postgresql_adapter.rb

# frozen_string_literal: true

gem "pg", "~> 1.1"
require "pg"

require "active_support/core_ext/object/try"
require "active_record/connection_adapters/abstract_adapter"
require "active_record/connection_adapters/statement_pool"
require "active_record/connection_adapters/postgresql/column"
require "active_record/connection_adapters/postgresql/database_statements"
require "active_record/connection_adapters/postgresql/explain_pretty_printer"
require "active_record/connection_adapters/postgresql/oid"
require "active_record/connection_adapters/postgresql/quoting"
require "active_record/connection_adapters/postgresql/referential_integrity"
require "active_record/connection_adapters/postgresql/schema_creation"
require "active_record/connection_adapters/postgresql/schema_definitions"
require "active_record/connection_adapters/postgresql/schema_dumper"
require "active_record/connection_adapters/postgresql/schema_statements"
require "active_record/connection_adapters/postgresql/type_metadata"
require "active_record/connection_adapters/postgresql/utils"

module ActiveRecord
  module ConnectionHandling # :nodoc:
    # Establishes a connection to the database that's used by all Active Record objects
    def postgresql_connection(config)
      conn_params = config.symbolize_keys.compact

      # Map ActiveRecords param names to PGs.
      conn_params[:user] = conn_params.delete(:username) if conn_params[:username]
      conn_params[:dbname] = conn_params.delete(:database) if conn_params[:database]

      # Forward only valid config params to PG::Connection.connect.
      valid_conn_param_keys = PG::Connection.conndefaults_hash.keys + [:requiressl]
      conn_params.slice!(*valid_conn_param_keys)

      ConnectionAdapters::PostgreSQLAdapter.new(
        ConnectionAdapters::PostgreSQLAdapter.new_client(conn_params),
        logger,
        conn_params,
        config,
      )
    end
  end

  module ConnectionAdapters
    # The PostgreSQL adapter works with the native C (https://github.com/ged/ruby-pg) driver.
    #
    # Options:
    #
    # * <tt>:host</tt> - Defaults to a Unix-domain socket in /tmp. On machines without Unix-domain sockets,
    #   the default is to connect to localhost.
    # * <tt>:port</tt> - Defaults to 5432.
    # * <tt>:username</tt> - Defaults to be the same as the operating system name of the user running the application.
    # * <tt>:password</tt> - Password to be used if the server demands password authentication.
    # * <tt>:database</tt> - Defaults to be the same as the username.
    # * <tt>:schema_search_path</tt> - An optional schema search path for the connection given
    #   as a string of comma-separated schema names. This is backward-compatible with the <tt>:schema_order</tt> option.
    # * <tt>:encoding</tt> - An optional client encoding that is used in a <tt>SET client_encoding TO
    #   <encoding></tt> call on the connection.
    # * <tt>:min_messages</tt> - An optional client min messages that is used in a
    #   <tt>SET client_min_messages TO <min_messages></tt> call on the connection.
    # * <tt>:variables</tt> - An optional hash of additional parameters that
    #   will be used in <tt>SET SESSION key = val</tt> calls on the connection.
    # * <tt>:insert_returning</tt> - An optional boolean to control the use of <tt>RETURNING</tt> for <tt>INSERT</tt> statements
    #   defaults to true.
    #
    # Any further options are used as connection parameters to libpq. See
    # https://www.postgresql.org/docs/current/static/libpq-connect.html for the
    # list of parameters.
    #
    # In addition, default connection parameters of libpq can be set per environment variables.
    # See https://www.postgresql.org/docs/current/static/libpq-envars.html .
    class PostgreSQLAdapter < AbstractAdapter
      ADAPTER_NAME = "PostgreSQL"

      class << self
        def new_client(conn_params)
          PG.connect(**conn_params)
        rescue ::PG::Error => error
          if conn_params && conn_params[:dbname] && error.message.include?(conn_params[:dbname])
            raise ActiveRecord::NoDatabaseError.db_error(conn_params[:dbname])
          elsif conn_params && conn_params[:user] && error.message.include?(conn_params[:user])
            raise ActiveRecord::DatabaseConnectionError.username_error(conn_params[:user])
          elsif conn_params && conn_params[:hostname] && error.message.include?(conn_params[:hostname])
            raise ActiveRecord::DatabaseConnectionError.hostname_error(conn_params[:hostname])
          else
            raise ActiveRecord::ConnectionNotEstablished, error.message
          end
        end
      end

      ##
      # :singleton-method:
      # PostgreSQL allows the creation of "unlogged" tables, which do not record
      # data in the PostgreSQL Write-Ahead Log. This can make the tables faster,
      # but significantly increases the risk of data loss if the database
      # crashes. As a result, this should not be used in production
      # environments. If you would like all created tables to be unlogged in
      # the test environment you can add the following line to your test.rb
      # file:
      #
      #   ActiveRecord::ConnectionAdapters::PostgreSQLAdapter.create_unlogged_tables = true
      class_attribute :create_unlogged_tables, default: false

      ##
      # :singleton-method:
      # PostgreSQL supports multiple types for DateTimes. By default, if you use +datetime+
      # in migrations, Rails will translate this to a PostgreSQL "timestamp without time zone".
      # Change this in an initializer to use another NATIVE_DATABASE_TYPES. For example, to
      # store DateTimes as "timestamp with time zone":
      #
      #   ActiveRecord::ConnectionAdapters::PostgreSQLAdapter.datetime_type = :timestamptz
      #
      # Or if you are adding a custom type:
      #
      #   ActiveRecord::ConnectionAdapters::PostgreSQLAdapter::NATIVE_DATABASE_TYPES[:my_custom_type] = { name: "my_custom_type_name" }
      #   ActiveRecord::ConnectionAdapters::PostgreSQLAdapter.datetime_type = :my_custom_type
      #
      # If you're using +:ruby+ as your +config.active_record.schema_format+ and you change this
      # setting, you should immediately run <tt>bin/rails db:migrate</tt> to update the types in your schema.rb.
      class_attribute :datetime_type, default: :timestamp

      NATIVE_DATABASE_TYPES = {
        primary_key: "bigserial primary key",
        string:      { name: "character varying" },
        text:        { name: "text" },
        integer:     { name: "integer", limit: 4 },
        bigint:      { name: "bigint" },
        float:       { name: "float" },
        decimal:     { name: "decimal" },
        datetime:    {}, # set dynamically based on datetime_type
        timestamp:   { name: "timestamp" },
        timestamptz: { name: "timestamptz" },
        time:        { name: "time" },
        date:        { name: "date" },
        daterange:   { name: "daterange" },
        numrange:    { name: "numrange" },
        tsrange:     { name: "tsrange" },
        tstzrange:   { name: "tstzrange" },
        int4range:   { name: "int4range" },
        int8range:   { name: "int8range" },
        binary:      { name: "bytea" },
        boolean:     { name: "boolean" },
        xml:         { name: "xml" },
        tsvector:    { name: "tsvector" },
        hstore:      { name: "hstore" },
        inet:        { name: "inet" },
        cidr:        { name: "cidr" },
        macaddr:     { name: "macaddr" },
        uuid:        { name: "uuid" },
        json:        { name: "json" },
        jsonb:       { name: "jsonb" },
        ltree:       { name: "ltree" },
        citext:      { name: "citext" },
        point:       { name: "point" },
        line:        { name: "line" },
        lseg:        { name: "lseg" },
        box:         { name: "box" },
        path:        { name: "path" },
        polygon:     { name: "polygon" },
        circle:      { name: "circle" },
        bit:         { name: "bit" },
        bit_varying: { name: "bit varying" },
        money:       { name: "money" },
        interval:    { name: "interval" },
        oid:         { name: "oid" },
        enum:        {} # special type https://www.postgresql.org/docs/current/datatype-enum.html
      }

      OID = PostgreSQL::OID # :nodoc:

      include PostgreSQL::Quoting
      include PostgreSQL::ReferentialIntegrity
      include PostgreSQL::SchemaStatements
      include PostgreSQL::DatabaseStatements

      def supports_bulk_alter?
        true
      end

      def supports_index_sort_order?
        true
      end

      def supports_partitioned_indexes?
        database_version >= 110_000 # >= 11.0
      end

      def supports_partial_index?
        true
      end

      def supports_expression_index?
        true
      end

      def supports_transaction_isolation?
        true
      end

      def supports_foreign_keys?
        true
      end

      def supports_check_constraints?
        true
      end

      def supports_validate_constraints?
        true
      end

      def supports_deferrable_constraints?
        true
      end

      def supports_views?
        true
      end

      def supports_datetime_with_precision?
        true
      end

      def supports_json?
        true
      end

      def supports_comments?
        true
      end

      def supports_savepoints?
        true
      end

      def supports_insert_returning?
        true
      end

      def supports_insert_on_conflict?
        database_version >= 90500 # >= 9.5
      end
      alias supports_insert_on_duplicate_skip? supports_insert_on_conflict?
      alias supports_insert_on_duplicate_update? supports_insert_on_conflict?
      alias supports_insert_conflict_target? supports_insert_on_conflict?

      def supports_virtual_columns?
        database_version >= 120_000 # >= 12.0
      end

      def index_algorithms
        { concurrently: "CONCURRENTLY" }
      end

      class StatementPool < ConnectionAdapters::StatementPool # :nodoc:
        def initialize(connection, max)
          super(max)
          @connection = connection
          @counter = 0
        end

        def next_key
          "a#{@counter += 1}"
        end

        private
          def dealloc(key)
            @connection.query "DEALLOCATE #{key}" if connection_active?
          rescue PG::Error
          end

          def connection_active?
            @connection.status == PG::CONNECTION_OK
          rescue PG::Error
            false
          end
      end

      # Initializes and connects a PostgreSQL adapter.
      def initialize(connection, logger, connection_parameters, config)
        super(connection, logger, config)

        @connection_parameters = connection_parameters || {}

        # @local_tz is initialized as nil to avoid warnings when connect tries to use it
        @local_tz = nil
        @max_identifier_length = nil

        configure_connection
        add_pg_encoders
        add_pg_decoders

        @type_map = Type::HashLookupTypeMap.new
        initialize_type_map
        @local_tz = execute("SHOW TIME ZONE", "SCHEMA").first["TimeZone"]
        @use_insert_returning = @config.key?(:insert_returning) ? self.class.type_cast_config_to_boolean(@config[:insert_returning]) : true
      end

      def self.database_exists?(config)
        !!ActiveRecord::Base.postgresql_connection(config)
      rescue ActiveRecord::NoDatabaseError
        false
      end

      # Is this connection alive and ready for queries?
      def active?
        @lock.synchronize do
          @connection.query ";"
        end
        true
      rescue PG::Error
        false
      end

      def reload_type_map # :nodoc:
        type_map.clear
        initialize_type_map
      end

      # Close then reopen the connection.
      def reconnect!
        @lock.synchronize do
          super
          @connection.reset
          configure_connection
          reload_type_map
        rescue PG::ConnectionBad
          connect
        end
      end

      def reset!
        @lock.synchronize do
          clear_cache!
          reset_transaction
          unless @connection.transaction_status == ::PG::PQTRANS_IDLE
            @connection.query "ROLLBACK"
          end
          @connection.query "DISCARD ALL"
          configure_connection
        end
      end

      # Disconnects from the database if already connected. Otherwise, this
      # method does nothing.
      def disconnect!
        @lock.synchronize do
          super
          @connection.close rescue nil
        end
      end

      def discard! # :nodoc:
        super
        @connection.socket_io.reopen(IO::NULL) rescue nil
        @connection = nil
      end

      def native_database_types # :nodoc:
        self.class.native_database_types
      end

      def self.native_database_types # :nodoc:
        @native_database_types ||= begin
          types = NATIVE_DATABASE_TYPES.dup
          types[:datetime] = types[datetime_type]
          types
        end
      end

      def set_standard_conforming_strings
        execute("SET standard_conforming_strings = on", "SCHEMA")
      end

      def supports_ddl_transactions?
        true
      end

      def supports_advisory_locks?
        true
      end

      def supports_explain?
        true
      end

      def supports_extensions?
        true
      end

      def supports_materialized_views?
        true
      end

      def supports_foreign_tables?
        true
      end

      def supports_pgcrypto_uuid?
        database_version >= 90400 # >= 9.4
      end

      def supports_optimizer_hints?
        unless defined?(@has_pg_hint_plan)
          @has_pg_hint_plan = extension_available?("pg_hint_plan")
        end
        @has_pg_hint_plan
      end

      def supports_common_table_expressions?
        true
      end

      def supports_lazy_transactions?
        true
      end

      def get_advisory_lock(lock_id) # :nodoc:
        unless lock_id.is_a?(Integer) && lock_id.bit_length <= 63
          raise(ArgumentError, "PostgreSQL requires advisory lock ids to be a signed 64 bit integer")
        end
        query_value("SELECT pg_try_advisory_lock(#{lock_id})")
      end

      def release_advisory_lock(lock_id) # :nodoc:
        unless lock_id.is_a?(Integer) && lock_id.bit_length <= 63
          raise(ArgumentError, "PostgreSQL requires advisory lock ids to be a signed 64 bit integer")
        end
        query_value("SELECT pg_advisory_unlock(#{lock_id})")
      end

      def enable_extension(name)
        exec_query("CREATE EXTENSION IF NOT EXISTS \"#{name}\"").tap {
          reload_type_map
        }
      end

      def disable_extension(name)
        exec_query("DROP EXTENSION IF EXISTS \"#{name}\" CASCADE").tap {
          reload_type_map
        }
      end

      def extension_available?(name)
        query_value("SELECT true FROM pg_available_extensions WHERE name = #{quote(name)}", "SCHEMA")
      end

      def extension_enabled?(name)
        query_value("SELECT installed_version IS NOT NULL FROM pg_available_extensions WHERE name = #{quote(name)}", "SCHEMA")
      end

      def extensions
        exec_query("SELECT extname FROM pg_extension", "SCHEMA").cast_values
      end

      # Returns a list of defined enum types, and their values.
      def enum_types
        query = <<~SQL
          SELECT
            type.typname AS name,
            string_agg(enum.enumlabel, ',' ORDER BY enum.enumsortorder) AS value
          FROM pg_enum AS enum
          JOIN pg_type AS type
            ON (type.oid = enum.enumtypid)
          GROUP BY type.typname;
        SQL
        exec_query(query, "SCHEMA").cast_values
      end

      # Given a name and an array of values, creates an enum type.
      def create_enum(name, values)
        sql_values = values.map { |s| "'#{s}'" }.join(", ")
        query = <<~SQL
          DO $$
          BEGIN
              IF NOT EXISTS (
                SELECT 1 FROM pg_type t
                WHERE t.typname = '#{name}'
              ) THEN
                  CREATE TYPE \"#{name}\" AS ENUM (#{sql_values});
              END IF;
          END
          $$;
        SQL
        exec_query(query)
      end

      # Returns the configured supported identifier length supported by PostgreSQL
      def max_identifier_length
        @max_identifier_length ||= query_value("SHOW max_identifier_length", "SCHEMA").to_i
      end

      # Set the authorized user for this session
      def session_auth=(user)
        clear_cache!
        execute("SET SESSION AUTHORIZATION #{user}")
      end

      def use_insert_returning?
        @use_insert_returning
      end

      # Returns the version of the connected PostgreSQL server.
      def get_database_version # :nodoc:
        @connection.server_version
      end
      alias :postgresql_version :database_version

      def default_index_type?(index) # :nodoc:
        index.using == :btree || super
      end

      def build_insert_sql(insert) # :nodoc:
        sql = +"INSERT #{insert.into} #{insert.values_list}"

        if insert.skip_duplicates?
          sql << " ON CONFLICT #{insert.conflict_target} DO NOTHING"
        elsif insert.update_duplicates?
          sql << " ON CONFLICT #{insert.conflict_target} DO UPDATE SET "
          if insert.raw_update_sql?
            sql << insert.raw_update_sql
          else
            sql << insert.touch_model_timestamps_unless { |column| "#{insert.model.quoted_table_name}.#{column} IS NOT DISTINCT FROM excluded.#{column}" }
            sql << insert.updatable_columns.map { |column| "#{column}=excluded.#{column}" }.join(",")
          end
        end

        sql << " RETURNING #{insert.returning}" if insert.returning
        sql
      end

      def check_version # :nodoc:
        if database_version < 90300 # < 9.3
          raise "Your version of PostgreSQL (#{database_version}) is too old. Active Record supports PostgreSQL >= 9.3."
        end
      end

      class << self
        def initialize_type_map(m) # :nodoc:
          m.register_type "int2", Type::Integer.new(limit: 2)
          m.register_type "int4", Type::Integer.new(limit: 4)
          m.register_type "int8", Type::Integer.new(limit: 8)
          m.register_type "oid", OID::Oid.new
          m.register_type "float4", Type::Float.new
          m.alias_type "float8", "float4"
          m.register_type "text", Type::Text.new
          register_class_with_limit m, "varchar", Type::String
          m.alias_type "char", "varchar"
          m.alias_type "name", "varchar"
          m.alias_type "bpchar", "varchar"
          m.register_type "bool", Type::Boolean.new
          register_class_with_limit m, "bit", OID::Bit
          register_class_with_limit m, "varbit", OID::BitVarying
          m.register_type "date", OID::Date.new

          m.register_type "money", OID::Money.new
          m.register_type "bytea", OID::Bytea.new
          m.register_type "point", OID::Point.new
          m.register_type "hstore", OID::Hstore.new
          m.register_type "json", Type::Json.new
          m.register_type "jsonb", OID::Jsonb.new
          m.register_type "cidr", OID::Cidr.new
          m.register_type "inet", OID::Inet.new
          m.register_type "uuid", OID::Uuid.new
          m.register_type "xml", OID::Xml.new
          m.register_type "tsvector", OID::SpecializedString.new(:tsvector)
          m.register_type "macaddr", OID::Macaddr.new
          m.register_type "citext", OID::SpecializedString.new(:citext)
          m.register_type "ltree", OID::SpecializedString.new(:ltree)
          m.register_type "line", OID::SpecializedString.new(:line)
          m.register_type "lseg", OID::SpecializedString.new(:lseg)
          m.register_type "box", OID::SpecializedString.new(:box)
          m.register_type "path", OID::SpecializedString.new(:path)
          m.register_type "polygon", OID::SpecializedString.new(:polygon)
          m.register_type "circle", OID::SpecializedString.new(:circle)

          register_class_with_precision m, "time", Type::Time
          register_class_with_precision m, "timestamp", OID::Timestamp
          register_class_with_precision m, "timestamptz", OID::TimestampWithTimeZone

          m.register_type "numeric" do |_, fmod, sql_type|
            precision = extract_precision(sql_type)
            scale = extract_scale(sql_type)

            # The type for the numeric depends on the width of the field,
            # so we'll do something special here.
            #
            # When dealing with decimal columns:
            #
            # places after decimal  = fmod - 4 & 0xffff
            # places before decimal = (fmod - 4) >> 16 & 0xffff
            if fmod && (fmod - 4 & 0xffff).zero?
              # FIXME: Remove this class, and the second argument to
              # lookups on PG
              Type::DecimalWithoutScale.new(precision: precision)
            else
              OID::Decimal.new(precision: precision, scale: scale)
            end
          end

          m.register_type "interval" do |*args, sql_type|
            precision = extract_precision(sql_type)
            OID::Interval.new(precision: precision)
          end
        end
      end

      private
        def type_map
          @type_map ||= Type::HashLookupTypeMap.new
        end

        def initialize_type_map(m = type_map)
          self.class.initialize_type_map(m)
          load_additional_types
        end

        # Extracts the value from a PostgreSQL column default definition.
        def extract_value_from_default(default)
          case default
            # Quoted types
          when /\A[(B]?'(.*)'.*::"?([\w. ]+)"?(?:\[\])?\z/m
            # The default 'now'::date is CURRENT_DATE
            if $1 == "now" && $2 == "date"
              nil
            else
              $1.gsub("''", "'")
            end
            # Boolean types
          when "true", "false"
            default
            # Numeric types
          when /\A\(?(-?\d+(\.\d*)?)\)?(::bigint)?\z/
            $1
            # Object identifier types
          when /\A-?\d+\z/
            $1
          else
            # Anything else is blank, some user type, or some function
            # and we can't know the value of that, so return nil.
            nil
          end
        end

        def extract_default_function(default_value, default)
          default if has_default_function?(default_value, default)
        end

        def has_default_function?(default_value, default)
          !default_value && %r{\w+\(.*\)|\(.*\)::\w+|CURRENT_DATE|CURRENT_TIMESTAMP}.match?(default)
        end

        # See https://www.postgresql.org/docs/current/static/errcodes-appendix.html
        VALUE_LIMIT_VIOLATION = "22001"
        NUMERIC_VALUE_OUT_OF_RANGE = "22003"
        NOT_NULL_VIOLATION    = "23502"
        FOREIGN_KEY_VIOLATION = "23503"
        UNIQUE_VIOLATION      = "23505"
        SERIALIZATION_FAILURE = "40001"
        DEADLOCK_DETECTED     = "40P01"
        DUPLICATE_DATABASE    = "42P04"
        LOCK_NOT_AVAILABLE    = "55P03"
        QUERY_CANCELED        = "57014"

        def translate_exception(exception, message:, sql:, binds:)
          return exception unless exception.respond_to?(:result)

          case exception.result.try(:error_field, PG::PG_DIAG_SQLSTATE)
          when nil
            if exception.message.match?(/connection is closed/i)
              ConnectionNotEstablished.new(exception)
            else
              super
            end
          when UNIQUE_VIOLATION
            RecordNotUnique.new(message, sql: sql, binds: binds)
          when FOREIGN_KEY_VIOLATION
            InvalidForeignKey.new(message, sql: sql, binds: binds)
          when VALUE_LIMIT_VIOLATION
            ValueTooLong.new(message, sql: sql, binds: binds)
          when NUMERIC_VALUE_OUT_OF_RANGE
            RangeError.new(message, sql: sql, binds: binds)
          when NOT_NULL_VIOLATION
            NotNullViolation.new(message, sql: sql, binds: binds)
          when SERIALIZATION_FAILURE
            SerializationFailure.new(message, sql: sql, binds: binds)
          when DEADLOCK_DETECTED
            Deadlocked.new(message, sql: sql, binds: binds)
          when DUPLICATE_DATABASE
            DatabaseAlreadyExists.new(message, sql: sql, binds: binds)
          when LOCK_NOT_AVAILABLE
            LockWaitTimeout.new(message, sql: sql, binds: binds)
          when QUERY_CANCELED
            QueryCanceled.new(message, sql: sql, binds: binds)
          else
            super
          end
        end

        def get_oid_type(oid, fmod, column_name, sql_type = "")
          if !type_map.key?(oid)
            load_additional_types([oid])
          end

          type_map.fetch(oid, fmod, sql_type) {
            warn "unknown OID #{oid}: failed to recognize type of '#{column_name}'. It will be treated as String."
            Type.default_value.tap do |cast_type|
              type_map.register_type(oid, cast_type)
            end
          }
        end

        def load_additional_types(oids = nil)
          initializer = OID::TypeMapInitializer.new(type_map)
          load_types_queries(initializer, oids) do |query|
            execute_and_clear(query, "SCHEMA", []) do |records|
              initializer.run(records)
            end
          end
        end

        def load_types_queries(initializer, oids)
          query = <<~SQL
            SELECT t.oid, t.typname, t.typelem, t.typdelim, t.typinput, r.rngsubtype, t.typtype, t.typbasetype
            FROM pg_type as t
            LEFT JOIN pg_range as r ON oid = rngtypid
          SQL
          if oids
            yield query + "WHERE t.oid IN (%s)" % oids.join(", ")
          else
            yield query + initializer.query_conditions_for_known_type_names
            yield query + initializer.query_conditions_for_known_type_types
            yield query + initializer.query_conditions_for_array_types
          end
        end

        FEATURE_NOT_SUPPORTED = "0A000" # :nodoc:

        def execute_and_clear(sql, name, binds, prepare: false, async: false)
          sql = transform_query(sql)
          check_if_write_query(sql)

          if !prepare || without_prepared_statement?(binds)
            result = exec_no_cache(sql, name, binds, async: async)
          else
            result = exec_cache(sql, name, binds, async: async)
          end
          begin
            ret = yield result
          ensure
            result.clear
          end
          ret
        end

        def exec_no_cache(sql, name, binds, async: false)
          materialize_transactions
          mark_transaction_written_if_write(sql)

          # make sure we carry over any changes to ActiveRecord.default_timezone that have been
          # made since we established the connection
          update_typemap_for_default_timezone

          type_casted_binds = type_casted_binds(binds)
          log(sql, name, binds, type_casted_binds, async: async) do
            ActiveSupport::Dependencies.interlock.permit_concurrent_loads do
              @connection.exec_params(sql, type_casted_binds)
            end
          end
        end

        def exec_cache(sql, name, binds, async: false)
          materialize_transactions
          mark_transaction_written_if_write(sql)
          update_typemap_for_default_timezone

          stmt_key = prepare_statement(sql, binds)
          type_casted_binds = type_casted_binds(binds)

          log(sql, name, binds, type_casted_binds, stmt_key, async: async) do
            ActiveSupport::Dependencies.interlock.permit_concurrent_loads do
              @connection.exec_prepared(stmt_key, type_casted_binds)
            end
          end
        rescue ActiveRecord::StatementInvalid => e
          raise unless is_cached_plan_failure?(e)

          # Nothing we can do if we are in a transaction because all commands
          # will raise InFailedSQLTransaction
          if in_transaction?
            raise ActiveRecord::PreparedStatementCacheExpired.new(e.cause.message)
          else
            @lock.synchronize do
              # outside of transactions we can simply flush this query and retry
              @statements.delete sql_key(sql)
            end
            retry
          end
        end

        # Annoyingly, the code for prepared statements whose return value may
        # have changed is FEATURE_NOT_SUPPORTED.
        #
        # This covers various different error types so we need to do additional
        # work to classify the exception definitively as a
        # ActiveRecord::PreparedStatementCacheExpired
        #
        # Check here for more details:
        # https://git.postgresql.org/gitweb/?p=postgresql.git;a=blob;f=src/backend/utils/cache/plancache.c#l573
        def is_cached_plan_failure?(e)
          pgerror = e.cause
          pgerror.result.result_error_field(PG::PG_DIAG_SQLSTATE) == FEATURE_NOT_SUPPORTED &&
            pgerror.result.result_error_field(PG::PG_DIAG_SOURCE_FUNCTION) == "RevalidateCachedQuery"
        rescue
          false
        end

        def in_transaction?
          open_transactions > 0
        end

        # Returns the statement identifier for the client side cache
        # of statements
        def sql_key(sql)
          "#{schema_search_path}-#{sql}"
        end

        # Prepare the statement if it hasn't been prepared, return
        # the statement key.
        def prepare_statement(sql, binds)
          @lock.synchronize do
            sql_key = sql_key(sql)
            unless @statements.key? sql_key
              nextkey = @statements.next_key
              begin
                @connection.prepare nextkey, sql
              rescue => e
                raise translate_exception_class(e, sql, binds)
              end
              # Clear the queue
              @connection.get_last_result
              @statements[sql_key] = nextkey
            end
            @statements[sql_key]
          end
        end

        # Connects to a PostgreSQL server and sets up the adapter depending on the
        # connected server's characteristics.
        def connect
          @connection = self.class.new_client(@connection_parameters)
          configure_connection
          add_pg_encoders
          add_pg_decoders
        end

        # Configures the encoding, verbosity, schema search path, and time zone of the connection.
        # This is called by #connect and should not be called manually.
        def configure_connection
          if @config[:encoding]
            @connection.set_client_encoding(@config[:encoding])
          end
          self.client_min_messages = @config[:min_messages] || "warning"
          self.schema_search_path = @config[:schema_search_path] || @config[:schema_order]

          # Use standard-conforming strings so we don't have to do the E'...' dance.
          set_standard_conforming_strings

          variables = @config.fetch(:variables, {}).stringify_keys

          # If using Active Record's time zone support configure the connection to return
          # TIMESTAMP WITH ZONE types in UTC.
          unless variables["timezone"]
            if ActiveRecord.default_timezone == :utc
              variables["timezone"] = "UTC"
            elsif @local_tz
              variables["timezone"] = @local_tz
            end
          end

          # Set interval output format to ISO 8601 for ease of parsing by ActiveSupport::Duration.parse
          execute("SET intervalstyle = iso_8601", "SCHEMA")

          # SET statements from :variables config hash
          # https://www.postgresql.org/docs/current/static/sql-set.html
          variables.map do |k, v|
            if v == ":default" || v == :default
              # Sets the value to the global or compile default
              execute("SET SESSION #{k} TO DEFAULT", "SCHEMA")
            elsif !v.nil?
              execute("SET SESSION #{k} TO #{quote(v)}", "SCHEMA")
            end
          end
        end

        # Returns the list of a table's column names, data types, and default values.
        #
        # The underlying query is roughly:
        #  SELECT column.name, column.type, default.value, column.comment
        #    FROM column LEFT JOIN default
        #      ON column.table_id = default.table_id
        #     AND column.num = default.column_num
        #   WHERE column.table_id = get_table_id('table_name')
        #     AND column.num > 0
        #     AND NOT column.is_dropped
        #   ORDER BY column.num
        #
        # If the table name is not prefixed with a schema, the database will
        # take the first match from the schema search path.
        #
        # Query implementation notes:
        #  - format_type includes the column size constraint, e.g. varchar(50)
        #  - ::regclass is a function that gives the id for a table name
        def column_definitions(table_name)
          query(<<~SQL, "SCHEMA")
              SELECT a.attname, format_type(a.atttypid, a.atttypmod),
                     pg_get_expr(d.adbin, d.adrelid), a.attnotnull, a.atttypid, a.atttypmod,
                     c.collname, col_description(a.attrelid, a.attnum) AS comment,
                     #{supports_virtual_columns? ? 'attgenerated' : quote('')} as attgenerated
                FROM pg_attribute a
                LEFT JOIN pg_attrdef d ON a.attrelid = d.adrelid AND a.attnum = d.adnum
                LEFT JOIN pg_type t ON a.atttypid = t.oid
                LEFT JOIN pg_collation c ON a.attcollation = c.oid AND a.attcollation <> t.typcollation
               WHERE a.attrelid = #{quote(quote_table_name(table_name))}::regclass
                 AND a.attnum > 0 AND NOT a.attisdropped
               ORDER BY a.attnum
          SQL
        end

        def extract_table_ref_from_insert_sql(sql)
          sql[/into\s("[A-Za-z0-9_."\[\]\s]+"|[A-Za-z0-9_."\[\]]+)\s*/im]
          $1.strip if $1
        end

        def arel_visitor
          Arel::Visitors::PostgreSQL.new(self)
        end

        def build_statement_pool
          StatementPool.new(@connection, self.class.type_cast_config_to_integer(@config[:statement_limit]))
        end

        def can_perform_case_insensitive_comparison_for?(column)
          @case_insensitive_cache ||= {}
          @case_insensitive_cache[column.sql_type] ||= begin
            sql = <<~SQL
              SELECT exists(
                SELECT * FROM pg_proc
                WHERE proname = 'lower'
                  AND proargtypes = ARRAY[#{quote column.sql_type}::regtype]::oidvector
              ) OR exists(
                SELECT * FROM pg_proc
                INNER JOIN pg_cast
                  ON ARRAY[casttarget]::oidvector = proargtypes
                WHERE proname = 'lower'
                  AND castsource = #{quote column.sql_type}::regtype
              )
            SQL
            execute_and_clear(sql, "SCHEMA", []) do |result|
              result.getvalue(0, 0)
            end
          end
        end

        def add_pg_encoders
          map = PG::TypeMapByClass.new
          map[Integer] = PG::TextEncoder::Integer.new
          map[TrueClass] = PG::TextEncoder::Boolean.new
          map[FalseClass] = PG::TextEncoder::Boolean.new
          @connection.type_map_for_queries = map
        end

        def update_typemap_for_default_timezone
          if @default_timezone != ActiveRecord.default_timezone && @timestamp_decoder
            decoder_class = ActiveRecord.default_timezone == :utc ?
              PG::TextDecoder::TimestampUtc :
              PG::TextDecoder::TimestampWithoutTimeZone

            @timestamp_decoder = decoder_class.new(**@timestamp_decoder.to_h)
            @connection.type_map_for_results.add_coder(@timestamp_decoder)

            @default_timezone = ActiveRecord.default_timezone

            # if default timezone has changed, we need to reconfigure the connection
            # (specifically, the session time zone)
            configure_connection
          end
        end

        def add_pg_decoders
          @default_timezone = nil
          @timestamp_decoder = nil

          coders_by_name = {
            "int2" => PG::TextDecoder::Integer,
            "int4" => PG::TextDecoder::Integer,
            "int8" => PG::TextDecoder::Integer,
            "oid" => PG::TextDecoder::Integer,
            "float4" => PG::TextDecoder::Float,
            "float8" => PG::TextDecoder::Float,
            "numeric" => PG::TextDecoder::Numeric,
            "bool" => PG::TextDecoder::Boolean,
            "timestamp" => PG::TextDecoder::TimestampUtc,
            "timestamptz" => PG::TextDecoder::TimestampWithTimeZone,
          }

          known_coder_types = coders_by_name.keys.map { |n| quote(n) }
          query = <<~SQL % known_coder_types.join(", ")
            SELECT t.oid, t.typname
            FROM pg_type as t
            WHERE t.typname IN (%s)
          SQL
          coders = execute_and_clear(query, "SCHEMA", []) do |result|
            result.filter_map { |row| construct_coder(row, coders_by_name[row["typname"]]) }
          end

          map = PG::TypeMapByOid.new
          coders.each { |coder| map.add_coder(coder) }
          @connection.type_map_for_results = map

          @type_map_for_results = PG::TypeMapByOid.new
          @type_map_for_results.default_type_map = map
          @type_map_for_results.add_coder(PG::TextDecoder::Bytea.new(oid: 17, name: "bytea"))
          @type_map_for_results.add_coder(MoneyDecoder.new(oid: 790, name: "money"))

          # extract timestamp decoder for use in update_typemap_for_default_timezone
          @timestamp_decoder = coders.find { |coder| coder.name == "timestamp" }
          update_typemap_for_default_timezone
        end

        def construct_coder(row, coder_class)
          return unless coder_class
          coder_class.new(oid: row["oid"].to_i, name: row["typname"])
        end

        class MoneyDecoder < PG::SimpleDecoder # :nodoc:
          TYPE = OID::Money.new

          def decode(value, tuple = nil, field = nil)
            TYPE.deserialize(value)
          end
        end

        ActiveRecord::Type.add_modifier({ array: true }, OID::Array, adapter: :postgresql)
        ActiveRecord::Type.add_modifier({ range: true }, OID::Range, adapter: :postgresql)
        ActiveRecord::Type.register(:bit, OID::Bit, adapter: :postgresql)
        ActiveRecord::Type.register(:bit_varying, OID::BitVarying, adapter: :postgresql)
        ActiveRecord::Type.register(:binary, OID::Bytea, adapter: :postgresql)
        ActiveRecord::Type.register(:cidr, OID::Cidr, adapter: :postgresql)
        ActiveRecord::Type.register(:date, OID::Date, adapter: :postgresql)
        ActiveRecord::Type.register(:datetime, OID::DateTime, adapter: :postgresql)
        ActiveRecord::Type.register(:decimal, OID::Decimal, adapter: :postgresql)
        ActiveRecord::Type.register(:enum, OID::Enum, adapter: :postgresql)
        ActiveRecord::Type.register(:hstore, OID::Hstore, adapter: :postgresql)
        ActiveRecord::Type.register(:inet, OID::Inet, adapter: :postgresql)
        ActiveRecord::Type.register(:interval, OID::Interval, adapter: :postgresql)
        ActiveRecord::Type.register(:jsonb, OID::Jsonb, adapter: :postgresql)
        ActiveRecord::Type.register(:money, OID::Money, adapter: :postgresql)
        ActiveRecord::Type.register(:point, OID::Point, adapter: :postgresql)
        ActiveRecord::Type.register(:legacy_point, OID::LegacyPoint, adapter: :postgresql)
        ActiveRecord::Type.register(:uuid, OID::Uuid, adapter: :postgresql)
        ActiveRecord::Type.register(:vector, OID::Vector, adapter: :postgresql)
        ActiveRecord::Type.register(:xml, OID::Xml, adapter: :postgresql)
    end
    ActiveSupport.run_load_hooks(:active_record_postgresqladapter, PostgreSQLAdapter)
  end
end