psycopg2/lib/extras.py

"""Miscellaneous goodies for psycopg2

This module is a generic place used to hold little helper functions
and classes untill a better place in the distribution is found.
"""
# psycopg/extras.py - miscellaneous extra goodies for psycopg
#
# Copyright (C) 2003-2010 Federico Di Gregorio  <fog@debian.org>
#
# psycopg2 is free software: you can redistribute it and/or modify it
# under the terms of the GNU Lesser General Public License as published
# by the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# In addition, as a special exception, the copyright holders give
# permission to link this program with the OpenSSL library (or with
# modified versions of OpenSSL that use the same license as OpenSSL),
# and distribute linked combinations including the two.
#
# You must obey the GNU Lesser General Public License in all respects for
# all of the code used other than OpenSSL.
#
# psycopg2 is distributed in the hope that it will be useful, but WITHOUT
# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
# FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
# License for more details.

import os
import sys
import time
import warnings
import re as regex

try:
    import logging
except:
    logging = None

import psycopg2
from psycopg2 import extensions as _ext
from psycopg2.extensions import cursor as _cursor
from psycopg2.extensions import connection as _connection
from psycopg2.extensions import adapt as _A
from psycopg2.extensions import b


class DictCursorBase(_cursor):
    """Base class for all dict-like cursors."""

    def __init__(self, *args, **kwargs):
        if 'row_factory' in kwargs:
            row_factory = kwargs['row_factory']
            del kwargs['row_factory']
        else:
            raise NotImplementedError(
                "DictCursorBase can't be instantiated without a row factory.")
        super(DictCursorBase, self).__init__(*args, **kwargs)
        self._query_executed = 0
        self._prefetch = 0
        self.row_factory = row_factory

    def fetchone(self):
        if self._prefetch:
            res = super(DictCursorBase, self).fetchone()
        if self._query_executed:
            self._build_index()
        if not self._prefetch:
            res = super(DictCursorBase, self).fetchone()
        return res

    def fetchmany(self, size=None):
        if self._prefetch:
            res = super(DictCursorBase, self).fetchmany(size)
        if self._query_executed:
            self._build_index()
        if not self._prefetch:
            res = super(DictCursorBase, self).fetchmany(size)
        return res

    def fetchall(self):
        if self._prefetch:
            res = super(DictCursorBase, self).fetchall()
        if self._query_executed:
            self._build_index()
        if not self._prefetch:
            res = super(DictCursorBase, self).fetchall()
        return res

    def __iter__(self):
        if self._prefetch:
            res = super(DictCursorBase, self).__iter__()
            first = res.next()
        if self._query_executed:
            self._build_index()
        if not self._prefetch:
            res = super(DictCursorBase, self).__iter__()
            first = res.next()

        yield first
        while 1:
            yield res.next()


class DictConnection(_connection):
    """A connection that uses `DictCursor` automatically."""
    def cursor(self, *args, **kwargs):
        kwargs.setdefault('cursor_factory', DictCursor)
        return super(DictConnection, self).cursor(*args, **kwargs)

class DictCursor(DictCursorBase):
    """A cursor that keeps a list of column name -> index mappings."""

    def __init__(self, *args, **kwargs):
        kwargs['row_factory'] = DictRow
        super(DictCursor, self).__init__(*args, **kwargs)
        self._prefetch = 1

    def execute(self, query, vars=None):
        self.index = {}
        self._query_executed = 1
        return super(DictCursor, self).execute(query, vars)

    def callproc(self, procname, vars=None):
        self.index = {}
        self._query_executed = 1
        return super(DictCursor, self).callproc(procname, vars)

    def _build_index(self):
        if self._query_executed == 1 and self.description:
            for i in range(len(self.description)):
                self.index[self.description[i][0]] = i
            self._query_executed = 0

class DictRow(list):
    """A row object that allow by-colmun-name access to data."""

    __slots__ = ('_index',)

    def __init__(self, cursor):
        self._index = cursor.index
        self[:] = [None] * len(cursor.description)

    def __getitem__(self, x):
        if not isinstance(x, (int, slice)):
            x = self._index[x]
        return list.__getitem__(self, x)

    def __setitem__(self, x, v):
        if not isinstance(x, (int, slice)):
            x = self._index[x]
        list.__setitem__(self, x, v)

    def items(self):
        return list(self.iteritems())

    def keys(self):
        return self._index.keys()

    def values(self):
        return tuple(self[:])

    def has_key(self, x):
        return x in self._index

    def get(self, x, default=None):
        try:
            return self[x]
        except:
            return default

    def iteritems(self):
        for n, v in self._index.iteritems():
            yield n, list.__getitem__(self, v)

    def iterkeys(self):
        return self._index.iterkeys()

    def itervalues(self):
        return list.__iter__(self)

    def copy(self):
        return dict(self.iteritems())

    def __contains__(self, x):
        return x in self._index

    # grop the crusty Py2 methods
    if sys.version_info[0] > 2:
        items = iteritems; del iteritems
        keys = iterkeys; del iterkeys
        values = itervalues; del itervalues
        del has_key


class RealDictConnection(_connection):
    """A connection that uses `RealDictCursor` automatically."""
    def cursor(self, *args, **kwargs):
        kwargs.setdefault('cursor_factory', RealDictCursor)
        return super(RealDictConnection, self).cursor(*args, **kwargs)

class RealDictCursor(DictCursorBase):
    """A cursor that uses a real dict as the base type for rows.

    Note that this cursor is extremely specialized and does not allow
    the normal access (using integer indices) to fetched data. If you need
    to access database rows both as a dictionary and a list, then use
    the generic `DictCursor` instead of `!RealDictCursor`.
    """
    def __init__(self, *args, **kwargs):
        kwargs['row_factory'] = RealDictRow
        super(RealDictCursor, self).__init__(*args, **kwargs)
        self._prefetch = 0

    def execute(self, query, vars=None):
        self.column_mapping = []
        self._query_executed = 1
        return super(RealDictCursor, self).execute(query, vars)

    def callproc(self, procname, vars=None):
        self.column_mapping = []
        self._query_executed = 1
        return super(RealDictCursor, self).callproc(procname, vars)

    def _build_index(self):
        if self._query_executed == 1 and self.description:
            for i in range(len(self.description)):
                self.column_mapping.append(self.description[i][0])
            self._query_executed = 0

class RealDictRow(dict):
    """A `!dict` subclass representing a data record."""

    __slots__ = ('_column_mapping')

    def __init__(self, cursor):
        dict.__init__(self)
        # Required for named cursors
        if cursor.description and not cursor.column_mapping:
            cursor._build_index()

        self._column_mapping = cursor.column_mapping

    def __setitem__(self, name, value):
        if type(name) == int:
            name = self._column_mapping[name]
        return dict.__setitem__(self, name, value)


class NamedTupleConnection(_connection):
    """A connection that uses `NamedTupleCursor` automatically."""
    def cursor(self, *args, **kwargs):
        kwargs.setdefault('cursor_factory', NamedTupleCursor)
        return super(NamedTupleConnection, self).cursor(*args, **kwargs)

class NamedTupleCursor(_cursor):
    """A cursor that generates results as `~collections.namedtuple`.

    `!fetch*()` methods will return named tuples instead of regular tuples, so
    their elements can be accessed both as regular numeric items as well as
    attributes.

        >>> nt_cur = conn.cursor(cursor_factory=psycopg2.extras.NamedTupleCursor)
        >>> rec = nt_cur.fetchone()
        >>> rec
        Record(id=1, num=100, data="abc'def")
        >>> rec[1]
        100
        >>> rec.data
        "abc'def"
    """
    Record = None

    def execute(self, query, vars=None):
        self.Record = None
        return super(NamedTupleCursor, self).execute(query, vars)

    def executemany(self, query, vars):
        self.Record = None
        return super(NamedTupleCursor, self).executemany(query, vars)

    def callproc(self, procname, vars=None):
        self.Record = None
        return super(NamedTupleCursor, self).callproc(procname, vars)

    def fetchone(self):
        t = super(NamedTupleCursor, self).fetchone()
        if t is not None:
            nt = self.Record
            if nt is None:
                nt = self.Record = self._make_nt()
            return nt(*t)

    def fetchmany(self, size=None):
        ts = super(NamedTupleCursor, self).fetchmany(size)
        nt = self.Record
        if nt is None:
            nt = self.Record = self._make_nt()
        return [nt(*t) for t in ts]

    def fetchall(self):
        ts = super(NamedTupleCursor, self).fetchall()
        nt = self.Record
        if nt is None:
            nt = self.Record = self._make_nt()
        return [nt(*t) for t in ts]

    def __iter__(self):
        it = super(NamedTupleCursor, self).__iter__()
        t = it.next()

        nt = self.Record
        if nt is None:
            nt = self.Record = self._make_nt()

        yield nt(*t)

        while 1:
            yield nt(*it.next())

    try:
        from collections import namedtuple
    except ImportError, _exc:
        def _make_nt(self):
            raise self._exc
    else:
        def _make_nt(self, namedtuple=namedtuple):
            return namedtuple("Record", [d[0] for d in self.description or ()])


class LoggingConnection(_connection):
    """A connection that logs all queries to a file or logger__ object.

    .. __: http://docs.python.org/library/logging.html
    """

    def initialize(self, logobj):
        """Initialize the connection to log to `!logobj`.

        The `!logobj` parameter can be an open file object or a Logger
        instance from the standard logging module.
        """
        self._logobj = logobj
        if logging and isinstance(logobj, logging.Logger):
            self.log = self._logtologger
        else:
            self.log = self._logtofile

    def filter(self, msg, curs):
        """Filter the query before logging it.

        This is the method to overwrite to filter unwanted queries out of the
        log or to add some extra data to the output. The default implementation
        just does nothing.
        """
        return msg

    def _logtofile(self, msg, curs):
        msg = self.filter(msg, curs)
        if msg: self._logobj.write(msg + os.linesep)

    def _logtologger(self, msg, curs):
        msg = self.filter(msg, curs)
        if msg: self._logobj.debug(msg)

    def _check(self):
        if not hasattr(self, '_logobj'):
            raise self.ProgrammingError(
                "LoggingConnection object has not been initialize()d")

    def cursor(self, *args, **kwargs):
        self._check()
        kwargs.setdefault('cursor_factory', LoggingCursor)
        return super(LoggingConnection, self).cursor(*args, **kwargs)

class LoggingCursor(_cursor):
    """A cursor that logs queries using its connection logging facilities."""

    def execute(self, query, vars=None):
        try:
            return super(LoggingCursor, self).execute(query, vars)
        finally:
            self.connection.log(self.query, self)

    def callproc(self, procname, vars=None):
        try:
            return super(LoggingCursor, self).callproc(procname, vars)
        finally:
            self.connection.log(self.query, self)


class MinTimeLoggingConnection(LoggingConnection):
    """A connection that logs queries based on execution time.

    This is just an example of how to sub-class `LoggingConnection` to
    provide some extra filtering for the logged queries. Both the
    `inizialize()` and `filter()` methods are overwritten to make sure
    that only queries executing for more than ``mintime`` ms are logged.

    Note that this connection uses the specialized cursor
    `MinTimeLoggingCursor`.
    """
    def initialize(self, logobj, mintime=0):
        LoggingConnection.initialize(self, logobj)
        self._mintime = mintime

    def filter(self, msg, curs):
        t = (time.time() - curs.timestamp) * 1000
        if t > self._mintime:
            return msg + os.linesep + "  (execution time: %d ms)" % t

    def cursor(self, *args, **kwargs):
        kwargs.setdefault('cursor_factory', MinTimeLoggingCursor)
        return LoggingConnection.cursor(self, *args, **kwargs)

class MinTimeLoggingCursor(LoggingCursor):
    """The cursor sub-class companion to `MinTimeLoggingConnection`."""

    def execute(self, query, vars=None):
        self.timestamp = time.time()
        return LoggingCursor.execute(self, query, vars)

    def callproc(self, procname, vars=None):
        self.timestamp = time.time()
        return LoggingCursor.execute(self, procname, vars)


# a dbtype and adapter for Python UUID type

class UUID_adapter(object):
    """Adapt Python's uuid.UUID__ type to PostgreSQL's uuid__.

    .. __: http://docs.python.org/library/uuid.html
    .. __: http://www.postgresql.org/docs/current/static/datatype-uuid.html
    """

    def __init__(self, uuid):
        self._uuid = uuid

    def prepare(self, conn):
        pass

    def getquoted(self):
        return "'"+str(self._uuid)+"'::uuid"

    __str__ = getquoted

def register_uuid(oids=None, conn_or_curs=None):
    """Create the UUID type and an uuid.UUID adapter.

    :param oids: oid for the PostgreSQL :sql:`uuid` type, or 2-items sequence
        with oids of the type and the array. If not specified, use PostgreSQL
        standard oids.
    :param conn_or_curs: where to register the typecaster. If not specified,
        register it globally.
    """

    import uuid

    if not oids:
        oid1 = 2950
        oid2 = 2951
    elif isinstance(oids, (list, tuple)):
        oid1, oid2 = oids
    else:
        oid1 = oids
        oid2 = 2951

    _ext.UUID = _ext.new_type((oid1, ), "UUID",
            lambda data, cursor: data and uuid.UUID(data) or None)
    _ext.UUIDARRAY = _ext.new_array_type((oid2,), "UUID[]", _ext.UUID)

    _ext.register_type(_ext.UUID, conn_or_curs)
    _ext.register_type(_ext.UUIDARRAY, conn_or_curs)
    _ext.register_adapter(uuid.UUID, UUID_adapter)

    return _ext.UUID


# a type, dbtype and adapter for PostgreSQL inet type

class Inet(object):
    """Wrap a string to allow for correct SQL-quoting of inet values.

    Note that this adapter does NOT check the passed value to make
    sure it really is an inet-compatible address but DOES call adapt()
    on it to make sure it is impossible to execute an SQL-injection
    by passing an evil value to the initializer.
    """
    def __init__(self, addr):
        self.addr = addr

    def __repr__(self):
        return "%s(%r)" % (self.__class__.__name__, self.addr)

    def prepare(self, conn):
        self._conn = conn

    def getquoted(self):
        obj = _A(self.addr)
        if hasattr(obj, 'prepare'):
            obj.prepare(self._conn)
        return obj.getquoted() + b("::inet")

    def __conform__(self, foo):
        if foo is _ext.ISQLQuote:
            return self

    def __str__(self):
        return str(self.addr)

def register_inet(oid=None, conn_or_curs=None):
    """Create the INET type and an Inet adapter.

    :param oid: oid for the PostgreSQL :sql:`inet` type, or 2-items sequence
        with oids of the type and the array. If not specified, use PostgreSQL
        standard oids.
    :param conn_or_curs: where to register the typecaster. If not specified,
        register it globally.
    """
    if not oid:
        oid1 = 869
        oid2 = 1041
    elif isinstance(oid, (list, tuple)):
        oid1, oid2 = oid
    else:
        oid1 = oid
        oid2 = 1041

    _ext.INET = _ext.new_type((oid1, ), "INET",
            lambda data, cursor: data and Inet(data) or None)
    _ext.INETARRAY = _ext.new_array_type((oid2, ), "INETARRAY", _ext.INET)

    _ext.register_type(_ext.INET, conn_or_curs)
    _ext.register_type(_ext.INETARRAY, conn_or_curs)

    return _ext.INET


def register_tstz_w_secs(oids=None, conn_or_curs=None):
    """The function used to register an alternate type caster for
    :sql:`TIMESTAMP WITH TIME ZONE` to deal with historical time zones with
    seconds in the UTC offset.

    These are now correctly handled by the default type caster, so currently
    the function doesn't do anything.
    """
    warnings.warn("deprecated", DeprecationWarning)


import select
from psycopg2.extensions import POLL_OK, POLL_READ, POLL_WRITE
from psycopg2 import OperationalError

def wait_select(conn):
    """Wait until a connection or cursor has data available.

    The function is an example of a wait callback to be registered with
    `~psycopg2.extensions.set_wait_callback()`. This function uses `!select()`
    to wait for data available.
    """
    while 1:
        state = conn.poll()
        if state == POLL_OK:
            break
        elif state == POLL_READ:
            select.select([conn.fileno()], [], [])
        elif state == POLL_WRITE:
            select.select([], [conn.fileno()], [])
        else:
            raise OperationalError("bad state from poll: %s" % state)


def _solve_conn_curs(conn_or_curs):
    """Return the connection and a DBAPI cursor from a connection or cursor."""
    if hasattr(conn_or_curs, 'execute'):
        conn = conn_or_curs.connection
        curs = conn.cursor(cursor_factory=_cursor)
    else:
        conn = conn_or_curs
        curs = conn.cursor(cursor_factory=_cursor)

    return conn, curs


class HstoreAdapter(object):
    """Adapt a Python dict to the hstore syntax."""
    def __init__(self, wrapped):
        self.wrapped = wrapped

    def prepare(self, conn):
        self.conn = conn

        # use an old-style getquoted implementation if required
        if conn.server_version < 90000:
            self.getquoted = self._getquoted_8

    def _getquoted_8(self):
        """Use the operators available in PG pre-9.0."""
        if not self.wrapped:
            return b("''::hstore")

        adapt = _ext.adapt
        rv = []
        for k, v in self.wrapped.iteritems():
            k = adapt(k)
            k.prepare(self.conn)
            k = k.getquoted()

            if v is not None:
                v = adapt(v)
                v.prepare(self.conn)
                v = v.getquoted()
            else:
                v = b('NULL')

            # XXX this b'ing is painfully inefficient!
            rv.append(b("(") + k + b(" => ") + v + b(")"))

        return b("(") + b('||').join(rv) + b(")")

    def _getquoted_9(self):
        """Use the hstore(text[], text[]) function."""
        if not self.wrapped:
            return b("''::hstore")

        k = _ext.adapt(self.wrapped.keys())
        k.prepare(self.conn)
        v = _ext.adapt(self.wrapped.values())
        v.prepare(self.conn)
        return b("hstore(") + k.getquoted() + b(", ") + v.getquoted() + b(")")

    getquoted = _getquoted_9

    _re_hstore = regex.compile(r"""
        # hstore key:
        # a string of normal or escaped chars
        "((?: [^"\\] | \\. )*)"
        \s*=>\s* # hstore value
        (?:
            NULL # the value can be null - not catched
            # or a quoted string like the key
            | "((?: [^"\\] | \\. )*)"
        )
        (?:\s*,\s*|$) # pairs separated by comma or end of string.
    """, regex.VERBOSE)

    @classmethod
    def parse(self, s, cur, _bsdec=regex.compile(r"\\(.)")):
        """Parse an hstore representation in a Python string.

        The hstore is represented as something like::

            "a"=>"1", "b"=>"2"

        with backslash-escaped strings.
        """
        if s is None:
            return None

        rv = {}
        start = 0
        for m in self._re_hstore.finditer(s):
            if m is None or m.start() != start:
                raise psycopg2.InterfaceError(
                    "error parsing hstore pair at char %d" % start)
            k = _bsdec.sub(r'\1', m.group(1))
            v = m.group(2)
            if v is not None:
                v = _bsdec.sub(r'\1', v)

            rv[k] = v
            start = m.end()

        if start < len(s):
            raise psycopg2.InterfaceError(
                "error parsing hstore: unparsed data after char %d" % start)

        return rv

    @classmethod
    def parse_unicode(self, s, cur):
        """Parse an hstore returning unicode keys and values."""
        if s is None:
            return None

        s = s.decode(_ext.encodings[cur.connection.encoding])
        return self.parse(s, cur)

    @classmethod
    def get_oids(self, conn_or_curs):
        """Return the lists of OID of the hstore and hstore[] types.
        """
        conn, curs = _solve_conn_curs(conn_or_curs)

        # Store the transaction status of the connection to revert it after use
        conn_status = conn.status

        # column typarray not available before PG 8.3
        typarray = conn.server_version >= 80300 and "typarray" or "NULL"

        rv0, rv1 = [], []

        # get the oid for the hstore
        curs.execute("""\
SELECT t.oid, %s
FROM pg_type t JOIN pg_namespace ns
    ON typnamespace = ns.oid
WHERE typname = 'hstore';
""" % typarray)
        for oids in curs:
            rv0.append(oids[0])
            rv1.append(oids[1])

        # revert the status of the connection as before the command
        if (conn_status != _ext.STATUS_IN_TRANSACTION
        and not conn.autocommit):
            conn.rollback()

        return tuple(rv0), tuple(rv1)

def register_hstore(conn_or_curs, globally=False, unicode=False,
        oid=None, array_oid=None):
    """Register adapter and typecaster for `!dict`\-\ |hstore| conversions.

    :param conn_or_curs: a connection or cursor: the typecaster will be
        registered only on this object unless *globally* is set to `!True`
    :param globally: register the adapter globally, not only on *conn_or_curs*
    :param unicode: if `!True`, keys and values returned from the database
        will be `!unicode` instead of `!str`. The option is not available on
        Python 3
    :param oid: the OID of the |hstore| type if known. If not, it will be
        queried on *conn_or_curs*.
    :param array_oid: the OID of the |hstore| array type if known. If not, it
        will be queried on *conn_or_curs*.

    The connection or cursor passed to the function will be used to query the
    database and look for the OID of the |hstore| type (which may be different
    across databases). If querying is not desirable (e.g. with
    :ref:`asynchronous connections <async-support>`) you may specify it in the
    *oid* parameter, which can be found using a query such as :sql:`SELECT
    'hstore'::regtype::oid`. Analogously you can obtain a value for *array_oid*
    using a query such as :sql:`SELECT 'hstore[]'::regtype::oid`.

    Note that, when passing a dictionary from Python to the database, both
    strings and unicode keys and values are supported. Dictionaries returned
    from the database have keys/values according to the *unicode* parameter.

    The |hstore| contrib module must be already installed in the database
    (executing the ``hstore.sql`` script in your ``contrib`` directory).
    Raise `~psycopg2.ProgrammingError` if the type is not found.
    """
    if oid is None:
        oid = HstoreAdapter.get_oids(conn_or_curs)
        if oid is None or not oid[0]:
            raise psycopg2.ProgrammingError(
                "hstore type not found in the database. "
                "please install it from your 'contrib/hstore.sql' file")
        else:
            array_oid = oid[1]
            oid = oid[0]

    if isinstance(oid, int):
        oid = (oid,)

    if array_oid is not None:
        if isinstance(array_oid, int):
            array_oid = (array_oid,)
        else:
            array_oid = tuple([x for x in array_oid if x])

    # create and register the typecaster
    if sys.version_info[0] < 3 and unicode:
        cast = HstoreAdapter.parse_unicode
    else:
        cast = HstoreAdapter.parse

    HSTORE = _ext.new_type(oid, "HSTORE", cast)
    _ext.register_type(HSTORE, not globally and conn_or_curs or None)
    _ext.register_adapter(dict, HstoreAdapter)

    if array_oid:
        HSTOREARRAY = _ext.new_array_type(array_oid, "HSTOREARRAY", HSTORE)
        _ext.register_type(HSTOREARRAY, not globally and conn_or_curs or None)


class CompositeCaster(object):
    """Helps conversion of a PostgreSQL composite type into a Python object.

    The class is usually created by the `register_composite()` function.
    You may want to create and register manually instances of the class if
    querying the database at registration time is not desirable (such as when
    using an :ref:`asynchronous connections <async-support>`).

    .. attribute:: name

        The name of the PostgreSQL type.

    .. attribute:: oid

        The oid of the PostgreSQL type.

    .. attribute:: array_oid

        The oid of the PostgreSQL array type, if available.

    .. attribute:: type

        The type of the Python objects returned. If :py:func:`collections.namedtuple()`
        is available, it is a named tuple with attributes equal to the type
        components. Otherwise it is just the `!tuple` object.

    .. attribute:: attnames

        List of component names of the type to be casted.

    .. attribute:: atttypes

        List of component type oids of the type to be casted.

    """
    def __init__(self, name, oid, attrs, array_oid=None):
        self.name = name
        self.oid = oid
        self.array_oid = array_oid

        self.attnames = [ a[0] for a in attrs ]
        self.atttypes = [ a[1] for a in attrs ]
        self._create_type(name, self.attnames)
        self.typecaster = _ext.new_type((oid,), name, self.parse)
        if array_oid:
            self.array_typecaster = _ext.new_array_type(
                (array_oid,), "%sARRAY" % name, self.typecaster)
        else:
            self.array_typecaster = None

    def parse(self, s, curs):
        if s is None:
            return None

        tokens = self.tokenize(s)
        if len(tokens) != len(self.atttypes):
            raise psycopg2.DataError(
                "expecting %d components for the type %s, %d found instead" %
                (len(self.atttypes), self.name, len(tokens)))

        attrs = [ curs.cast(oid, token)
            for oid, token in zip(self.atttypes, tokens) ]
        return self._ctor(*attrs)

    _re_tokenize = regex.compile(r"""
  \(? ([,)])                        # an empty token, representing NULL
| \(? " ((?: [^"] | "")*) " [,)]    # or a quoted string
| \(? ([^",)]+) [,)]                # or an unquoted string
    """, regex.VERBOSE)

    _re_undouble = regex.compile(r'(["\\])\1')

    @classmethod
    def tokenize(self, s):
        rv = []
        for m in self._re_tokenize.finditer(s):
            if m is None:
                raise psycopg2.InterfaceError("can't parse type: %r" % s)
            if m.group(1):
                rv.append(None)
            elif m.group(2):
                rv.append(self._re_undouble.sub(r"\1", m.group(2)))
            else:
                rv.append(m.group(3))

        return rv

    def _create_type(self, name, attnames):
        try:
            from collections import namedtuple
        except ImportError:
            self.type = tuple
            self._ctor = lambda *args: tuple(args)
        else:
            self.type = namedtuple(name, attnames)
            self._ctor = self.type

    @classmethod
    def _from_db(self, name, conn_or_curs):
        """Return a `CompositeCaster` instance for the type *name*.

        Raise `ProgrammingError` if the type is not found.
        """
        conn, curs = _solve_conn_curs(conn_or_curs)

        # Store the transaction status of the connection to revert it after use
        conn_status = conn.status

        # Use the correct schema
        if '.' in name:
            schema, tname = name.split('.', 1)
        else:
            tname = name
            schema = 'public'

        # column typarray not available before PG 8.3
        typarray = conn.server_version >= 80300 and "typarray" or "NULL"

        # get the type oid and attributes
        curs.execute("""\
SELECT t.oid, %s, attname, atttypid
FROM pg_type t
JOIN pg_namespace ns ON typnamespace = ns.oid
JOIN pg_attribute a ON attrelid = typrelid
WHERE typname = %%s AND nspname = %%s
    AND attnum > 0 AND NOT attisdropped
ORDER BY attnum;
""" % typarray, (tname, schema))

        recs = curs.fetchall()

        # revert the status of the connection as before the command
        if (conn_status != _ext.STATUS_IN_TRANSACTION
        and not conn.autocommit):
            conn.rollback()

        if not recs:
            raise psycopg2.ProgrammingError(
                "PostgreSQL type '%s' not found" % name)

        type_oid = recs[0][0]
        array_oid = recs[0][1]
        type_attrs = [ (r[2], r[3]) for r in recs ]

        return CompositeCaster(tname, type_oid, type_attrs,
            array_oid=array_oid)

def register_composite(name, conn_or_curs, globally=False):
    """Register a typecaster to convert a composite type into a tuple.

    :param name: the name of a PostgreSQL composite type, e.g. created using
        the |CREATE TYPE|_ command
    :param conn_or_curs: a connection or cursor used to find the type oid and
        components; the typecaster is registered in a scope limited to this
        object, unless *globally* is set to `!True`
    :param globally: if `!False` (default) register the typecaster only on
        *conn_or_curs*, otherwise register it globally
    :return: the registered `CompositeCaster` instance responsible for the
        conversion

    .. versionchanged:: 2.4.3
        added support for array of composite types

    """
    caster = CompositeCaster._from_db(name, conn_or_curs)
    _ext.register_type(caster.typecaster, not globally and conn_or_curs or None)

    if caster.array_typecaster is not None:
        _ext.register_type(caster.array_typecaster, not globally and conn_or_curs or None)

    return caster


__all__ = filter(lambda k: not k.startswith('_'), locals().keys())