Dnspython now uses a single source for Python 2 and Python 3,
eliminating the painful merging between the Python 2 and Python 3
branches.  Thank you so much to Arthur Gautier for taking on this
challenge and making it work!  It was a big job!

Support for Python older than 2.6 dropped.

Support for Python older than 3.3 dropped.

Zone origin can be specified as a string.

A rich string representation for all DNSExceptions.

setuptools has replaced distutils

Added support for CAA, CDS, CDNSKEY, EUI48, EUI64, and URI RR
types.

Names now support the pickle protocol.

Ports can be specified per-nameserver in the stub resolver.

A number of Unicode name bugs have been fixed.

resolv.conf processing now rejects lines with too few tokens.

NameDicts now keep the max-depth value correct, and update properly.

Added dns.zone.to_text().

Added support for "options rotate" in /etc/resolv.conf.

dns.rdtypes.ANY.DNSKEY now has helpers functions to convert
between the numeric form of the flags and a set of
human-friendly strings

The reverse name of an IPv6 mapped IPv4 address is now in the
IPv4 reverse namespace.

The test system can now run the tests without requiring
dnspython to be installed.

Preliminary Elliptic Curve DNSSEC Validation (requires ecdsa module)

dnspython raised an exception when reading a masterfile starting
with leading whitespace

dnspython was affected by a python slicing API bug present on
64-bit windows.

Unicode escaping was applied at the wrong time.

RRSIG to_text() did not respect the relativize setting.

APL RRs with zero rdlength were rejected.

The tokenizer could put back an unescaped token.

Making a response to a message signed with TSIG was broken.

The IXFR state machine didn't handle long IXFR diffs.

Nothing

dns.resolver.Resolver erroneously referred to 'retry_servfail'
instead of 'self.retry_servfail'.

 	dns.tsigkeyring.to_text() would fail trying to convert the
 	keyname to text.

Multi-message TSIGs were broken for algorithms other than
HMAC-MD5 because we weren't passing the right digest module to
the HMAC code.

dns.dnssec._find_candidate_keys() tried to extract the key
from the wrong variable name.

$GENERATE tests were not backward compatible with python 2.4.

    $GENERATE support

TLSA RR support

Added set_flags() method to dns.resolver.Resolver

    Names with offsets >= 2^14 are no longer added to the compression
table.

The "::" syntax is not used to shorten a single 16-bit section of
the text form an IPv6 address.

Caches are now locked.

YXDOMAIN is raised if seen by the resolver.

Empty rdatasets are not printed.

DNSKEY key tags are no longer assumed to be unique.

   	Added dns.resolver.LRUCache.  In this cache implementation,
the cache size is limited to a user-specified number of nodes,
and when adding a new node to a full cache the least-recently
used node is removed.  If you're crawling the web or otherwise
doing lots of resolutions and you are using a cache, switching
to the LRUCache is recommended.

dns.resolver.query() will try TCP if a UDP response is
truncated.

The python socket module's DNS methods can be now be overriden
with implementations that use dnspython's resolver.

Old DNSSEC types KEY, NXT, and SIG have been removed.

Whitespace is allowed in SSHFP fingerprints.

Origin checking in dns.zone.from_xfr() can be disabled.

Trailing junk checking can be disabled.

A source port can be specified when creating a resolver query.

All EDNS values may now be specified to dns.message.make_query().

 	IPv4 and IPv6 address processing is now stricter.

Bounds checking of slices in rdata wire processing is now more
strict, and bounds errors (e.g. we got less data than was
expected) now raise dns.exception.FormError rather than
IndexError.

Specifying a source port without specifying source used to
have no effect, but now uses the wildcard address and the
specified port.

    Nothing.

 	The rdata _wire_cmp() routine now handles relative names.

The SIG RR implementation was missing 'import struct'.

	A boolean parameter, 'raise_on_no_answer', has been added to
the query() methods.  In no-error, no-data situations, this
parameter determines whether NoAnswer should be raised or not.
If True, NoAnswer is raised.  If False, then an Answer()
object with a None rrset will be returned.

Resolver Answer() objects now have a canonical_name field.

Rdata now have a __hash__ method.

   	Dnspython was erroneously doing case-insensitive comparisons
of the names in NSEC and RRSIG RRs.

We now use "is" and not "==" when testing what section an RR
is in.

The resolver now disallows metaqueries.

	Nothing.

The dns.dnssec module didn't work at all due to missing
imports that escaped detection in testing because the test
suite also did the imports.  The third time is the charm!

	Nothing.

    The dns.dnssec module didn't work with DSA due to namespace
contamination from a "from"-style import.

	dnspython now uses poll() instead of select() when available.

Basic DNSSEC validation can be done using dns.dnsec.validate()
and dns.dnssec.validate_rrsig() if you have PyCrypto 2.3 or
later installed.  Complete secure resolution is not yet
available.

Added key_id() to the DNSSEC module, which computes the DNSSEC
key id of a DNSKEY rdata.

Added make_ds() to the DNSSEC module, which returns the DS RR
for a given DNSKEY rdata.

dnspython now raises an exception if HMAC-SHA284 or
HMAC-SHA512 are used with a Python older than 2.5.2.  (Older
Pythons do not compute the correct value.)

Symbolic constants are now available for TSIG algorithm names.

    dns.resolver.zone_for_name() didn't handle a query response
with a CNAME or DNAME correctly in some cases.

    When specifying rdata types and classes as text, Unicode
strings may now be used.

Hashlib compatibility issues have been fixed.

dns.message now imports dns.edns.

The TSIG algorithm value was passed incorrectly to use_tsig()
in some cases.

	Support for hmac-sha1, hmac-sha224, hmac-sha256, hmac-sha384
and hmac-sha512 has been contributed by Kevin Chen.

The tokenizer's tokens are now Token objects instead of (type,
value) tuples.

    Escapes in masterfiles now work correctly.  Previously they
were only working correctly when the text involved was part of
a domain name.

 	When constructing a DDNS update, if the present() method was
used with a single rdata, a zero TTL was not added.

The entropy pool needed locking to be thread safe.

The entropy pool's reading of /dev/random could cause
dnspython to block.

The entropy pool did buffered reads, potentially consuming more
randomness than we needed.

The entropy pool did not seed with high quality randomness on
Windows.

SRV records were compared incorrectly.

In the e164 query function, the resolver parameter was not
used.

	Nothing

 	The 1.7.0 kitting process inadventently omitted the code for the
DLV RR.

Negative DDNS prerequisites are now handled correctly.

	Rdatas now have a to_digestable() method, which returns the
DNSSEC canonical form of the rdata, suitable for use in
signature computations.

The NSEC3, NSEC3PARAM, DLV, and HIP RR types are now supported.

An entropy module has been added and is used to randomize query ids.

EDNS0 options are now supported.

UDP IXFR is now supported.

The wire format parser now has a 'one_rr_per_rrset' mode, which
suppresses the usual coalescing of all RRs of a given type into a
single RRset.

Various helpful DNSSEC-related constants are now defined.

The resolver's query() method now has an optional 'source' parameter,
    allowing the source IP address to be specified.

 	On Windows, the resolver set the domain incorrectly.

DS RR parsing only allowed one Base64 chunk.

TSIG validation didn't always use absolute names.

NSEC.to_text() only printed the last window.

We did not canonicalize IPv6 addresses before comparing them; we
would thus treat equivalent but different textual forms, e.g.
"1:00::1" and "1::1" as being non-equivalent.

If the peer set a TSIG error, we didn't raise an exception.

Some EDNS bugs in the message code have been fixed (see the ChangeLog
for details).

If select() raises an exception due to EINTR, we should just
select() again.

If the queried address is a multicast address, then don't
check that the address of the response is the same as the
address queried.

NAPTR comparisons didn't compare the preference field due to a
typo.

Testing of whether a Windows NIC is enabled now works on Vista
thanks to code contributed by Paul Marks.

	Answer objects now support more of the python sequence
protocol, forwarding the requests to the answer rrset.
E.g. "for a in answer" is equivalent to "for a in
answer.rrset", "answer[i]" is equivalent to "answer.rrset[i]",
and "answer[i:j]" is equivalent to "answer.rrset[i:j]".

Making requests using EDNS, including indicating DNSSEC awareness,
is now easier.  For example, you can now say:

   q = dns.message.make_query('www.dnspython.org', 'MX',
			      want_dnssec=True)

dns.query.xfr() can now be used for IXFR.

Support has been added for the DHCID, IPSECKEY, and SPF RR types.

UDP messages from unexpected sources can now be ignored by
setting ignore_unexpected to True when calling dns.query.udp.

    If /etc/resolv.conf didn't exist, we raised an exception
instead of simply using the default resolver configuration.

In dns.resolver.Resolver._config_win32_fromkey(), we were
passing the wrong variable to self._config_win32_search().

    You can now convert E.164 numbers to/from their ENUM name
    forms:

      >>> import dns.e164
      >>> n = dns.e164.from_e164("+1 555 1212")
      >>> n
      <DNS name 2.1.2.1.5.5.5.1.e164.arpa.>
      >>> dns.e164.to_e164(n)
      '+15551212'

You can now convert IPv4 and IPv6 address to/from their
corresponding DNS reverse map names:

      >>> import dns.reversename
      >>> n = dns.reversename.from_address("127.0.0.1")
      >>> n
      <DNS name 1.0.0.127.in-addr.arpa.>
      >>> dns.reversename.to_address(n)
      '127.0.0.1'

You can now convert between Unicode strings and their IDN ACE
form:

      >>> n = dns.name.from_text(u'les-\u00e9l\u00e8ves.example.')
      >>> n
      <DNS name xn--les-lves-50ai.example.>
      >>> n.to_unicode()
      u'les-\xe9l\xe8ves.example.'

The origin parameter to dns.zone.from_text() and dns.zone.to_text()
is now optional.  If not specified, the origin will be taken from
the first $ORIGIN statement in the master file.

Sanity checking of a zone can be disabled; this is useful when
working with files which are zone fragments.

 	The correct delimiter was not used when retrieving the
list of nameservers from the registry in certain versions of
windows.

    The floating-point version of latitude and longitude in LOC RRs
(float_latitude and float_longitude) had incorrect signs for
south latitudes and west longitudes.

BIND 8 TTL syntax is now accepted in all TTL-like places (i.e.
SOA fields refresh, retry, expire, and minimum; SIG/RRSIG
field original_ttl).

TTLs are now bounds checked when their text form is parsed,
and their values must be in the closed interval [0, 2^31 - 1].

 	In the resolver, if time goes backward a little bit, ignore
	it.

zone_for_name() has been added to the resolver module.  It
returns the zone which is authoritative for the specified
name, which is handy for dynamic update.  E.g.

      import dns.resolver
      print dns.resolver.zone_for_name('www.dnspython.org')

will output "dnspython.org." and

      print dns.resolver.zone_for_name('a.b.c.d.e.f.example.')

will output ".".

The default resolver can be fetched with the
get_default_resolver() method.

	You can now get the parent (immediate superdomain) of a name
by using the parent() method.

Zone.iterate_rdatasets() and Zone.iterate_rdatas() now have
a default rdtype of dns.rdatatype.ANY like the documentation
says.

A Dynamic DNS example, ddns.py, has been added.

The source address and port may now be specified when calling
dns.query.{udp,tcp,xfr}.

The resolver now does exponential backoff each time it runs
through all of the nameservers.

Rcodes which indicate a nameserver is likely to be a
"permanent failure" for a query cause the nameserver to be removed
from the mix for that query.

	dns.message.Message.find_rrset() now uses an index, vastly
improving the from_wire() performance of large messages such
as zone transfers.

Added dns.message.make_response(), which creates a skeletal
response for the specified query.

Added opcode() and set_opcode() convenience methods to the
dns.message.Message class.  Added the request_payload
attribute to the Message class.

    The 'file' parameter of dns.name.Name.to_wire() is now
optional; if omitted, the wire form will be returned as the
value of the function.

dns.zone.from_xfr() in relativization mode incorrectly set
zone.origin to the empty name.

The masterfile parser incorrectly rejected TXT records where a
value was not quoted.

The NSEC format doesn't allow specifying types by number, so
we shouldn't either.  (Using the unknown type format is still
OK though.)

The resolver wasn't catching dns.exception.Timeout, so a timeout
erroneously caused the whole resolution to fail instead of just
going on to the next server.

The renderer module didn't import random, causing an exception
to be raised if a query id wasn't provided when a Renderer was
created.

    The conversion of LOC milliseconds values from text to binary was
incorrect if the length of the milliseconds string was not 3.

Added support for the SSHFP type.

Added support for new DNSSEC types RRSIG, NSEC, and DNSKEY.

http://www.dnspython.org/