [m[1], m[2], nil]
+      else
+        [nil, s, nil]
+      end
+    end
+
+    # SQL fragement specifying a table name.
+    def table_ref(t)
+      case t
+      when Dataset
+        t.to_table_reference
+      when Hash
+        t.map {|k, v| "#{table_ref(k)} #{table_ref(v)}"}.join(COMMA_SEPARATOR)
+      when Symbol
+        symbol_to_column_ref(t)
+      when String
+        quote_identifier(t)
+      else
+        literal(t)
+      end
+    end
+  end
+end