# -*- ruby -*- require 'pg' unless defined?( PG ) # This module defines the mapping between OID and encoder/decoder classes for PG::BasicTypeMapForResults, PG::BasicTypeMapForQueries and PG::BasicTypeMapBasedOnResult. # # Additional types can be added like so: # # require 'pg' # require 'ipaddr' # # class InetDecoder < PG::SimpleDecoder # def decode(string, tuple=nil, field=nil) # IPAddr.new(string) # end # end # class InetEncoder < PG::SimpleEncoder # def encode(ip_addr) # ip_addr.to_s # end # end # # # 0 if for text format, can also be 1 for binary # PG::BasicTypeRegistry.register_type(0, 'inet', InetEncoder, InetDecoder) module PG::BasicTypeRegistry # An instance of this class stores the coders that should be used for a given wire format (text or binary) # and type cast direction (encoder or decoder). class CoderMap # Hash of text types that don't require quotation, when used within composite types. # type.name => true DONT_QUOTE_TYPES = %w[ int2 int4 int8 float4 float8 oid bool date timestamp timestamptz ].inject({}){|h,e| h[e] = true; h } def initialize(result, coders_by_name, format, arraycoder) coder_map = {} _ranges, nodes = result.partition { |row| row['typinput'] == 'range_in' } leaves, nodes = nodes.partition { |row| row['typelem'].to_i == 0 } arrays, nodes = nodes.partition { |row| row['typinput'] == 'array_in' } # populate the enum types _enums, leaves = leaves.partition { |row| row['typinput'] == 'enum_in' } # enums.each do |row| # coder_map[row['oid'].to_i] = OID::Enum.new # end # populate the base types leaves.find_all { |row| coders_by_name.key?(row['typname']) }.each do |row| coder = coders_by_name[row['typname']].dup coder.oid = row['oid'].to_i coder.name = row['typname'] coder.format = format coder_map[coder.oid] = coder end _records_by_oid = result.group_by { |row| row['oid'] } # populate composite types # nodes.each do |row| # add_oid row, records_by_oid, coder_map # end if arraycoder # populate array types arrays.each do |row| elements_coder = coder_map[row['typelem'].to_i] next unless elements_coder coder = arraycoder.new coder.oid = row['oid'].to_i coder.name = row['typname'] coder.format = format coder.elements_type = elements_coder coder.needs_quotation = !DONT_QUOTE_TYPES[elements_coder.name] coder_map[coder.oid] = coder end end # populate range types # ranges.find_all { |row| coder_map.key? row['rngsubtype'].to_i }.each do |row| # subcoder = coder_map[row['rngsubtype'].to_i] # range = OID::Range.new subcoder # coder_map[row['oid'].to_i] = range # end @coders = coder_map.values @coders_by_name = @coders.inject({}){|h, t| h[t.name] = t; h } @coders_by_oid = @coders.inject({}){|h, t| h[t.oid] = t; h } @typenames_by_oid = result.inject({}){|h, t| h[t['oid'].to_i] = t['typname']; h } end attr_reader :coders attr_reader :coders_by_oid attr_reader :coders_by_name attr_reader :typenames_by_oid def coder_by_name(name) @coders_by_name[name] end def coder_by_oid(oid) @coders_by_oid[oid] end end private def supports_ranges?(connection) connection.server_version >= 90200 end def build_coder_maps(connection) if supports_ranges?(connection) result = connection.exec <<-SQL SELECT t.oid, t.typname, t.typelem, t.typdelim, t.typinput, r.rngsubtype FROM pg_type as t LEFT JOIN pg_range as r ON oid = rngtypid SQL else result = connection.exec <<-SQL SELECT t.oid, t.typname, t.typelem, t.typdelim, t.typinput FROM pg_type as t SQL end [ [0, :encoder, PG::TextEncoder::Array], [0, :decoder, PG::TextDecoder::Array], [1, :encoder, nil], [1, :decoder, nil], ].inject([]) do |h, (format, direction, arraycoder)| h[format] ||= {} h[format][direction] = CoderMap.new result, CODERS_BY_NAME[format][direction], format, arraycoder h end end ValidFormats = { 0 => true, 1 => true } ValidDirections = { :encoder => true, :decoder => true } def check_format_and_direction(format, direction) raise(ArgumentError, "Invalid format value %p" % format) unless ValidFormats[format] raise(ArgumentError, "Invalid direction %p" % direction) unless ValidDirections[direction] end protected :check_format_and_direction # The key of this hash maps to the `typname` column from the table. # encoder_map is then dynamically built with oids as the key and Type # objects as values. CODERS_BY_NAME = [] # Register an OID type named +name+ with a typecasting encoder and decoder object in # +type+. +name+ should correspond to the `typname` column in # the `pg_type` table. # +format+ can be 0 for text format and 1 for binary. def self.register_type(format, name, encoder_class, decoder_class) CODERS_BY_NAME[format] ||= { encoder: {}, decoder: {} } CODERS_BY_NAME[format][:encoder][name] = encoder_class.new(name: name, format: format) if encoder_class CODERS_BY_NAME[format][:decoder][name] = decoder_class.new(name: name, format: format) if decoder_class end # Alias the +old+ type to the +new+ type. def self.alias_type(format, new, old) [:encoder, :decoder].each do |ende| enc = CODERS_BY_NAME[format][ende][old] if enc CODERS_BY_NAME[format][ende][new] = enc else CODERS_BY_NAME[format][ende].delete(new) end end end register_type 0, 'int2', PG::TextEncoder::Integer, PG::TextDecoder::Integer alias_type 0, 'int4', 'int2' alias_type 0, 'int8', 'int2' alias_type 0, 'oid', 'int2' register_type 0, 'numeric', PG::TextEncoder::Numeric, PG::TextDecoder::Numeric register_type 0, 'text', PG::TextEncoder::String, PG::TextDecoder::String alias_type 0, 'varchar', 'text' alias_type 0, 'char', 'text' alias_type 0, 'bpchar', 'text' alias_type 0, 'xml', 'text' # FIXME: why are we keeping these types as strings? # alias_type 'tsvector', 'text' # alias_type 'interval', 'text' # alias_type 'macaddr', 'text' # alias_type 'uuid', 'text' # # register_type 'money', OID::Money.new # There is no PG::TextEncoder::Bytea, because it's simple and more efficient to send bytea-data # in binary format, either with PG::BinaryEncoder::Bytea or in Hash param format. register_type 0, 'bytea', nil, PG::TextDecoder::Bytea register_type 0, 'bool', PG::TextEncoder::Boolean, PG::TextDecoder::Boolean # register_type 'bit', OID::Bit.new # register_type 'varbit', OID::Bit.new register_type 0, 'float4', PG::TextEncoder::Float, PG::TextDecoder::Float alias_type 0, 'float8', 'float4' register_type 0, 'timestamp', PG::TextEncoder::TimestampWithoutTimeZone, PG::TextDecoder::TimestampWithoutTimeZone register_type 0, 'timestamptz', PG::TextEncoder::TimestampWithTimeZone, PG::TextDecoder::TimestampWithTimeZone register_type 0, 'date', PG::TextEncoder::Date, PG::TextDecoder::Date # register_type 'time', OID::Time.new # # register_type 'path', OID::Text.new # register_type 'point', OID::Point.new # register_type 'polygon', OID::Text.new # register_type 'circle', OID::Text.new # register_type 'hstore', OID::Hstore.new register_type 0, 'json', PG::TextEncoder::JSON, PG::TextDecoder::JSON alias_type 0, 'jsonb', 'json' # register_type 'citext', OID::Text.new # register_type 'ltree', OID::Text.new # register_type 0, 'inet', PG::TextEncoder::Inet, PG::TextDecoder::Inet alias_type 0, 'cidr', 'inet' register_type 1, 'int2', PG::BinaryEncoder::Int2, PG::BinaryDecoder::Integer register_type 1, 'int4', PG::BinaryEncoder::Int4, PG::BinaryDecoder::Integer register_type 1, 'int8', PG::BinaryEncoder::Int8, PG::BinaryDecoder::Integer alias_type 1, 'oid', 'int2' register_type 1, 'text', PG::BinaryEncoder::String, PG::BinaryDecoder::String alias_type 1, 'varchar', 'text' alias_type 1, 'char', 'text' alias_type 1, 'bpchar', 'text' alias_type 1, 'xml', 'text' register_type 1, 'bytea', PG::BinaryEncoder::Bytea, PG::BinaryDecoder::Bytea register_type 1, 'bool', PG::BinaryEncoder::Boolean, PG::BinaryDecoder::Boolean register_type 1, 'float4', nil, PG::BinaryDecoder::Float register_type 1, 'float8', nil, PG::BinaryDecoder::Float register_type 1, 'timestamp', nil, PG::BinaryDecoder::TimestampUtc register_type 1, 'timestamptz', nil, PG::BinaryDecoder::TimestampUtcToLocal end # Simple set of rules for type casting common PostgreSQL types to Ruby. # # OIDs of supported type casts are not hard-coded in the sources, but are retrieved from the # PostgreSQL's pg_type table in PG::BasicTypeMapForResults.new . # # Result values are type casted based on the type OID of the given result column. # # Higher level libraries will most likely not make use of this class, but use their # own set of rules to choose suitable encoders and decoders. # # Example: # conn = PG::Connection.new # # Assign a default ruleset for type casts of output values. # conn.type_map_for_results = PG::BasicTypeMapForResults.new(conn) # # Execute a query. # res = conn.exec_params( "SELECT $1::INT", ['5'] ) # # Retrieve and cast the result value. Value format is 0 (text) and OID is 20. Therefore typecasting # # is done by PG::TextDecoder::Integer internally for all value retrieval methods. # res.values # => [[5]] # # PG::TypeMapByOid#fit_to_result(result, false) can be used to generate # a result independent PG::TypeMapByColumn type map, which can subsequently be used # to cast #get_copy_data fields: # # For the following table: # conn.exec( "CREATE TABLE copytable AS VALUES('a', 123, '{5,4,3}'::INT[])" ) # # # Retrieve table OIDs per empty result set. # res = conn.exec( "SELECT * FROM copytable LIMIT 0" ) # # Build a type map for common database to ruby type decoders. # btm = PG::BasicTypeMapForResults.new(conn) # # Build a PG::TypeMapByColumn with decoders suitable for copytable. # tm = btm.build_column_map( res ) # row_decoder = PG::TextDecoder::CopyRow.new type_map: tm # # conn.copy_data( "COPY copytable TO STDOUT", row_decoder ) do |res| # while row=conn.get_copy_data # p row # end # end # This prints the rows with type casted columns: # ["a", 123, [5, 4, 3]] # # See also PG::BasicTypeMapBasedOnResult for the encoder direction and PG::BasicTypeRegistry for the definition of additional types. class PG::BasicTypeMapForResults < PG::TypeMapByOid include PG::BasicTypeRegistry class WarningTypeMap < PG::TypeMapInRuby def initialize(typenames) @already_warned = Hash.new{|h, k| h[k] = {} } @typenames_by_oid = typenames end def typecast_result_value(result, _tuple, field) format = result.fformat(field) oid = result.ftype(field) unless @already_warned[format][oid] STDERR.puts "Warning: no type cast defined for type #{@typenames_by_oid[format][oid].inspect} with oid #{oid}. Please cast this type explicitly to TEXT to be safe for future changes." @already_warned[format][oid] = true end super end end def initialize(connection) @coder_maps = build_coder_maps(connection) # Populate TypeMapByOid hash with decoders @coder_maps.map{|f| f[:decoder].coders }.flatten.each do |coder| add_coder(coder) end typenames = @coder_maps.map{|f| f[:decoder].typenames_by_oid } self.default_type_map = WarningTypeMap.new(typenames) end end # Simple set of rules for type casting common PostgreSQL types from Ruby # to PostgreSQL. # # OIDs of supported type casts are not hard-coded in the sources, but are retrieved from the # PostgreSQL's pg_type table in PG::BasicTypeMapBasedOnResult.new . # # This class works equal to PG::BasicTypeMapForResults, but does not define decoders for # the given result OIDs, but encoders. So it can be used to type cast field values based on # the type OID retrieved by a separate SQL query. # # PG::TypeMapByOid#build_column_map(result) can be used to generate a result independent # PG::TypeMapByColumn type map, which can subsequently be used to cast query bind parameters # or #put_copy_data fields. # # Example: # conn.exec( "CREATE TEMP TABLE copytable (t TEXT, i INT, ai INT[])" ) # # # Retrieve table OIDs per empty result set. # res = conn.exec( "SELECT * FROM copytable LIMIT 0" ) # # Build a type map for common ruby to database type encoders. # btm = PG::BasicTypeMapBasedOnResult.new(conn) # # Build a PG::TypeMapByColumn with encoders suitable for copytable. # tm = btm.build_column_map( res ) # row_encoder = PG::TextEncoder::CopyRow.new type_map: tm # # conn.copy_data( "COPY copytable FROM STDIN", row_encoder ) do |res| # conn.put_copy_data ['a', 123, [5,4,3]] # end # This inserts a single row into copytable with type casts from ruby to # database types. class PG::BasicTypeMapBasedOnResult < PG::TypeMapByOid include PG::BasicTypeRegistry def initialize(connection) @coder_maps = build_coder_maps(connection) # Populate TypeMapByOid hash with encoders @coder_maps.map{|f| f[:encoder].coders }.flatten.each do |coder| add_coder(coder) end end end # Simple set of rules for type casting common Ruby types to PostgreSQL. # # OIDs of supported type casts are not hard-coded in the sources, but are retrieved from the # PostgreSQL's pg_type table in PG::BasicTypeMapForQueries.new . # # Query params are type casted based on the class of the given value. # # Higher level libraries will most likely not make use of this class, but use their # own derivation of PG::TypeMapByClass or another set of rules to choose suitable # encoders and decoders for the values to be sent. # # Example: # conn = PG::Connection.new # # Assign a default ruleset for type casts of input and output values. # conn.type_map_for_queries = PG::BasicTypeMapForQueries.new(conn) # # Execute a query. The Integer param value is typecasted internally by PG::BinaryEncoder::Int8. # # The format of the parameter is set to 1 (binary) and the OID of this parameter is set to 20 (int8). # res = conn.exec_params( "SELECT $1", [5] ) class PG::BasicTypeMapForQueries < PG::TypeMapByClass include PG::BasicTypeRegistry def initialize(connection) @coder_maps = build_coder_maps(connection) populate_encoder_list @array_encoders_by_klass = array_encoders_by_klass @anyarray_encoder = coder_by_name(0, :encoder, '_any') end private def coder_by_name(format, direction, name) check_format_and_direction(format, direction) @coder_maps[format][direction].coder_by_name(name) end def populate_encoder_list DEFAULT_TYPE_MAP.each do |klass, selector| if Array === selector format, name, oid_name = selector coder = coder_by_name(format, :encoder, name).dup if oid_name coder.oid = coder_by_name(format, :encoder, oid_name).oid else coder.oid = 0 end self[klass] = coder else self[klass] = selector end end end def array_encoders_by_klass DEFAULT_ARRAY_TYPE_MAP.inject({}) do |h, (klass, (format, name))| h[klass] = coder_by_name(format, :encoder, name) h end end def get_array_type(value) elem = value while elem.kind_of?(Array) elem = elem.first end @array_encoders_by_klass[elem.class] || elem.class.ancestors.lazy.map{|ancestor| @array_encoders_by_klass[ancestor] }.find{|a| a } || @anyarray_encoder end DEFAULT_TYPE_MAP = { TrueClass => [1, 'bool', 'bool'], FalseClass => [1, 'bool', 'bool'], # We use text format and no type OID for numbers, because setting the OID can lead # to unnecessary type conversions on server side. Integer => [0, 'int8'], Float => [0, 'float8'], BigDecimal => [0, 'numeric'], # We use text format and no type OID for IPAddr, because setting the OID can lead # to unnecessary inet/cidr conversions on the server side. IPAddr => [0, 'inet'], Array => :get_array_type, } DEFAULT_ARRAY_TYPE_MAP = { TrueClass => [0, '_bool'], FalseClass => [0, '_bool'], Integer => [0, '_int8'], String => [0, '_text'], Float => [0, '_float8'], } end