activerecord 5.0.7.2 → 6.1.1

data/lib/active_record/relation/batches.rb

@@ -1,8 +1,10 @@
+# frozen_string_literal: true
+
 require "active_record/relation/batches/batch_enumerator"
 
 module ActiveRecord
   module Batches
-    ORDER_OR_LIMIT_IGNORED_MESSAGE = "Scoped order and limit are ignored, it's forced to be batch order and batch size."
+    ORDER_IGNORE_MESSAGE = "Scoped order is ignored, it's forced to be batch order."
 
     # Looping through a collection of records from the database
     # (using the Scoping::Named::ClassMethods.all method, for example)
@@ -34,34 +36,44 @@ module ActiveRecord
     # * <tt>:start</tt> - Specifies the primary key value to start from, inclusive of the value.
     # * <tt>:finish</tt> - Specifies the primary key value to end at, inclusive of the value.
     # * <tt>:error_on_ignore</tt> - Overrides the application config to specify if an error should be raised when
-    #   the order and limit have to be ignored due to batching.
+    #   an order is present in the relation.
+    # * <tt>:order</tt> - Specifies the primary key order (can be :asc or :desc). Defaults to :asc.
+    #
+    # Limits are honored, and if present there is no requirement for the batch
+    # size: it can be less than, equal to, or greater than the limit.
     #
-    # This is especially useful if you want multiple workers dealing with
-    # the same processing queue. You can make worker 1 handle all the records
-    # between id 0 and 10,000 and worker 2 handle from 10,000 and beyond
-    # (by setting the +:start+ and +:finish+ option on each worker).
+    # The options +start+ and +finish+ are especially useful if you want
+    # multiple workers dealing with the same processing queue. You can make
+    # worker 1 handle all the records between id 1 and 9999 and worker 2
+    # handle from 10000 and beyond by setting the +:start+ and +:finish+
+    # option on each worker.
     #
-    #   # Let's process for a batch of 2000 records, skipping the first 2000 rows
-    #   Person.find_each(start: 2000, batch_size: 2000) do |person|
+    #   # In worker 1, let's process until 9999 records.
+    #   Person.find_each(finish: 9_999) do |person|
     #     person.party_all_night!
     #   end
     #
-    # NOTE: It's not possible to set the order. That is automatically set to
-    # ascending on the primary key ("id ASC") to make the batch ordering
-    # work. This also means that this method only works when the primary key is
+    #   # In worker 2, let's process from record 10_000 and onwards.
+    #   Person.find_each(start: 10_000) do |person|
+    #     person.party_all_night!
+    #   end
+    #
+    # NOTE: Order can be ascending (:asc) or descending (:desc). It is automatically set to
+    # ascending on the primary key ("id ASC").
+    # This also means that this method only works when the primary key is
     # orderable (e.g. an integer or string).
     #
-    # NOTE: You can't set the limit either, that's used to control
-    # the batch sizes.
-    def find_each(start: nil, finish: nil, batch_size: 1000, error_on_ignore: nil)
+    # NOTE: By its nature, batch processing is subject to race conditions if
+    # other processes are modifying the database.
+    def find_each(start: nil, finish: nil, batch_size: 1000, error_on_ignore: nil, order: :asc)
       if block_given?
-        find_in_batches(start: start, finish: finish, batch_size: batch_size, error_on_ignore: error_on_ignore) do |records|
+        find_in_batches(start: start, finish: finish, batch_size: batch_size, error_on_ignore: error_on_ignore, order: order) do |records|
           records.each { |record| yield record }
         end
       else
-        enum_for(:find_each, start: start, finish: finish, batch_size: batch_size, error_on_ignore: error_on_ignore) do
+        enum_for(:find_each, start: start, finish: finish, batch_size: batch_size, error_on_ignore: error_on_ignore, order: order) do
           relation = self
-          apply_limits(relation, start, finish).size
+          apply_limits(relation, start, finish, order).size
         end
       end
     end
@@ -89,35 +101,40 @@ module ActiveRecord
     # * <tt>:start</tt> - Specifies the primary key value to start from, inclusive of the value.
     # * <tt>:finish</tt> - Specifies the primary key value to end at, inclusive of the value.
     # * <tt>:error_on_ignore</tt> - Overrides the application config to specify if an error should be raised when
-    #   the order and limit have to be ignored due to batching.
+    #   an order is present in the relation.
+    # * <tt>:order</tt> - Specifies the primary key order (can be :asc or :desc). Defaults to :asc.
+    #
+    # Limits are honored, and if present there is no requirement for the batch
+    # size: it can be less than, equal to, or greater than the limit.
     #
-    # This is especially useful if you want multiple workers dealing with
-    # the same processing queue. You can make worker 1 handle all the records
-    # between id 0 and 10,000 and worker 2 handle from 10,000 and beyond
-    # (by setting the +:start+ and +:finish+ option on each worker).
+    # The options +start+ and +finish+ are especially useful if you want
+    # multiple workers dealing with the same processing queue. You can make
+    # worker 1 handle all the records between id 1 and 9999 and worker 2
+    # handle from 10000 and beyond by setting the +:start+ and +:finish+
+    # option on each worker.
     #
-    #   # Let's process the next 2000 records
-    #   Person.find_in_batches(start: 2000, batch_size: 2000) do |group|
+    #   # Let's process from record 10_000 on.
+    #   Person.find_in_batches(start: 10_000) do |group|
     #     group.each { |person| person.party_all_night! }
     #   end
     #
-    # NOTE: It's not possible to set the order. That is automatically set to
-    # ascending on the primary key ("id ASC") to make the batch ordering
-    # work. This also means that this method only works when the primary key is
+    # NOTE: Order can be ascending (:asc) or descending (:desc). It is automatically set to
+    # ascending on the primary key ("id ASC").
+    # This also means that this method only works when the primary key is
     # orderable (e.g. an integer or string).
     #
-    # NOTE: You can't set the limit either, that's used to control
-    # the batch sizes.
-    def find_in_batches(start: nil, finish: nil, batch_size: 1000, error_on_ignore: nil)
+    # NOTE: By its nature, batch processing is subject to race conditions if
+    # other processes are modifying the database.
+    def find_in_batches(start: nil, finish: nil, batch_size: 1000, error_on_ignore: nil, order: :asc)
       relation = self
       unless block_given?
-        return to_enum(:find_in_batches, start: start, finish: finish, batch_size: batch_size, error_on_ignore: error_on_ignore) do
-          total = apply_limits(relation, start, finish).size
+        return to_enum(:find_in_batches, start: start, finish: finish, batch_size: batch_size, error_on_ignore: error_on_ignore, order: order) do
+          total = apply_limits(relation, start, finish, order).size
           (total - 1).div(batch_size) + 1
         end
       end
 
-      in_batches(of: batch_size, start: start, finish: finish, load: true, error_on_ignore: error_on_ignore) do |batch|
+      in_batches(of: batch_size, start: start, finish: finish, load: true, error_on_ignore: error_on_ignore, order: order) do |batch|
         yield batch.to_a
       end
     end
@@ -149,17 +166,20 @@ module ActiveRecord
     # * <tt>:start</tt> - Specifies the primary key value to start from, inclusive of the value.
     # * <tt>:finish</tt> - Specifies the primary key value to end at, inclusive of the value.
     # * <tt>:error_on_ignore</tt> - Overrides the application config to specify if an error should be raised when
-    #   the order and limit have to be ignored due to batching.
+    #   an order is present in the relation.
+    # * <tt>:order</tt> - Specifies the primary key order (can be :asc or :desc). Defaults to :asc.
+    #
+    # Limits are honored, and if present there is no requirement for the batch
+    # size, it can be less than, equal, or greater than the limit.
     #
-    # This is especially useful if you want to work with the
-    # ActiveRecord::Relation object instead of the array of records, or if
-    # you want multiple workers dealing with the same processing queue. You can
-    # make worker 1 handle all the records between id 0 and 10,000 and worker 2
-    # handle from 10,000 and beyond (by setting the +:start+ and +:finish+
-    # option on each worker).
+    # The options +start+ and +finish+ are especially useful if you want
+    # multiple workers dealing with the same processing queue. You can make
+    # worker 1 handle all the records between id 1 and 9999 and worker 2
+    # handle from 10000 and beyond by setting the +:start+ and +:finish+
+    # option on each worker.
     #
-    #   # Let's process the next 2000 records
-    #   Person.in_batches(of: 2000, start: 2000).update_all(awesome: true)
+    #   # Let's process from record 10_000 on.
+    #   Person.in_batches(start: 10_000).update_all(awesome: true)
     #
     # An example of calling where query method on the relation:
     #
@@ -174,36 +194,47 @@ module ActiveRecord
     #
     #   Person.in_batches.each_record(&:party_all_night!)
     #
-    # NOTE: It's not possible to set the order. That is automatically set to
-    # ascending on the primary key ("id ASC") to make the batch ordering
-    # consistent. Therefore the primary key must be orderable, e.g. an integer
-    # or a string.
+    # NOTE: Order can be ascending (:asc) or descending (:desc). It is automatically set to
+    # ascending on the primary key ("id ASC").
+    # This also means that this method only works when the primary key is
+    # orderable (e.g. an integer or string).
     #
-    # NOTE: You can't set the limit either, that's used to control the batch
-    # sizes.
-    def in_batches(of: 1000, start: nil, finish: nil, load: false, error_on_ignore: nil)
+    # NOTE: By its nature, batch processing is subject to race conditions if
+    # other processes are modifying the database.
+    def in_batches(of: 1000, start: nil, finish: nil, load: false, error_on_ignore: nil, order: :asc)
       relation = self
       unless block_given?
         return BatchEnumerator.new(of: of, start: start, finish: finish, relation: self)
       end
 
-      if arel.orders.present? || arel.taken.present?
-        act_on_order_or_limit_ignored(error_on_ignore)
+      unless [:asc, :desc].include?(order)
+        raise ArgumentError, ":order must be :asc or :desc, got #{order.inspect}"
+      end
+
+      if arel.orders.present?
+        act_on_ignored_order(error_on_ignore)
       end
 
-      relation = relation.reorder(batch_order).limit(of)
-      relation = apply_limits(relation, start, finish)
+      batch_limit = of
+      if limit_value
+        remaining   = limit_value
+        batch_limit = remaining if remaining < batch_limit
+      end
+
+      relation = relation.reorder(batch_order(order)).limit(batch_limit)
+      relation = apply_limits(relation, start, finish, order)
+      relation.skip_query_cache! # Retaining the results in the query cache would undermine the point of batching
       batch_relation = relation
 
       loop do
         if load
           records = batch_relation.records
           ids = records.map(&:id)
-          yielded_relation = self.where(primary_key => ids)
+          yielded_relation = where(primary_key => ids)
           yielded_relation.load_records(records)
         else
           ids = batch_relation.pluck(primary_key)
-          yielded_relation = self.where(primary_key => ids)
+          yielded_relation = where(primary_key => ids)
         end
 
         break if ids.empty?
@@ -213,31 +244,53 @@ module ActiveRecord
 
         yield yielded_relation
 
-        break if ids.length < of
-        batch_relation = relation.where(arel_attribute(primary_key).gt(primary_key_offset))
+        break if ids.length < batch_limit
+
+        if limit_value
+          remaining -= ids.length
+
+          if remaining == 0
+            # Saves a useless iteration when the limit is a multiple of the
+            # batch size.
+            break
+          elsif remaining < batch_limit
+            relation = relation.limit(remaining)
+          end
+        end
+
+        batch_relation = relation.where(
+          predicate_builder[primary_key, primary_key_offset, order == :desc ? :lt : :gt]
+        )
       end
     end
 
     private
+      def apply_limits(relation, start, finish, order)
+        relation = apply_start_limit(relation, start, order) if start
+        relation = apply_finish_limit(relation, finish, order) if finish
+        relation
+      end
 
-    def apply_limits(relation, start, finish)
-      relation = relation.where(arel_attribute(primary_key).gteq(start)) if start
-      relation = relation.where(arel_attribute(primary_key).lteq(finish)) if finish
-      relation
-    end
+      def apply_start_limit(relation, start, order)
+        relation.where(predicate_builder[primary_key, start, order == :desc ? :lteq : :gteq])
+      end
 
-    def batch_order
-      "#{quoted_table_name}.#{quoted_primary_key} ASC"
-    end
+      def apply_finish_limit(relation, finish, order)
+        relation.where(predicate_builder[primary_key, finish, order == :desc ? :gteq : :lteq])
+      end
+
+      def batch_order(order)
+        table[primary_key].public_send(order)
+      end
 
-    def act_on_order_or_limit_ignored(error_on_ignore)
-      raise_error = (error_on_ignore.nil? ? self.klass.error_on_ignored_order_or_limit : error_on_ignore)
+      def act_on_ignored_order(error_on_ignore)
+        raise_error = (error_on_ignore.nil? ? klass.error_on_ignored_order : error_on_ignore)
 
-      if raise_error
-        raise ArgumentError.new(ORDER_OR_LIMIT_IGNORED_MESSAGE)
-      elsif logger
-        logger.warn(ORDER_OR_LIMIT_IGNORED_MESSAGE)
+        if raise_error
+          raise ArgumentError.new(ORDER_IGNORE_MESSAGE)
+        elsif logger
+          logger.warn(ORDER_IGNORE_MESSAGE)
+        end
       end
-    end
   end
 end

data/lib/active_record/relation/batches/batch_enumerator.rb

@@ -1,3 +1,5 @@
+# frozen_string_literal: true
+
 module ActiveRecord
   module Batches
     class BatchEnumerator
@@ -7,7 +9,7 @@ module ActiveRecord
         @of       = of
         @relation = relation
         @start = start
-        @finish   = finish
+        @finish = finish
       end
 
       # Looping through a collection of records from the database (using the
@@ -39,19 +41,35 @@ module ActiveRecord
         end
       end
 
-      # Delegates #delete_all, #update_all, #destroy_all methods to each batch.
+      # Deletes records in batches. Returns the total number of rows affected.
+      #
+      #   Person.in_batches.delete_all
+      #
+      # See Relation#delete_all for details of how each batch is deleted.
+      def delete_all
+        sum(&:delete_all)
+      end
+
+      # Updates records in batches. Returns the total number of rows affected.
       #
-      #   People.in_batches.delete_all
-      #   People.where('age < 10').in_batches.destroy_all
-      #   People.in_batches.update_all('age = age + 1')
-      [:delete_all, :update_all, :destroy_all].each do |method|
-        define_method(method) do |*args, &block|
-          @relation.to_enum(:in_batches, of: @of, start: @start, finish: @finish, load: false).each do |relation|
-            relation.send(method, *args, &block)
-          end
+      #   Person.in_batches.update_all("age = age + 1")
+      #
+      # See Relation#update_all for details of how each batch is updated.
+      def update_all(updates)
+        sum do |relation|
+          relation.update_all(updates)
         end
       end
 
+      # Destroys records in batches.
+      #
+      #   Person.where("age < 10").in_batches.destroy_all
+      #
+      # See Relation#destroy_all for details of how each batch is destroyed.
+      def destroy_all
+        each(&:destroy_all)
+      end
+
       # Yields an ActiveRecord::Relation object for each batch of records.
       #
       #   Person.in_batches.each do |relation|

data/lib/active_record/relation/calculations.rb

@@ -1,3 +1,7 @@
+# frozen_string_literal: true
+
+require "active_support/core_ext/enumerable"
+
 module ActiveRecord
   module Calculations
     # Count the records.
@@ -37,7 +41,15 @@ module ActiveRecord
     # Note: not all valid {Relation#select}[rdoc-ref:QueryMethods#select] expressions are valid #count expressions. The specifics differ
     # between databases. In invalid cases, an error from the database is thrown.
     def count(column_name = nil)
-      calculate(:count, column_name)
+      if block_given?
+        unless column_name.nil?
+          raise ArgumentError, "Column name argument is not supported when a block is passed."
+        end
+
+        super()
+      else
+        calculate(:count, column_name)
+      end
     end
 
     # Calculates the average value on a given column. Returns +nil+ if there's
@@ -71,9 +83,16 @@ module ActiveRecord
     # #calculate for examples with options.
     #
     #   Person.sum(:age) # => 4562
-    def sum(column_name = nil, &block)
-      return super(&block) if block_given?
-      calculate(:sum, column_name)
+    def sum(column_name = nil)
+      if block_given?
+        unless column_name.nil?
+          raise ArgumentError, "Column name argument is not supported when a block is passed."
+        end
+
+        super()
+      else
+        calculate(:sum, column_name)
+      end
     end
 
     # This calculates aggregate values in the given column. Methods for #count, #sum, #average,
@@ -108,13 +127,17 @@ module ActiveRecord
     #        ...
     #      end
     def calculate(operation, column_name)
-      if column_name.is_a?(Symbol) && attribute_alias?(column_name)
-        column_name = attribute_alias(column_name)
-      end
-
       if has_include?(column_name)
-        relation = construct_relation_for_association_calculations
-        relation = relation.distinct if operation.to_s.downcase == "count"
+        relation = apply_join_dependency
+
+        if operation.to_s.downcase == "count"
+          unless distinct_value || distinct_select?(column_name || select_for_count)
+            relation.distinct!
+            relation.select_values = [ klass.primary_key || table[Arel.star] ]
+          end
+          # PostgreSQL: ORDER BY expressions must appear in SELECT list when using DISTINCT
+          relation.order_values = [] if group_values.empty?
+        end
 
         relation.calculate(operation, column_name)
       else
@@ -151,29 +174,58 @@ module ActiveRecord
     #   # SELECT people.id FROM people WHERE people.age = 21 LIMIT 5
     #   # => [2, 3]
     #
-    #   Person.pluck('DATEDIFF(updated_at, created_at)')
+    #   Person.pluck(Arel.sql('DATEDIFF(updated_at, created_at)'))
     #   # SELECT DATEDIFF(updated_at, created_at) FROM people
     #   # => ['0', '27761', '173']
     #
     # See also #ids.
     #
     def pluck(*column_names)
-      if loaded? && (column_names.map(&:to_s) - @klass.attribute_names - @klass.attribute_aliases.keys).empty?
+      if loaded? && all_attributes?(column_names)
         return records.pluck(*column_names)
       end
 
       if has_include?(column_names.first)
-        construct_relation_for_association_calculations.pluck(*column_names)
+        relation = apply_join_dependency
+        relation.pluck(*column_names)
       else
+        klass.disallow_raw_sql!(column_names)
+        columns = arel_columns(column_names)
         relation = spawn
-        relation.select_values = column_names.map { |cn|
-          @klass.has_attribute?(cn) || @klass.attribute_alias?(cn) ? arel_attribute(cn) : cn
-        }
-        result = klass.connection.select_all(relation.arel, nil, bound_attributes)
-        result.cast_values(klass.attribute_types)
+        relation.select_values = columns
+        result = skip_query_cache_if_necessary do
+          if where_clause.contradiction?
+            ActiveRecord::Result.new([], [])
+          else
+            klass.connection.select_all(relation.arel, nil)
+          end
+        end
+        type_cast_pluck_values(result, columns)
       end
     end
 
+    # Pick the value(s) from the named column(s) in the current relation.
+    # This is short-hand for <tt>relation.limit(1).pluck(*column_names).first</tt>, and is primarily useful
+    # when you have a relation that's already narrowed down to a single row.
+    #
+    # Just like #pluck, #pick will only load the actual value, not the entire record object, so it's also
+    # more efficient. The value is, again like with pluck, typecast by the column type.
+    #
+    #   Person.where(id: 1).pick(:name)
+    #   # SELECT people.name FROM people WHERE id = 1 LIMIT 1
+    #   # => 'David'
+    #
+    #   Person.where(id: 1).pick(:name, :email_address)
+    #   # SELECT people.name, people.email_address FROM people WHERE id = 1 LIMIT 1
+    #   # => [ 'David', 'david@loudthinking.com' ]
+    def pick(*column_names)
+      if loaded? && all_attributes?(column_names)
+        return records.pick(*column_names)
+      end
+
+      limit(1).pluck(*column_names).first
+    end
+
     # Pluck all the ID's for the relation using the table's primary key
     #
     #   Person.ids # SELECT people.id FROM people
@@ -183,203 +235,247 @@ module ActiveRecord
     end
 
     private
+      def all_attributes?(column_names)
+        (column_names.map(&:to_s) - @klass.attribute_names - @klass.attribute_aliases.keys).empty?
+      end
 
-    def has_include?(column_name)
-      eager_loading? || (includes_values.present? && column_name && column_name != :all)
-    end
+      def has_include?(column_name)
+        eager_loading? || (includes_values.present? && column_name && column_name != :all)
+      end
+
+      def perform_calculation(operation, column_name)
+        operation = operation.to_s.downcase
 
-    def perform_calculation(operation, column_name)
-      operation = operation.to_s.downcase
+        # If #count is used with #distinct (i.e. `relation.distinct.count`) it is
+        # considered distinct.
+        distinct = distinct_value
 
-      # If #count is used with #distinct (i.e. `relation.distinct.count`) it is
-      # considered distinct.
-      distinct = self.distinct_value
+        if operation == "count"
+          column_name ||= select_for_count
+          if column_name == :all
+            if !distinct
+              distinct = distinct_select?(select_for_count) if group_values.empty?
+            elsif group_values.any? || select_values.empty? && order_values.empty?
+              column_name = primary_key
+            end
+          elsif distinct_select?(column_name)
+            distinct = nil
+          end
+        end
 
-      if operation == "count"
-        column_name ||= select_for_count
-        column_name = primary_key if column_name == :all && distinct
-        distinct = nil if column_name =~ /\s*DISTINCT[\s(]+/i
+        if group_values.any?
+          execute_grouped_calculation(operation, column_name, distinct)
+        else
+          execute_simple_calculation(operation, column_name, distinct)
+        end
       end
 
-      if group_values.any?
-        execute_grouped_calculation(operation, column_name, distinct)
-      else
-        execute_simple_calculation(operation, column_name, distinct)
+      def distinct_select?(column_name)
+        column_name.is_a?(::String) && /\bDISTINCT[\s(]/i.match?(column_name)
       end
-    end
 
-    def aggregate_column(column_name)
-      return column_name if Arel::Expressions === column_name
+      def aggregate_column(column_name)
+        return column_name if Arel::Expressions === column_name
 
-      if @klass.column_names.include?(column_name.to_s)
-        Arel::Attribute.new(@klass.unscoped.table, column_name)
-      else
-        Arel.sql(column_name == :all ? "*" : column_name.to_s)
+        arel_column(column_name.to_s) do |name|
+          Arel.sql(column_name == :all ? "*" : name)
+        end
       end
-    end
 
-    def operation_over_aggregate_column(column, operation, distinct)
-      operation == 'count' ? column.count(distinct) : column.send(operation)
-    end
-
-    def execute_simple_calculation(operation, column_name, distinct) #:nodoc:
-      # PostgreSQL doesn't like ORDER BY when there are no GROUP BY
-      relation = unscope(:order)
+      def operation_over_aggregate_column(column, operation, distinct)
+        operation == "count" ? column.count(distinct) : column.public_send(operation)
+      end
 
-      column_alias = column_name
+      def execute_simple_calculation(operation, column_name, distinct) #:nodoc:
+        if operation == "count" && (column_name == :all && distinct || has_limit_or_offset?)
+          # Shortcut when limit is zero.
+          return 0 if limit_value == 0
 
-      if operation == "count" && (relation.limit_value || relation.offset_value)
-        # Shortcut when limit is zero.
-        return 0 if relation.limit_value == 0
+          query_builder = build_count_subquery(spawn, column_name, distinct)
+        else
+          # PostgreSQL doesn't like ORDER BY when there are no GROUP BY
+          relation = unscope(:order).distinct!(false)
 
-        query_builder = build_count_subquery(relation, column_name, distinct)
-      else
-        column = aggregate_column(column_name)
+          column = aggregate_column(column_name)
+          select_value = operation_over_aggregate_column(column, operation, distinct)
+          select_value.distinct = true if operation == "sum" && distinct
 
-        select_value = operation_over_aggregate_column(column, operation, distinct)
+          relation.select_values = [select_value]
 
-        if operation == "sum" && distinct
-          select_value.distinct = true
+          query_builder = relation.arel
         end
 
-        column_alias = select_value.alias
-        column_alias ||= @klass.connection.column_name_for_operation(operation, select_value)
-        relation.select_values = [select_value]
+        result = skip_query_cache_if_necessary { @klass.connection.select_all(query_builder) }
 
-        query_builder = relation.arel
+        type_cast_calculated_value(result.cast_values.first, operation) do |value|
+          type = column.try(:type_caster) ||
+            lookup_cast_type_from_join_dependencies(column_name.to_s) || Type.default_value
+          type.deserialize(value)
+        end
       end
 
-      result = @klass.connection.select_all(query_builder, nil, bound_attributes)
-      row    = result.first
-      value  = row && row.values.first
-      column = result.column_types.fetch(column_alias) do
-        type_for(column_name)
-      end
+      def execute_grouped_calculation(operation, column_name, distinct) #:nodoc:
+        group_fields = group_values
+        group_fields = group_fields.uniq if group_fields.size > 1
+
+        unless group_fields == group_values
+          ActiveSupport::Deprecation.warn(<<-MSG.squish)
+            `#{operation}` with group by duplicated fields does no longer affect to result in Rails 6.2.
+            To migrate to Rails 6.2's behavior, use `uniq!(:group)` to deduplicate group fields
+            (`#{klass.name&.tableize || klass.table_name}.uniq!(:group).#{operation}(#{column_name.inspect})`).
+          MSG
+          group_fields = group_values
+        end
 
-      type_cast_calculated_value(value, column, operation)
-    end
+        if group_fields.size == 1 && group_fields.first.respond_to?(:to_sym)
+          association  = klass._reflect_on_association(group_fields.first)
+          associated   = association && association.belongs_to? # only count belongs_to associations
+          group_fields = Array(association.foreign_key) if associated
+        end
+        group_fields = arel_columns(group_fields)
 
-    def execute_grouped_calculation(operation, column_name, distinct) #:nodoc:
-      group_attrs = group_values
+        group_aliases = group_fields.map { |field|
+          field = connection.visitor.compile(field) if Arel.arel_node?(field)
+          column_alias_for(field.to_s.downcase)
+        }
+        group_columns = group_aliases.zip(group_fields)
 
-      if group_attrs.first.respond_to?(:to_sym)
-        association  = @klass._reflect_on_association(group_attrs.first)
-        associated   = group_attrs.size == 1 && association && association.belongs_to? # only count belongs_to associations
-        group_fields = Array(associated ? association.foreign_key : group_attrs)
-      else
-        group_fields = group_attrs
-      end
-      group_fields = arel_columns(group_fields)
+        column = aggregate_column(column_name)
+        column_alias = column_alias_for("#{operation} #{column_name.to_s.downcase}")
+        select_value = operation_over_aggregate_column(column, operation, distinct)
+        select_value.as(column_alias)
 
-      group_aliases = group_fields.map { |field| column_alias_for(field) }
-      group_columns = group_aliases.zip(group_fields)
+        select_values = [select_value]
+        select_values += self.select_values unless having_clause.empty?
 
-      if operation == 'count' && column_name == :all
-        aggregate_alias = 'count_all'
-      else
-        aggregate_alias = column_alias_for([operation, column_name].join(' '))
-      end
+        select_values.concat group_columns.map { |aliaz, field|
+          if field.respond_to?(:as)
+            field.as(aliaz)
+          else
+            "#{field} AS #{aliaz}"
+          end
+        }
 
-      select_values = [
-        operation_over_aggregate_column(
-          aggregate_column(column_name),
-          operation,
-          distinct).as(aggregate_alias)
-      ]
-      select_values += self.select_values unless having_clause.empty?
-
-      select_values.concat group_columns.map { |aliaz, field|
-        if field.respond_to?(:as)
-          field.as(aliaz)
-        else
-          "#{field} AS #{aliaz}"
-        end
-      }
+        relation = except(:group).distinct!(false)
+        relation.group_values  = group_fields
+        relation.select_values = select_values
 
-      relation = except(:group)
-      relation.group_values  = group_fields
-      relation.select_values = select_values
+        calculated_data = skip_query_cache_if_necessary { @klass.connection.select_all(relation.arel, nil) }
 
-      calculated_data = @klass.connection.select_all(relation, nil, relation.bound_attributes)
+        if association
+          key_ids     = calculated_data.collect { |row| row[group_aliases.first] }
+          key_records = association.klass.base_class.where(association.klass.base_class.primary_key => key_ids)
+          key_records = key_records.index_by(&:id)
+        end
 
-      if association
-        key_ids     = calculated_data.collect { |row| row[group_aliases.first] }
-        key_records = association.klass.base_class.where(association.klass.base_class.primary_key => key_ids)
-        key_records = Hash[key_records.map { |r| [r.id, r] }]
-      end
+        key_types = group_columns.each_with_object({}) do |(aliaz, col_name), types|
+          types[aliaz] = type_for(col_name) do
+            calculated_data.column_types.fetch(aliaz, Type.default_value)
+          end
+        end
 
-      Hash[calculated_data.map do |row|
-        key = group_columns.map { |aliaz, col_name|
-          column = type_for(col_name) do
-            calculated_data.column_types.fetch(aliaz) do
-              Type::Value.new
-            end
+        hash_rows = calculated_data.cast_values(key_types).map! do |row|
+          calculated_data.columns.each_with_object({}).with_index do |(col_name, hash), i|
+            hash[col_name] = row[i]
           end
-          type_cast_calculated_value(row[aliaz], column)
-        }
-        key = key.first if key.size == 1
-        key = key_records[key] if associated
+        end
 
-        column_type = calculated_data.column_types.fetch(aggregate_alias) { type_for(column_name) }
-        [key, type_cast_calculated_value(row[aggregate_alias], column_type, operation)]
-      end]
-    end
+        type = nil
+        hash_rows.each_with_object({}) do |row, result|
+          key = group_aliases.map { |aliaz| row[aliaz] }
+          key = key.first if key.size == 1
+          key = key_records[key] if associated
 
-    # Converts the given keys to the value that the database adapter returns as
-    # a usable column name:
-    #
-    #   column_alias_for("users.id")                 # => "users_id"
-    #   column_alias_for("sum(id)")                  # => "sum_id"
-    #   column_alias_for("count(distinct users.id)") # => "count_distinct_users_id"
-    #   column_alias_for("count(*)")                 # => "count_all"
-    def column_alias_for(keys)
-      if keys.respond_to? :name
-        keys = "#{keys.relation.name}.#{keys.name}"
+          result[key] = type_cast_calculated_value(row[column_alias], operation) do |value|
+            type ||= column.try(:type_caster) ||
+              lookup_cast_type_from_join_dependencies(column_name.to_s) || Type.default_value
+            type.deserialize(value)
+          end
+        end
       end
 
-      table_name = keys.to_s.downcase
-      table_name.gsub!(/\*/, 'all')
-      table_name.gsub!(/\W+/, ' ')
-      table_name.strip!
-      table_name.gsub!(/ +/, '_')
+      # Converts the given field to the value that the database adapter returns as
+      # a usable column name:
+      #
+      #   column_alias_for("users.id")                 # => "users_id"
+      #   column_alias_for("sum(id)")                  # => "sum_id"
+      #   column_alias_for("count(distinct users.id)") # => "count_distinct_users_id"
+      #   column_alias_for("count(*)")                 # => "count_all"
+      def column_alias_for(field)
+        column_alias = +field
+        column_alias.gsub!(/\*/, "all")
+        column_alias.gsub!(/\W+/, " ")
+        column_alias.strip!
+        column_alias.gsub!(/ +/, "_")
+
+        connection.table_alias_for(column_alias)
+      end
 
-      @klass.connection.table_alias_for(table_name)
-    end
+      def type_for(field, &block)
+        field_name = field.respond_to?(:name) ? field.name.to_s : field.to_s.split(".").last
+        @klass.type_for_attribute(field_name, &block)
+      end
 
-    def type_for(field, &block)
-      field_name = field.respond_to?(:name) ? field.name.to_s : field.to_s.split('.').last
-      @klass.type_for_attribute(field_name, &block)
-    end
+      def lookup_cast_type_from_join_dependencies(name, join_dependencies = build_join_dependencies)
+        each_join_dependencies(join_dependencies) do |join|
+          type = join.base_klass.attribute_types.fetch(name, nil)
+          return type if type
+        end
+        nil
+      end
 
-    def type_cast_calculated_value(value, type, operation = nil)
-      case operation
-        when 'count'   then value.to_i
-        when 'sum'     then type.deserialize(value || 0)
-        when 'average' then value.respond_to?(:to_d) ? value.to_d : value
-        else type.deserialize(value)
+      def type_cast_pluck_values(result, columns)
+        cast_types = if result.columns.size != columns.size
+          klass.attribute_types
+        else
+          join_dependencies = nil
+          columns.map.with_index do |column, i|
+            column.try(:type_caster) ||
+              klass.attribute_types.fetch(name = result.columns[i]) do
+                join_dependencies ||= build_join_dependencies
+                lookup_cast_type_from_join_dependencies(name, join_dependencies) ||
+                  result.column_types[name] || Type.default_value
+              end
+          end
+        end
+        result.cast_values(cast_types)
       end
-    end
 
-    def select_for_count
-      if select_values.present?
-        return select_values.first if select_values.one?
-        select_values.join(", ")
-      else
-        :all
+      def type_cast_calculated_value(value, operation)
+        case operation
+        when "count"
+          value.to_i
+        when "sum"
+          yield value || 0
+        when "average"
+          value&.respond_to?(:to_d) ? value.to_d : value
+        else # "minimum", "maximum"
+          yield value
+        end
       end
-    end
 
-    def build_count_subquery(relation, column_name, distinct)
-      column_alias = Arel.sql('count_column')
-      subquery_alias = Arel.sql('subquery_for_count')
+      def select_for_count
+        if select_values.present?
+          return select_values.first if select_values.one?
+          select_values.join(", ")
+        else
+          :all
+        end
+      end
 
-      aliased_column = aggregate_column(column_name == :all ? 1 : column_name).as(column_alias)
-      relation.select_values = [aliased_column]
-      subquery = relation.arel.as(subquery_alias)
+      def build_count_subquery(relation, column_name, distinct)
+        if column_name == :all
+          column_alias = Arel.star
+          relation.select_values = [ Arel.sql(FinderMethods::ONE_AS_ONE) ] unless distinct
+        else
+          column_alias = Arel.sql("count_column")
+          relation.select_values = [ aggregate_column(column_name).as(column_alias) ]
+        end
 
-      sm = Arel::SelectManager.new relation.engine
-      select_value = operation_over_aggregate_column(column_alias, 'count', distinct)
-      sm.project(select_value).from(subquery)
-    end
+        subquery_alias = Arel.sql("subquery_for_count")
+        select_value = operation_over_aggregate_column(column_alias, "count", false)
+
+        relation.build_subquery(subquery_alias, select_value)
+      end
   end
 end