freeipa/ipapython/dn.py

# Authors:
#   John Dennis <jdennis@redhat.com>
#
# Copyright (C) 2011  Red Hat
# see file 'COPYING' for use and warranty information
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# This program 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 General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program.  If not, see <http://www.gnu.org/licenses/>.

'''

Goal
----

To allow a Python programmer the ability to operate on DN's
(Distinguished Names) in a simple intuitive manner supporting all the
Pythonic mechanisms for manipulating objects such that the simple
majority case remains simple with simple code, yet the corner cases
are fully supported. With the result both simple and complex cases are
100% correct.

This is achieved with a fair of amount of syntax sugar which is best
described as "Do What I Mean" (i.e. DWIM). The class implementations
take simple expressions and internally convert them to their more
complex full definitions hiding much of the complexity from the
programmer.

Anatomy of a DN
---------------

Some definitions:

AVA
    An AVA is an Attribute Value Assertion. In more simple terms it's
    an attribute value pair typically expressed as attr=value
    (e.g. cn=Bob). Both the attr and value in an AVA when expressed in
    a string representation are subject to encoding rules.

RDN
    A RDN is a Relative Distinguished Name. A RDN is a non-empty set of
    AVA's. In the common case a RDN is single valued consisting of 1
    AVA (e.g. cn=Bob). But a RDN may be multi-valued consisting of
    more than one AVA. Because the RDN is a set of AVA's the AVA's are
    unordered when they appear in a multi-valued RDN. In the string
    representation of a RDN AVA's are separated by the plus sign (+).

DN
    A DN is a ordered sequence of 1 or more RDN's. In the string
    representation of a DN each RDN is separated by a comma (,)

Thus a DN is:

Sequence of set of <encoded attr, encoded value> pairs

The following are valid DN's

# 1 RDN with 1 AVA (e.g. cn=Bob)
RDN(AVA)

# 2 RDN's each with 1 AVA (e.g. cn=Bob,dc=redhat.com)
RDN(AVA),RDN(AVA)

# 2 RDN's the first RDN is multi-valued with 2 AVA's
# the second RDN is singled valued with 1 AVA
# (e.g. cn=Bob+ou=people,dc=redhat.com
RDN({AVA,AVA}),RDN(AVA)

Common programming mistakes
---------------------------

DN's present a pernicious problem for programmers. They appear to have
a very simple string format in the majority case, a sequence of
attr=value pairs separated by commas. For example:

dn='cn=Bob,ou=people,dc=redhat,dc=com'

As such there is a tendency to believe you can form DN's by simple
string manipulations such as:

dn='%s=%s' % ('cn','Bob') + ',ou=people,dc=redhat,dc=com'

Or to extract a attr & value by searching the string, for example:

attr=dn[0 : dn.find('=')]
value=dn[dn.find('=')+1 : dn.find(',')]

Or compare a value returned by an LDAP query to a known value:

if value == 'Bob'

All of these simple coding assumptions are WRONG and will FAIL when a
DN is not one of the simple DN's (simple DN's are probably the 95% of
all DN's). This is what makes DN handling pernicious. What works in
95% of the cases and is simple, fails for the 5% of DN's which are not
simple.

Examples of where the simple assumptions fail are:

* A RDN may be multi-valued

* A multi-valued RDN has no ordering on it's components

* Attr's and values must be UTF-8 encoded

* String representations of AVA's, RDN's and DN's must be completely UTF-8

* An attr or value may have reserved characters which must be escaped.

* Whitespace needs special handling

To complicate matters a bit more the RFC for the string representation
of DN's (RFC 4514) permits a variety of different syntax's each of
which can evaluate to exactly the same DN but have different string
representations. For example, the attr "r,w" which contains a reserved
character (the comma) can be encoded as a string in these different
ways:

'r\,w'          # backslash escape
'r\2cw'         # hexadecimal ascii escape
'#722C77'       # binary encoded

It should be clear a DN string may NOT be a simple string, rather a DN
string is ENCODED. For simple strings the encoding of the DN is
identical to the simple string value (this common case leads to
erroneous assumptions and bugs because it does not account for
encodings).

The openldap library we use at the client level uses the backslash
escape form. The LDAP server we use uses the hexadecimal ascii escape
form. Thus 'r,w' appears as 'r\,w' when sent from the client to the
LDAP server as part of a DN. But when it's returned as a DN from the
server in an LDAP search it's returned as 'r\2cw'. Any attempt to
compare 'r\,w' to 'r\2cw' for equality will fail despite the fact they
are indeed equal once decoded. Such a test fails because you're
comparing two different encodings of the same value. In MIME you
wouldn't expect the base64 encoding of a string to be equal to the
same string encoded as quoted-printable would you?

When you are comparing attrs or values which are part of a DN and
other string you MUST:

* Know if either of the strings have been encoded and make sure you're
  comparing only decoded components component-wise.

* Extract the component from the DN and decode it. You CANNOT decode
  the entire DN as a string and operate on it. Why? Consider a value
  with a comma embedded in it. For example:

  cn=r\2cw,cn=privilege

  Is a DN with 2 RDN components: cn=r,w followed by "cn=privilege"

  But if you decode the entire DN string as a whole you would get:

  cn=r,w,cn=privilege

  Which is a malformed DN with 3 RDN's, the 2nd RDN is invalid.

* Determine if a RDN is multi-valued, if so you must account
  for the fact each AVA component in the multi-valued RDN can appear
  in any order and still be equivalent. For example the following two
  RDN's are equal:

  cn=Bob+ou=people
  ou=people+cn=Bob

  In addition each AVA (cn=Bob & ou=people) needs to be
  INDEPENDENTLY decoded prior to comparing the unordered set of AVA's
  in the multi-valued RDN.

If you are trying to form a new DN or RDN from a raw string you cannot
simply do string concatenation or string formatting unless you ESCAPE
the components independently prior to concatenation, for example:

  base = 'dc=redhat,dc=com'
  value = 'r,w'
  dn = 'cn=%s,%s' % (value, base)

Will result in the malformed DN 'cn=r,w,dc=redhat,dc=com'

Syntax Sugar
------------

The majority of DN's have a simple string form:

attr=value,attr=value

We want the programmer to be able to create DN's, compare them, and
operate on their components as simply and concisely as possible so
the classes are implemented to provide a lot of syntax sugar.

The classes automatically handle UTF-8 <-> Unicode conversions. Every
attr and value which is returned from a class will be Unicode. Every
attr and value assigned into an object will be promoted to
Unicode. All string representations in RFC 4514 format will be UTF-8
and properly escaped. Thus at the "user" or "API" level every string
is Unicode with the single exception that the str() method returns RFC
compliant escaped UTF-8.

RDN's are assumed to be single-valued. If you need a multi-valued RDN
(an exception) you must explicitly create a multi-valued RDN.

Thus DN's are assumed to be a sequence of attr, value pairs, which is
equivalent to a sequence of RDN's. The attr and value in the pair MUST
be strings.

The DN and RDN constructors take a sequence, the constructor parses
the sequence to find items it knows about.

The DN constructor will accept in it's sequence:
  * tuple of 2 strings, converting it to an RDN
  * list of 2 strings, converting it to an RDN
  * a RDN object
  * a DN syntax string (e.g. 'cn=Bob,dc=redhat.com')

Note DN syntax strings should be avoided if possible when passing to a
constructor because they run afoul of the problems outlined above
which the DN, RDN & AVA classes are meant to overcome. But sometimes a
DN syntax string is all you have to work with. DN strings which come
from a LDAP library or server will be properly formed and it's safe to
use those. However DN strings provided via user input should be
treated suspiciously as they may be improperly formed. You can test
for this by passing the string to the DN constructor and see if it
throws an exception.

The sequence passed to the DN constructor takes each item in order,
produces one or more RDN's from it and appends those RDN in order to
its internal RDN sequence.

For example:

   DN(('cn', 'Bob'), ('dc', 'redhat.com'))

This is equivalent to the DN string:

    cn=Bob,dc=redhat.com

And is exactly equal to:

    DN(RDN(AVA('cn','Bob')),RDN(AVA('dc','redhat.com')))

The following are alternative syntax's which are all exactly
equivalent to the above example.

   DN(['cn', 'Bob'], ['dc', 'redhat.com'])
   DN(RDN('cn', 'Bob'), RDN('dc', 'redhat.com'))

You can provide a properly escaped string representation.

   DN('cn=Bob,dc=redhat.com')

You can mix and match any of the forms in the constructor parameter
list.

   DN(('cn', 'Bob'), 'dc=redhat.com')
   DN(('cn', 'Bob'), RDN('dc', 'redhat.com'))

AVA's have an attr and value property, thus if you have an AVA

# Get the attr and value
ava.attr  -> u'cn'
ava.value -> u'Bob'

# Set the attr and value
ava.attr  = 'cn'
ava.value = 'Bob'

Since RDN's are assumed to be single valued, exactly the same
behavior applies to an RDN. If the RDN is multi-valued then the attr
property returns the attr of the first AVA, likewise for the value.

# Get the attr and value
rdn.attr  -> u'cn'
rdn.value -> u'Bob'

# Set the attr and value
rdn.attr  = 'cn'
rdn.value = 'Bob'

Also RDN's can be indexed by name or position (see the RDN class doc
for details).

rdn['cn'] -> u'Bob'
rdn[0] -> AVA('cn', 'Bob')

A DN is a sequence of RDN's, as such any of Python's container
operators can be applied to a DN in a intuitive way.

# How many RDN's in a DN?
len(dn)

# WARNING, this a count of RDN's not how characters there are in the
# string representation the dn, instead that would be:
len(str(dn))

# Iterate over each RDN in a DN
for rdn in dn:

# Get the first RDN in a DN
dn[0] -> RDN('cn', 'Bob')

# Get the value of the first RDN in a DN
dn[0].value -> u'Bob'

# Get the value of the first RDN by indexing by attr name
dn['cn'] -> u'Bob'

# WARNING, when a string is used as an index key the FIRST RDN's value
# in the sequence whose attr matches the key is returned. Thus if you
# have a DN like this "cn=foo,cn=bar" then dn['cn'] will always return
# 'foo' even though there is another attr with the name 'cn'. This is
# almost always what the programmer wants. See the class doc for how
# you can override this default behavior and get a list of every value
# whose attr matches the key.

# Set the first RDN in the DN (all are equivalent)
dn[0] = ('cn', 'Bob')
dn[0] = ['cn', 'Bob']
dn[0] = RDN('cn', 'Bob')

dn[0].attr = 'cn'
dn[0].value = 'Bob'

# Get the first two RDN's using slices
dn[0:2]

# Get the last two RDN's using slices
dn[-2:]

# Get a list of all RDN's using slices
dn[:]

# Set the 2nd and 3rd RDN using slices (all are equivalent)
dn[1:3] = ('cn', 'Bob), ('dc', 'redhat.com')
dn[1:3] = RDN('cn', 'Bob), RDN('dc', 'redhat.com')

String representations and escapes:

# To get an RFC compliant string representation of a DN, RDN or AVA
# simply call str() on it or evaluate it in a string context.
str(dn) -> 'cn=Bob,dc=redhat.com'

# When working with attr's and values you do not have to worry about
# escapes, simply use the raw unescaped string in a natural fashion.

rdn = RDN('cn', 'r,w')

# Thus:
rdn.value == 'r,w' -> True

# But:
str(rdn) == 'cn=r,w' -> False
# Because:
str(rdn) -> 'cn=r\2cw' or 'cn='r\,w' # depending on the underlying LDAP library

Equality and Comparing:

# All DN's, RDN's and AVA's support equality testing in an intuitive
# manner.
dn1 = DN(('cn', 'Bob'))
dn2 = DN(RDN('cn', 'Bob'))
dn1 == dn2 -> True
dn1[0] == dn2[0] -> True
dn1[0].value = 'Bobby'
dn1 == dn2 -> False

DN objects implement startswith(), endswith() and the "in" membership
operator. You may pass a DN or RDN object to these. Examples:

if dn.endswith(base_dn):
if dn.startswith(rdn1):
if container_dn in dn:

# See the class doc for how DN's, RDN's and AVA's compare
# (e.g. cmp()). The general rule is for objects supporting multiple
# values first their lengths are compared, then if the lengths match
# the respective components of each are pair-wise compared until one
# is discovered to be  non-equal. The comparison is case insensitive.

Concatenation, In-Place Addition, Insertion:

# DN's and RDN's can be concatenated.
# Return a new DN by appending the RDN's of dn2 to dn1
dn3 = dn1 + dn2

# Append a RDN to DN's RDN sequence (all are equivalent)
dn += ('cn', 'Bob')
dn += RDN('cn', 'Bob')

# Append a DN to an existing DN
dn1 += dn2

# Prepend a RDN to an existing DN
dn1.insert(0, RDN('cn', 'Bob'))

Finally see the unittest for a more complete set of ways you can
manipulate these objects.

Immutability
------------

All the class types are immutable.
As with other immutable types (such as str and int), you must not rely on
the object identity operator ("is") for comparisons.

It is possible to "copy" an object by passing an object of the same type
to the constructor. The result may share underlying structure.

'''
from __future__ import print_function

import sys
import functools

from ldap.dn import str2dn, dn2str
from ldap import DECODING_ERROR
import six

if six.PY3:
    unicode = str

__all__ = 'AVA', 'RDN', 'DN'

def _adjust_indices(start, end, length):
    'helper to fixup start/end slice values'

    if end > length:
        end = length
    elif end < 0:
        end += length
        if end < 0:
            end = 0

    if start < 0:
        start += length
        if start < 0:
            start = 0

    return start, end


def _normalize_ava_input(val):
    if six.PY3 and isinstance(val, bytes):
        raise TypeError('expected str, got bytes: %s' % val)
    elif not isinstance(val, six.string_types):
        val = val_encode(six.text_type(val))
    elif six.PY2 and isinstance(val, unicode):
        val = val.encode('utf-8')
    return val


def str2rdn(value):
    try:
        rdns = str2dn(value.encode('utf-8'))
    except DECODING_ERROR:
        raise ValueError("malformed AVA string = \"%s\"" % value)
    if len(rdns) != 1:
        raise ValueError("multiple RDN's specified by \"%s\"" % (value))
    return rdns[0]


def get_ava(*args):
    """
    Get AVA from args in open ldap format(raw). Optimized for construction
    from openldap format.

    Allowed formats of argument list:
    1) three args - open ldap format (attr and value have to be utf-8 encoded):
        a) ['attr', 'value', 0]
    2) two args:
        a) ['attr', 'value']
    3) one arg:
        a) [('attr', 'value')]
        b) [['attr', 'value']]
        c) [AVA(..)]
        d) ['attr=value']
    """
    ava = None
    l = len(args)
    if l == 3:  # raw values - constructed FROM RDN
        ava = args
    elif l == 2:  # user defined values
        ava = [_normalize_ava_input(args[0]), _normalize_ava_input(args[1]), 0]
    elif l == 1:  # slow mode, tuple, string,
        arg = args[0]
        if isinstance(arg, AVA):
            ava = arg.to_openldap()
        elif isinstance(arg, (tuple, list)):
            if len(arg) != 2:
                raise ValueError("tuple or list must be 2-valued, not \"%s\"" % (arg))
            ava = [_normalize_ava_input(arg[0]), _normalize_ava_input(arg[1]), 0]
        elif isinstance(arg, six.string_types):
            rdn = str2rdn(arg)
            if len(rdn) > 1:
                raise TypeError("multiple AVA's specified by \"%s\"" % (arg))
            ava = list(rdn[0])
        else:
            raise TypeError("with 1 argument, argument must be str, unicode, tuple or list, got %s instead" %
                            arg.__class__.__name__)
    else:
        raise TypeError("invalid number of arguments. 1-3 allowed")
    return ava


def sort_avas(rdn):
    if len(rdn) <= 1:
        return
    rdn.sort(key=ava_key)


def ava_key(ava):
    return ava[0].lower(), ava[1].lower()


def cmp_rdns(a, b):
    key_a = rdn_key(a)
    key_b = rdn_key(b)
    if key_a == key_b:
        return 0
    elif key_a < key_b:
        return -1
    else:
        return 1


def rdn_key(rdn):
    return (len(rdn),) + tuple(ava_key(k) for k in rdn)


if six.PY2:
    # Python 2: Input/output is unicode; we store UTF-8 bytes
    def val_encode(s):
        return s.encode('utf-8')

    def val_decode(s):
        return s.decode('utf-8')
else:
    # Python 3: Everything is unicode (str)
    def val_encode(s):
        if isinstance(s, bytes):
            raise TypeError('expected str, got bytes: %s' % s)
        return s

    def val_decode(s):
        return s


@functools.total_ordering
class AVA(object):
    '''
    AVA(arg0, ...)

    An AVA is an LDAP Attribute Value Assertion. It is convenient to think of
    AVA's as a <attr,value> pair. AVA's are members of RDN's (Relative
    Distinguished Name).

    The AVA constructor is passed a sequence of args and a set of
    keyword parameters used for configuration.

    The arg sequence may be:

    1) With 2 arguments, the first argument will be the attr, the 2nd
    the value. Each argument must be scalar convertable to unicode.

    2) With a sigle list or tuple argument containing exactly 2 items.
    Each item must be scalar convertable to unicode.

    3) With a single string (or unicode) argument, in this case the string will
    be interpretted using the DN syntax described in RFC 4514 to yield a AVA
    <attr,value> pair. The parsing recognizes the DN syntax escaping rules.

    For example:

    ava = AVA('cn', 'Bob')	# case 1: two strings
    ava = AVA(('cn', 'Bob'))    # case 2: 2-valued tuple
    ava = AVA(['cn', 'Bob'])    # case 2: 2-valued list
    ava = AVA('cn=Bob')         # case 3: DN syntax

    AVA object have two properties for accessing their data:

    attr:  the attribute name, cn in our exmaple
    value: the attribute's value, Bob in our example

    When attr and value are returned they will always be unicode. When
    attr or value are set they will be promoted to unicode.

    AVA objects support indexing by name, e.g.

    ava['cn']

    returns the value (Bob in our example). If the index does key does not match
    the attr then a KeyError will be raised.

    AVA objects support equality testing and comparsion (e.g. cmp()). When they
    are compared the attr is compared first, if the 2 attr's are equal then the
    values are compared. The comparison is case insensitive (because attr's map
    to numeric OID's and their values derive from from the 'name' atribute type
    (OID 2.5.4.41) whose EQUALITY MATCH RULE is caseIgnoreMatch.

    The str method of an AVA returns the string representation in RFC 4514 DN
    syntax with proper escaping.
    '''
    def __init__(self, *args):
        self._ava = get_ava(*args)

    def _get_attr(self):
        return val_decode(self._ava[0])

    def _set_attr(self, new_attr):
        try:
            self._ava[0] = _normalize_ava_input(new_attr)
        except Exception as e:
            raise ValueError('unable to convert attr "%s": %s' % (new_attr, e))

    attr = property(_get_attr)

    def _get_value(self):
        return val_decode(self._ava[1])

    def _set_value(self, new_value):
        try:
            self._ava[1] = _normalize_ava_input(new_value)
        except Exception as e:
            raise ValueError('unable to convert value "%s": %s' % (new_value, e))

    value = property(_get_value)

    def to_openldap(self):
        return list(self._ava)

    def __str__(self):
        return dn2str([[self.to_openldap()]])

    def __repr__(self):
        return "%s.%s('%s')" % (self.__module__, self.__class__.__name__, self.__str__())

    def __getitem__(self, key):

        if key == 0:
            return self.attr
        elif key == 1:
            return self.value
        elif key == self.attr:
            return self.value
        else:
            raise KeyError("\"%s\" not found in %s" % (key, self.__str__()))

    def __hash__(self):
        # Hash is computed from AVA's string representation.
        #
        # Because attrs & values are comparison case-insensitive the
        # hash value between two objects which compare as equal but
        # differ in case must yield the same hash value.

        return hash(str(self).lower())

    def __eq__(self, other):
        '''
        The attr comparison is case insensitive because attr is
        really an LDAP attribute type which means it's specified with
        an OID (dotted number) and not a string. Since OID's are
        numeric the human readable name which maps to the OID is not
        significant in case.

        The value comparison is also case insensitive because the all
        attribute types used in a DN are derived from the 'name'
        atribute type (OID 2.5.4.41) whose EQUALITY MATCH RULE is
        caseIgnoreMatch.
        '''
        # Try coercing string to AVA, if successful compare to coerced object
        if isinstance(other, six.string_types):
            try:
                other_ava = AVA(other)
                return self.__eq__(other_ava)
            except Exception:
                return False

        # If it's not an AVA it can't be equal
        if not isinstance(other, AVA):
            return False

        # Perform comparison between objects of same type
        return ava_key(self._ava) == ava_key(other._ava)

    def __ne__(self, other):
        return not self.__eq__(other)

    def __lt__(self, other):
        'comparison is case insensitive, see __eq__ doc for explanation'

        if not isinstance(other, AVA):
            raise TypeError("expected AVA but got %s" % (other.__class__.__name__))

        return ava_key(self._ava) < ava_key(other._ava)


@functools.total_ordering
class RDN(object):
    '''
    RDN(arg0, ...)

    An RDN is a LDAP Relative Distinguished Name. RDN's are members of DN's
    (Distinguished Name). An RDN contains 1 or more AVA's. If the RDN contains
    more than one AVA it is said to be a multi-valued RDN. When an RDN is
    multi-valued the AVA's are unorderd comprising a set. However this
    implementation orders the AVA's according to the AVA comparison function to
    make equality and comparison testing easier. Think of this a canonical
    normalization (however LDAP does not impose any ordering on multiple AVA's
    within an RDN). Single valued RDN's are the norm and thus the RDN
    constructor has simple syntax for them.

    The RDN constructor is passed a sequence of args and a set of
    keyword parameters used for configuration.

    The constructor iterates though the sequence and adds AVA's to the RDN.

    The arg sequence may be:

    * A 2-valued tuple or list denotes the <attr,value> pair of an AVA. The
    first member is the attr and the second member is the value, both members
    must be strings (or unicode). The tuple or list is passed to the AVA
    constructor and the resulting AVA is added to the RDN. Multiple tuples or
    lists may appear in the argument list, each adds one additional AVA to the
    RDN.

    * A single string (or unicode) argument, in this case the string will
    be interpretted using the DN syntax described in RFC 4514 to yield one or
    more AVA <attr,value> pairs. The parsing recognizes the DN syntax escaping
    rules.

    * A AVA object, the AVA will be copied into the new RDN respecting
      the constructors keyword configuration parameters.

    * A RDN object, the AVA's in the RDN are copied into the new RDN
      respecting the constructors keyword configuration parameters.

    Single AVA Examples:

    RDN(('cn', 'Bob'))                  # tuple yields 1 AVA
    RDN('cn=Bob')                       # DN syntax with 1 AVA
    RDN(AVA('cn', 'Bob'))               # AVA object adds 1 AVA

    Multiple AVA Examples:

    RDN(('cn', 'Bob'),('ou', 'people')) # 2 tuples yields 2 AVA's
    RDN('cn=Bob+ou=people')             # DN syntax with 2 AVA's
    RDN(AVA('cn', 'Bob'),AVA('ou', 'people')) # 2 AVA objects adds 2 AVA's
    RDN(('cn', 'Bob'), 'ou=people')     # 2 args, 1st tuple forms 1 AVA,
                                        # 2nd DN syntax string adds 1 AVA,
                                        # 2 AVA's in total

    Note: The RHS of a slice assignment is interpreted exactly in the
    same manner as the constructor argument list (see above examples).

    RDN objects support iteration over their AVA members. You can iterate all
    AVA members via any Python iteration syntax. RDN objects support full Python
    indexing using bracket [] notation. Examples:

    len(rdn)            # return the number of AVA's
    rdn[0]              # indexing the first AVA
    rdn['cn']           # index by AVA attr, returns AVA value
    for ava in rdn:     # iterate over each AVA
    rdn[:]              # a slice, in this case a copy of each AVA

    WARNING: When indexing by attr (e.g. rdn['cn']) there is a possibility more
    than one AVA has the same attr name as the index key. The default behavior
    is to return the value of the first AVA whose attr matches the index
    key.

    RDN objects support the AVA attr and value properties as another programmer
    convenience because the vast majority of RDN's are single valued. The attr
    and value properties return the attr and value properties of the first AVA
    in the RDN, for example:

    rdn = RDN(('cn', 'Bob')) # rdn has 1 AVA whose attr == 'cn' and value == 'Bob'
    len(rdn) -> 1
    rdn.attr -> u'cn'      # exactly equivalent to rdn[0].attr
    rdn.value -> u'Bob'    # exactly equivalent to rdn[0].value

    When attr and value are returned they will always be unicode. When
    attr or value are set they will be promoted to unicode.

    If an RDN is multi-valued the attr and value properties still return only
    the first AVA's properties, programmer beware! Recall the AVA's in the RDN
    are sorted according the to AVA collating semantics.

    RDN objects support equality testing and comparison. See AVA for the
    definition of the comparison method.

    RDN objects support concatenation and addition with other RDN's or AVA's

    rdn1 + rdn2 # yields a new RDN object with the contents of each RDN.
    rdn1 + ava1 # yields a new RDN object with the contents of rdn1 and ava1

    RDN objects can add AVA's objects via in-place addition.

    rdn1 += rdn2 # rdn1 now contains the sum of rdn1 and rdn2
    rdn1 += ava1 # rdn1 has ava1 added to it.

    The str method of an RDN returns the string representation in RFC 4514 DN
    syntax with proper escaping.
    '''

    AVA_type = AVA

    def __init__(self, *args, **kwds):
        self._avas = self._avas_from_sequence(args, kwds.get('raw', False))

    def _avas_from_sequence(self, args, raw=False):
        avas = []
        sort = 0
        ava_count = len(args)

        if raw:  # fast raw mode
            avas = args
        elif ava_count == 1 and isinstance(args[0], six.string_types):
            avas = str2rdn(args[0])
            sort = 1
        elif ava_count == 1 and isinstance(args[0], RDN):
            avas = args[0].to_openldap()
        elif ava_count > 0:
            sort = 1
            for arg in args:
                avas.append(get_ava(arg))
        if sort:
            sort_avas(avas)
        return avas

    def to_openldap(self):
        return [list(a) for a in self._avas]

    def __str__(self):
        return dn2str([self.to_openldap()])

    def __repr__(self):
        return "%s.%s('%s')" % (self.__module__, self.__class__.__name__, self.__str__())

    def _get_ava(self, ava):
        return self.AVA_type(*ava)

    def _next(self):
        for ava in self._avas:
            yield self._get_ava(ava)

    def __iter__(self):
        return self._next()

    def __len__(self):
        return len(self._avas)

    def __getitem__(self, key):
        if isinstance(key, six.integer_types):
            return self._get_ava(self._avas[key])
        if isinstance(key, slice):
            return [self._get_ava(ava) for ava in self._avas[key]]
        elif isinstance(key, six.string_types):
            for ava in self._avas:
                if key == val_decode(ava[0]):
                    return val_decode(ava[1])
            raise KeyError("\"%s\" not found in %s" % (key, self.__str__()))
        else:
            raise TypeError("unsupported type for RDN indexing, must be int, basestring or slice; not %s" % \
                                (key.__class__.__name__))

    def _get_attr(self):
        if len(self._avas) == 0:
            raise IndexError("No AVA's in this RDN")
        return val_decode(self._avas[0][0])

    def _set_attr(self, new_attr):
        if len(self._avas) == 0:
            raise IndexError("No AVA's in this RDN")

        self._avas[0][0] = val_encode(six.text_type(new_attr))

    attr  = property(_get_attr)

    def _get_value(self):
        if len(self._avas) == 0:
            raise IndexError("No AVA's in this RDN")
        return val_decode(self._avas[0][1])

    def _set_value(self, new_value):
        if len(self._avas) == 0:
            raise IndexError("No AVA's in this RDN")
        self._avas[0][1] = val_encode(six.text_type(new_value))

    value = property(_get_value)

    def __hash__(self):
        # Hash is computed from RDN's string representation.
        #
        # Because attrs & values are comparison case-insensitive the
        # hash value between two objects which compare as equal but
        # differ in case must yield the same hash value.

        return hash(str(self).lower())

    def __eq__(self, other):
        # Try coercing string to RDN, if successful compare to coerced object
        if isinstance(other, six.string_types):
            try:
                other_rdn = RDN(other)
                return self.__eq__(other_rdn)
            except Exception:
                return False

        # If it's not an RDN it can't be equal
        if not isinstance(other, RDN):
            return False

        # Perform comparison between objects of same type
        return rdn_key(self._avas) == rdn_key(other._avas)

    def __ne__(self, other):
        return not self.__eq__(other)

    def __lt__(self, other):
        if not isinstance(other, RDN):
            raise TypeError("expected RDN but got %s" % (other.__class__.__name__))

        return rdn_key(self._avas) < rdn_key(other._avas)

    def __add__(self, other):
        result = self.__class__(self)
        if isinstance(other, RDN):
            for ava in other._avas:
                result._avas.append((ava[0], ava[1], ava[2]))
        elif isinstance(other, AVA):
            result._avas.append(other.to_openldap())
        elif isinstance(other, six.string_types):
            rdn = self.__class__(other)
            for ava in rdn._avas:
                result._avas.append((ava[0], ava[1], ava[2]))
        else:
            raise TypeError("expected RDN, AVA or basestring but got %s" % (other.__class__.__name__))

        sort_avas(result._avas)
        return result


@functools.total_ordering
class DN(object):
    '''
    DN(arg0, ...)

    A DN is a LDAP Distinguished Name. A DN is an ordered sequence of RDN's.

    The DN constructor is passed a sequence of args and a set of
    keyword parameters used for configuration. normalize means the
    attr and value will be converted to lower case.

    The constructor iterates through the sequence and adds the RDN's
    it finds in order to the DN object. Each item in the sequence may
    be:

    * A 2-valued tuple or list. The first member is the attr and the
      second member is the value of an RDN, both members must be
      strings (or unicode). The tuple or list is passed to the RDN
      constructor and the resulting RDN is appended to the
      DN. Multiple tuples or lists may appear in the argument list,
      each adds one additional RDN to the DN.

    * A single string (or unicode) argument, in this case the string
      will be interpretted using the DN syntax described in RFC 4514
      to yield one or more RDN's which will be appended in order to
      the DN. The parsing recognizes the DN syntax escaping rules.

    * A RDN object, the RDN will copied respecting the constructors
      keyword configuration parameters and appended in order.

    * A DN object, the RDN's in the DN are copied respecting the
      constructors keyword configuration parameters and appended in
      order.

    Single DN Examples:

    DN(('cn', 'Bob'))                   # tuple yields 1 RDN
    DN(['cn', 'Bob'])                   # list yields 1 RDN
    DN('cn=Bob')                        # DN syntax with 1 RDN
    DN(RDN('cn', 'Bob'))                # RDN object adds 1 RDN

    Multiple RDN Examples:

    DN(('cn', 'Bob'),('ou', 'people'))  # 2 tuples yields 2 RDN's
                                        # 2 RDN's total
    DN('cn=Bob,ou=people')              # DN syntax with 2 RDN's
                                        # 2 RDN's total
    DN(RDN('cn', 'Bob'),RDN('ou', 'people')) # 2 RDN objects
                                        # 2 RDN's total
    DN(('cn', 'Bob'), "ou=people')      # 1st tuple adds 1 RDN
                                        # 2nd DN syntax string adds 1 RDN
                                        # 2 RDN's total
    base_dn = DN('dc=redhat,dc=com')
    container_dn = DN('cn=sudorules,cn=sudo')
    DN(('cn', 'Bob'), container_dn, base_dn)
                                        # 1st arg adds 1 RDN, cn=Bob
                                        # 2nd arg adds 2 RDN's, cn=sudorules,cn=sudo
                                        # 3rd arg adds 2 RDN's, dc=redhat,dc=com
                                        # 5 RDN's total


    Note: The RHS of a slice assignment is interpreted exactly in the
    same manner as the constructor argument list (see above examples).

    DN objects support iteration over their RDN members. You can iterate all
    RDN members via any Python iteration syntax. DN objects support full Python
    indexing using bracket [] notation. Examples:

    len(rdn)            # return the number of RDN's
    rdn[0]              # indexing the first RDN
    rdn['cn']           # index by RDN attr, returns RDN value
    for ava in rdn:     # iterate over each RDN
    rdn[:]              # a slice, in this case a copy of each RDN

    WARNING: When indexing by attr (e.g. dn['cn']) there is a
    possibility more than one RDN has the same attr name as the index
    key. The default behavior is to return the value of the first RDN
    whose attr matches the index key. If it's important the attr
    belong to a specific RDN (e.g. the first) then this is the
    suggested construct:

        try:
            cn = dn[0]['cn']
        except (IndexError, KeyError):
            raise ValueError("dn '%s' missing expected cn as first attribute" % dn)

    The IndexError catches a DN which does not have the expected
    number of RDN's and the KeyError catches the case where the
    indexed RDN does not have the expected attr.

    DN object support slices.

    # Get the first two RDN's using slices
    dn[0:2]

    # Get the last two RDN's using slices
    dn[-2:]

    # Get a list of all RDN's using slices
    dn[:]

    # Set the 2nd and 3rd RDN using slices (all are equivalent)
    dn[1:3] = ('cn', 'Bob'), ('dc', 'redhat.com')
    dn[1:3] = [['cn', 'Bob'], ['dc', 'redhat.com']]
    dn[1:3] = RDN('cn', 'Bob'), RDN('dc', 'redhat.com')

    DN objects support the insert operation.

    dn.insert(i,x) is exactly equivalent to dn[i:i] = [x], thus the following
    are all equivalent:

    dn.insert(i, ('cn','Bob'))
    dn.insert(i, ['cn','Bob'])
    dn.insert(i, RDN(('cn','Bob')))
    dn[i:i] = [('cn','Bob')]

    DN objects support equality testing and comparison. See RDN for the
    definition of the comparison method.

    DN objects implement startswith(), endswith() and the "in" membership
    operator. You may pass a DN or RDN object to these. Examples:

    # Test if dn ends with the contents of base_dn
    if dn.endswith(base_dn):
    # Test if dn starts with a rdn
    if dn.startswith(rdn1):
    # Test if a container is present in a dn
    if container_dn in dn:

    DN objects support concatenation and addition with other DN's or RDN's
    or strings (interpreted as RFC 4514 DN syntax).

    # yields a new DN object with the RDN's of dn2 appended to the RDN's of dn1
    dn1 + dn2

    # yields a new DN object with the rdn1 appended to the RDN's of dn1
    dn1 + rdn1

    DN objects can add RDN's objects via in-place addition.

    dn1 += dn2  # dn2 RDN's are appended to the dn1's RDN's
    dn1 += rdn1 # dn1 has rdn appended to its RDN's
    dn1 += "dc=redhat.com" # string is converted to DN, then appended

    The str method of an DN returns the string representation in RFC 4514 DN
    syntax with proper escaping.
    '''

    AVA_type = AVA
    RDN_type = RDN

    def __init__(self, *args, **kwds):
        self.rdns = self._rdns_from_sequence(args)

    def _copy_rdns(self, rdns=None):
        if not rdns:
            rdns = self.rdns
        return [[list(a) for a in rdn] for rdn in rdns]

    def _rdns_from_value(self, value):
        if isinstance(value, six.string_types):
            try:
                if isinstance(value, six.text_type):
                    value = val_encode(value)
                rdns = str2dn(value)
            except DECODING_ERROR:
                raise ValueError("malformed RDN string = \"%s\"" % value)
            for rdn in rdns:
                sort_avas(rdn)
        elif isinstance(value, DN):
            rdns = value._copy_rdns()
        elif isinstance(value, (tuple, list, AVA)):
            ava = get_ava(value)
            rdns = [[ava]]
        elif isinstance(value, RDN):
            rdns = [value.to_openldap()]
        else:
            raise TypeError("must be str, unicode, tuple, or RDN or DN, got %s instead" %
                            type(value))
        return rdns

    def _rdns_from_sequence(self, seq):
        rdns = []

        for item in seq:
            rdn = self._rdns_from_value(item)
            rdns.extend(rdn)
        return rdns

    def __deepcopy__(self, memo):
        return self

    def _get_rdn(self, rdn):
        return self.RDN_type(*rdn, **{'raw': True})

    def __str__(self):
        return dn2str(self.rdns)

    def __repr__(self):
        return "%s.%s('%s')" % (self.__module__, self.__class__.__name__, self.__str__())

    def _next(self):
        for rdn in self.rdns:
            yield self._get_rdn(rdn)

    def __iter__(self):
        return self._next()

    def __len__(self):
        return len(self.rdns)

    def __getitem__(self, key):
        if isinstance(key, six.integer_types):
            return self._get_rdn(self.rdns[key])
        if isinstance(key, slice):
            cls = self.__class__
            new_dn = cls.__new__(cls)
            new_dn.rdns = self.rdns[key]
            return new_dn
        elif isinstance(key, six.string_types):
            for rdn in self.rdns:
                for ava in rdn:
                    if key == val_decode(ava[0]):
                        return val_decode(ava[1])
            raise KeyError("\"%s\" not found in %s" % (key, self.__str__()))
        else:
            raise TypeError("unsupported type for DN indexing, must be int, basestring or slice; not %s" % \
                                (key.__class__.__name__))

    def __hash__(self):
        # Hash is computed from DN's string representation.
        #
        # Because attrs & values are comparison case-insensitive the
        # hash value between two objects which compare as equal but
        # differ in case must yield the same hash value.

        str_dn = ';,'.join([
            '++'.join([
                '=='.join((atype, avalue or ''))
                for atype, avalue, _dummy in rdn
            ]) for rdn in self.rdns
        ])
        return hash(str_dn.lower())

    def __eq__(self, other):
        # Try coercing to DN, if successful compare to coerced object
        if isinstance(other, (six.string_types, RDN, AVA)):
            try:
                other_dn = DN(other)
                return self.__eq__(other_dn)
            except Exception:
                return False

        # If it's not an DN it can't be equal
        if not isinstance(other, DN):
            return False

        if len(self) != len(other):
            return False

        # Perform comparison between objects of same type
        return self._cmp_sequence(other, 0, len(self)) == 0

    def __ne__(self, other):
        return not self.__eq__(other)

    def __lt__(self, other):
        if not isinstance(other, DN):
            raise TypeError("expected DN but got %s" % (other.__class__.__name__))

        if len(self) != len(other):
            return len(self) < len(other)

        return self._cmp_sequence(other, 0, len(self)) < 0

    def _cmp_sequence(self, pattern, self_start, pat_len):
        self_idx = self_start
        pat_idx = 0
        while pat_idx < pat_len:
            r = cmp_rdns(self.rdns[self_idx], pattern.rdns[pat_idx])
            if r != 0:
                return r
            self_idx += 1
            pat_idx += 1
        return 0

    def __add__(self, other):
        return self.__class__(self, other)

    # The implementation of startswith, endswith, tailmatch, adjust_indices
    # was based on the Python's stringobject.c implementation

    def startswith(self, prefix, start=0, end=sys.maxsize):
        '''
        Return True if the dn starts with the specified prefix (either a DN or
        RDN object), False otherwise.  With optional start, test dn beginning at
        that position.  With optional end, stop comparing dn at that position.
        prefix can also be a tuple of dn's or rdn's to try.
        '''
        if isinstance(prefix, tuple):
            for pat in prefix:
                if self._tailmatch(pat, start, end, -1):
                    return True
            return False

        return self._tailmatch(prefix, start, end, -1)

    def endswith(self, suffix, start=0, end=sys.maxsize):
        '''
        Return True if dn ends with the specified suffix (either a DN or RDN
        object), False otherwise.  With optional start, test dn beginning at
        that position.  With optional end, stop comparing dn at that position.
        suffix can also be a tuple of dn's or rdn's to try.
        '''
        if isinstance(suffix, tuple):
            for pat in suffix:
                if self._tailmatch(pat, start, end, +1):
                    return True
            return False

        return self._tailmatch(suffix, start, end, +1)

    def _tailmatch(self, pattern, start, end, direction):
        '''
        Matches the end (direction >= 0) or start (direction < 0) of self
        against pattern (either a DN or RDN), using the start and end
        arguments. Returns 0 if not found and 1 if found.
        '''

        if isinstance(pattern, RDN):
            pattern = DN(pattern)
        if isinstance(pattern, DN):
            pat_len = len(pattern)
        else:
            raise TypeError("expected DN or RDN but got %s" % (pattern.__class__.__name__))

        self_len = len(self)

        start, end = _adjust_indices(start, end, self_len)

        if direction < 0:       # starswith
            if start+pat_len > self_len:
                return 0
        else:                   # endswith
            if end-start < pat_len or start > self_len:
                return 0

            if end-pat_len >= start:
                start = end - pat_len

        if end-start >= pat_len:
            return not self._cmp_sequence(pattern, start, pat_len)
        return 0


    def __contains__(self, other):
        'Return the outcome of the test other in self. Note the reversed operands.'

        if isinstance(other, RDN):
            other = DN(other)
        if isinstance(other, DN):
            other_len = len(other)
            end = len(self) - other_len
            i = 0
            while i <= end:
                result = self._cmp_sequence(other, i, other_len)
                if result == 0:
                    return True
                i += 1
            return False
        else:
            raise TypeError("expected DN or RDN but got %s" % (other.__class__.__name__))


    def find(self, pattern, start=None, end=None):
        '''
        Return the lowest index in the DN where pattern DN is found,
        such that pattern is contained in the range [start, end]. Optional
        arguments start and end are interpreted as in slice notation. Return
        -1 if pattern is not found.
        '''

        if isinstance(pattern, DN):
            pat_len = len(pattern)
        else:
            raise TypeError("expected DN but got %s" % (pattern.__class__.__name__))

        self_len = len(self)

        if start is None:
            start = 0
        if end is None:
            end = self_len

        start, end = _adjust_indices(start, end, self_len)

        i = start
        stop = max(start, end - pat_len)

        while i <= stop:
            result = self._cmp_sequence(pattern, i, pat_len)
            if result == 0:
                return i
            i += 1
        return -1


    def index(self, pattern, start=None, end=None):
        '''
        Like find() but raise ValueError when the pattern is not found.
        '''

        i = self.find(pattern, start, end)
        if i == -1:
            raise ValueError("pattern not found")
        return i

    def rfind(self, pattern, start=None, end=None):
        '''
        Return the highest index in the DN where pattern DN is found,
        such that pattern is contained in the range [start, end]. Optional
        arguments start and end are interpreted as in slice notation. Return
        -1 if pattern is not found.
        '''

        if isinstance(pattern, DN):
            pat_len = len(pattern)
        else:
            raise TypeError("expected DN but got %s" % (pattern.__class__.__name__))

        self_len = len(self)

        if start is None:
            start = 0
        if end is None:
            end = self_len

        start, end = _adjust_indices(start, end, self_len)

        i = max(start, min(end, self_len - pat_len))
        stop = start

        while i >= stop:
            result = self._cmp_sequence(pattern, i, pat_len)
            if result == 0:
                return i
            i -= 1
        return -1

    def rindex(self, pattern, start=None, end=None):
        '''
        Like rfind() but raise ValueError when the pattern is not found.
        '''

        i = self.rfind(pattern, start, end)
        if i == -1:
            raise ValueError("pattern not found")
        return i