lokesh2019/python-zeroconf: A pure python implementation of multicast DNS service discovery · GitHub

Documentation.

This is fork of pyzeroconf, Multicast DNS Service Discovery for Python, originally by Paul Scott-Murphy (https://github.com/paulsm/pyzeroconf), modified by William McBrine (https://github.com/wmcbrine/pyzeroconf).

The original William McBrine's fork note:

This fork is used in all of my TiVo-related projects: HME for Python
(and therefore HME/VLC), Network Remote, Remote Proxy, and pyTivo.
Before this, I was tracking the changes for zeroconf.py in three
separate repos. I figured I should have an authoritative source.

Although I make changes based on my experience with TiVos, I expect that
they're generally applicable. This version also includes patches found
on the now-defunct (?) Launchpad repo of pyzeroconf, and elsewhere
around the net -- not always well-documented, sorry.

Compatible with:

• Bonjour
• Avahi

Compared to some other Zeroconf/Bonjour/Avahi Python packages, python-zeroconf:

• isn't tied to Bonjour or Avahi
• doesn't use D-Bus
• doesn't force you to use particular event loop or Twisted (asyncio is used under the hood but not required)
• is pip-installable
• has PyPI distribution

• CPython 3.6+
• PyPy3 7.2+

This project's versions follow the following pattern: MAJOR.MINOR.PATCH.

• MAJOR version has been 0 so far
• MINOR version is incremented on backward incompatible changes
• PATCH version is incremented on backward compatible changes

This project is actively maintained.

Before version 0.32, most traffic reduction techniques described in https://datatracker.ietf.org/doc/html/rfc6762#section-7 where not implemented which could lead to excessive network traffic. It is highly recommended that version 0.32 or later is used if this is a concern.

IPv6 support is relatively new and currently limited, specifically:

• InterfaceChoice.All is an alias for InterfaceChoice.Default on non-POSIX systems.
• On Windows specific interfaces can only be requested as interface indexes, not as IP addresses.
• Dual-stack IPv6 sockets are used, which may not be supported everywhere (some BSD variants do not have them).
• Listening on localhost (::1) does not work. Help with understanding why is appreciated.

• PyPI page https://pypi.python.org/pypi/zeroconf
• GitHub project https://github.com/jstasiak/python-zeroconf

The easiest way to install python-zeroconf is using pip:

pip install zeroconf

Here's an example of browsing for a service:

from zeroconf import ServiceBrowser, Zeroconf


class MyListener:

    def remove_service(self, zeroconf, type, name):
        print("Service %s removed" % (name,))

    def add_service(self, zeroconf, type, name):
        info = zeroconf.get_service_info(type, name)
        print("Service %s added, service info: %s" % (name, info))


zeroconf = Zeroconf()
listener = MyListener()
browser = ServiceBrowser(zeroconf, "_http._tcp.local.", listener)
try:
    input("Press enter to exit...\n\n")
finally:
    zeroconf.close()

If you don't know the name of the service you need to browse for, try:

from zeroconf import ZeroconfServiceTypes
print('\n'.join(ZeroconfServiceTypes.find()))

See examples directory for more.

• Only reschedule types if the send next time changes (#958) @bdraco

  When the PTR response was seen again, the timer was being canceled and rescheduled even if the timer was for the same time. While this did not cause any breakage, it is quite inefficient.

• Cache DNS record and question hashes (#960) @bdraco

  The hash was being recalculated every time the object was being used in a set or dict. Since the hashes are effectively immutable, we only calculate them once now.

• Reduced chance of accidental synchronization of ServiceInfo requests (#955) @bdraco
• Sort aggregated responses to increase chance of name compression (#954) @bdraco

Technically backwards incompatible:

• Send unicast replies on the same socket the query was received (#952) @bdraco

  When replying to a QU question, we do not know if the sending host is reachable from all of the sending sockets. We now avoid this problem by replying via the receiving socket. This was the existing behavior when InterfaceChoice.Default is set.

  This change extends the unicast relay behavior to used with InterfaceChoice.Default to apply when InterfaceChoice.All or interfaces are explicitly passed when instantiating a Zeroconf instance.

  Fixes #951

• Fix sending immediate multicast responses (#949) @bdraco

• Coalesce aggregated multicast answers (#945) @bdraco

  When the random delay is shorter than the last scheduled response, answers are now added to the same outgoing time group.

  This reduces traffic when we already know we will be sending a group of answers inside the random delay window described in datatracker.ietf.org/doc/html/rfc6762#section-6.3

• Ensure ServiceInfo requests can be answered inside the default timeout with network protection (#946) @bdraco

  Adjust the time windows to ensure responses that have triggered the protection against against excessive packet flooding due to software bugs or malicious attack described in RFC6762 section 6 can respond in under 1350ms to ensure ServiceInfo can ask two questions within the default timeout of 3000ms

• Ensure multicast aggregation sends responses within 620ms (#942) @bdraco

  Responses that trigger the protection against against excessive packet flooding due to software bugs or malicious attack described in RFC6762 section 6 could cause the multicast aggregation response to be delayed longer than 620ms (The maximum random delay of 120ms and 500ms additional for aggregation).

  Only responses that trigger the protection are delayed longer than 620ms

• Implemented Multicast Response Aggregation (#940) @bdraco

  Responses are now aggregated when possible per rules in RFC6762 section 6.4

  Responses that trigger the protection against against excessive packet flooding due to software bugs or malicious attack described in RFC6762 section 6 are delayed instead of discarding as it was causing responders that implement Passive Observation Of Failures (POOF) to evict the records.

  Probe responses are now always sent immediately as there were cases where they would fail to be answered in time to defend a name.

• Ensure zeroconf can be loaded when the system disables IPv6 (#933) @che0

• Added support for forward dns compression pointers (#934) @bdraco
• Provide sockname when logging a protocol error (#935) @bdraco

• Handle duplicate goodbye answers in the same packet (#928) @bdraco

  Solves an exception being thrown when we tried to remove the known answer from the cache when the second goodbye answer in the same packet was processed

  Fixed #926

• Skip ipv6 interfaces that return ENODEV (#930) @bdraco

• Version number change only with less restrictive directory permissions

  Fixed #923

This release eliminates all threading locks as all non-threadsafe operations now happen in the event loop.

• Let connection_lost close the underlying socket (#918) @bdraco

  The socket was closed during shutdown before asyncio's connection_lost handler had a chance to close it which resulted in a traceback on windows.

  Fixed #917

Technically backwards incompatible:

• Removed duplicate unregister_all_services code (#910) @bdraco

  Calling Zeroconf.close from same asyncio event loop zeroconf is running in will now skip unregister_all_services and log a warning as this a blocking operation and is not async safe and never has been.

  Use AsyncZeroconf instead, or for legacy code call async_unregister_all_services before Zeroconf.close

• Increased timeout in ServiceInfo.request to handle loaded systems (#895) @bdraco

  It can take a few seconds for a loaded system to run the async_request coroutine when the event loop is busy, or the system is CPU bound (example being Home Assistant startup). We now add an additional _LOADED_SYSTEM_TIMEOUT (10s) to the run_coroutine_threadsafe calls to ensure the coroutine has the total amount of time to run up to its internal timeout (default of 3000ms).

  Ten seconds is a bit large of a timeout; however, it is only used in cases where we wrap other timeouts. We now expect the only instance the run_coroutine_threadsafe result timeout will happen in a production circumstance is when someone is running a ServiceInfo.request() in a thread and another thread calls Zeroconf.close() at just the right moment that the future is never completed unless the system is so loaded that it is nearly unresponsive.

  The timeout for run_coroutine_threadsafe is the maximum time a thread can cleanly shut down when zeroconf is closed out in another thread, which should always be longer than the underlying thread operation.

This release offers 100% line and branch coverage.

• Made ServiceInfo first question QU (#852) @bdraco

  We want an immediate response when requesting with ServiceInfo by asking a QU question; most responders will not delay the response and respond right away to our question. This also improves compatibility with split networks as we may not have been able to see the response otherwise. If the responder has not multicast the record recently, it may still choose to do so in addition to responding via unicast

  Reduces traffic when there are multiple zeroconf instances running on the network running ServiceBrowsers

  If we don't get an answer on the first try, we ask a QM question in the event, we can't receive a unicast response for some reason

  This change puts ServiceInfo inline with ServiceBrowser which also asks the first question as QU since ServiceInfo is commonly called from ServiceBrowser callbacks

• Limited duplicate packet suppression to 1s intervals (#841) @bdraco

  Only suppress duplicate packets that happen within the same second. Legitimate queriers will retry the question if they are suppressed. The limit was reduced to one second to be in line with rfc6762

• Made multipacket known answer suppression per interface (#836) @bdraco

  The suppression was happening per instance of Zeroconf instead of per interface. Since the same network can be seen on multiple interfaces (usually and wifi and ethernet), this would confuse the multi-packet known answer supression since it was not expecting to get the same data more than once

• New ServiceBrowsers now request QU in the first outgoing when unspecified (#812) @bdraco

  https://datatracker.ietf.org/doc/html/rfc6762#section-5.4 When we start a ServiceBrowser and zeroconf has just started up, the known answer list will be small. By asking a QU question first, it is likely that we have a large known answer list by the time we ask the QM question a second later (current default which is likely too low but would be a breaking change to increase). This reduces the amount of traffic on the network, and has the secondary advantage that most responders will answer a QU question without the typical delay answering QM questions.

• IPv6 link-local addresses are now qualified with scope_id (#343) @ibygrave

  When a service is advertised on an IPv6 address where the scope is link local, i.e. fe80::/64 (see RFC 4007) the resolved IPv6 address must be extended with the scope_id that identifies through the "%" symbol the local interface to be used when routing to that address. A new API parsed_scoped_addresses() is provided to return qualified addresses to avoid breaking compatibility on the existing parsed_addresses().

• Network adapters that are disconnected are now skipped (#327) @ZLJasonG

• Fixed listeners missing initial packets if Engine starts too quickly (#387) @bdraco

  When manually creating a zeroconf.Engine object, it is no longer started automatically. It must manually be started by calling .start() on the created object.

  The Engine thread is now started after all the listeners have been added to avoid a race condition where packets could be missed at startup.

• Fixed answering matching PTR queries with the ANY query (#618) @bdraco

• Fixed lookup of uppercase names in the registry (#597) @bdraco

  If the ServiceInfo was registered with an uppercase name and the query was for a lowercase name, it would not be found and vice-versa.

• Fixed unicast responses from any source port (#598) @bdraco

  Unicast responses were only being sent if the source port was 53, this prevented responses when testing with dig:

  dig -p 5353 @224.0.0.251 media-12.local

  The above query will now see a response

• Fixed queries for AAAA records not being answered (#616) @bdraco

• Removed second level caching from ServiceBrowsers (#737) @bdraco

  The ServiceBrowser had its own cache of the last time it saw a service that was reimplementing the DNSCache and presenting a source of truth problem that lead to unexpected queries when the two disagreed.

• Fixed server cache not being case-insensitive (#731) @bdraco

  If the server name had uppercase chars and any of the matching records were lowercase, and the server would not be found

• Fixed cache handling of records with different TTLs (#729) @bdraco

  There should only be one unique record in the cache at a time as having multiple unique records will different TTLs in the cache can result in unexpected behavior since some functions returned all matching records and some fetched from the right side of the list to return the newest record. Instead we now store the records in a dict to ensure that the newest record always replaces the same unique record, and we never have a source of truth problem determining the TTL of a record from the cache.

• Fixed ServiceInfo with multiple A records (#725) @bdraco

  If there were multiple A records for the host, ServiceInfo would always return the last one that was in the incoming packet, which was usually not the one that was wanted.

• Fixed stale unique records expiring too quickly (#706) @bdraco

  Records now expire 1s in the future instead of instant removal.

  tools.ietf.org/html/rfc6762#section-10.2 Queriers receiving a Multicast DNS response with a TTL of zero SHOULD NOT immediately delete the record from the cache, but instead record a TTL of 1 and then delete the record one second later. In the case of multiple Multicast DNS responders on the network described in Section 6.6 above, if one of the responders shuts down and incorrectly sends goodbye packets for its records, it gives the other cooperating responders one second to send out their own response to "rescue" the records before they expire and are deleted.

• Fixed exception when unregistering a service multiple times (#679) @bdraco

• Added an AsyncZeroconfServiceTypes to mirror ZeroconfServiceTypes to zeroconf.asyncio (#658) @bdraco

• Fixed interface_index_to_ip6_address not skiping ipv4 adapters (#651) @bdraco

• Added async_unregister_all_services to AsyncZeroconf (#649) @bdraco

• Fixed services not being removed from the registry when calling unregister_all_services (#644) @bdraco

  There was a race condition where a query could be answered for a service in the registry, while goodbye packets which could result in a fresh record being broadcast after the goodbye if a query came in at just the right time. To avoid this, we now remove the services from the registry right after we generate the goodbye packet

• Fixed zeroconf exception on load when the system disables IPv6 (#624) @bdraco

• Fixed the QU bit missing from for probe queries (#609) @bdraco

  The bit should be set per datatracker.ietf.org/doc/html/rfc6762#section-8.1

• Fixed the TC bit missing for query packets where the known answers span multiple packets (#494) @bdraco

• Fixed packets not being properly separated when exceeding maximum size (#498) @bdraco

  Ensure that questions that exceed the max packet size are moved to the next packet. This fixes DNSQuestions being sent in multiple packets in violation of: datatracker.ietf.org/doc/html/rfc6762#section-7.2

  Ensure only one resource record is sent when a record exceeds _MAX_MSG_TYPICAL datatracker.ietf.org/doc/html/rfc6762#section-17

• Fixed PTR questions asked in uppercase not being answered (#465) @bdraco

• Added Support for context managers in Zeroconf and AsyncZeroconf (#284) @shenek

• Implemented an AsyncServiceBrowser to compliment the sync ServiceBrowser (#429) @bdraco

• Added async_get_service_info to AsyncZeroconf and async_request to AsyncServiceInfo (#408) @bdraco

• Implemented allowing passing in a sync Zeroconf instance to AsyncZeroconf (#406) @bdraco

• Fixed IPv6 setup under MacOS when binding to "" (#392) @bdraco

• Fixed ZeroconfServiceTypes.find not always cancels the ServiceBrowser (#389) @bdraco

  There was a short window where the ServiceBrowser thread could be left running after Zeroconf is closed because the .join() was never waited for when a new Zeroconf object was created

• Fixed duplicate packets triggering duplicate updates (#376) @bdraco

  If TXT or SRV records update was already processed and then received again, it was possible for a second update to be called back in the ServiceBrowser

• Fixed ServiceStateChange.Updated event happening for IPs that already existed (#375) @bdraco

• Fixed RFC6762 Section 10.2 paragraph 2 compliance (#374) @bdraco

• Reduced length of ServiceBrowser thread name with many types (#373) @bdraco

• Fixed empty answers being added in ServiceInfo.request (#367) @bdraco

• Fixed ServiceInfo not populating all AAAA records (#366) @bdraco

  Use get_all_by_details to ensure all records are loaded into addresses.

  Only load A/AAAA records from the cache once in load_from_cache if there is a SRV record present

  Move duplicate code that checked if the ServiceInfo was complete into its own function

• Fixed a case where the cache list can change during iteration (#363) @bdraco

• Return task objects created by AsyncZeroconf (#360) @nocarryr

Traffic Reduction:

• Added support for handling QU questions (#621) @bdraco

  Implements RFC 6762 sec 5.4: Questions Requesting Unicast Responses datatracker.ietf.org/doc/html/rfc6762#section-5.4

• Implemented protect the network against excessive packet flooding (#619) @bdraco

• Additionals are now suppressed when they are already in the answers section (#617) @bdraco

• Additionals are no longer included when the answer is suppressed by known-answer suppression (#614) @bdraco

• Implemented multi-packet known answer supression (#687) @bdraco

  Implements datatracker.ietf.org/doc/html/rfc6762#section-7.2

• Implemented efficient bucketing of queries with known answers (#698) @bdraco

• Implemented duplicate question suppression (#770) @bdraco

  http://datatracker.ietf.org/doc/html/rfc6762#section-7.3

Technically backwards incompatible:

• Update internal version check to match docs (3.6+) (#491) @bdraco

  Python version earlier then 3.6 were likely broken with zeroconf already, however, the version is now explicitly checked.

• Update python compatibility as PyPy3 7.2 is required (#523) @bdraco

Backwards incompatible:

• Drop oversize packets before processing them (#826) @bdraco

  Oversized packets can quickly overwhelm the system and deny service to legitimate queriers. In practice, this is usually due to broken mDNS implementations rather than malicious actors.

• Guard against excessive ServiceBrowser queries from PTR records significantly lowerthan recommended (#824) @bdraco

  We now enforce a minimum TTL for PTR records to avoid ServiceBrowsers generating excessive queries refresh queries. Apple uses a 15s minimum TTL, however, we do not have the same level of rate limit and safeguards, so we use 1/4 of the recommended value.

• RecordUpdateListener now uses async_update_records instead of update_record (#419, #726) @bdraco

  This allows the listener to receive all the records that have been updated in a single transaction such as a packet or cache expiry.

  update_record has been deprecated in favor of async_update_records A compatibility shim exists to ensure classes that use RecordUpdateListener as a base class continue to have update_record called, however, they should be updated as soon as possible.

  A new method async_update_records_complete is now called on each listener when all listeners have completed processing updates and the cache has been updated. This allows ServiceBrowsers to delay calling handlers until they are sure the cache has been updated as its a common pattern to call for ServiceInfo when a ServiceBrowser handler fires.

  The async_ prefix was chosen to make it clear that these functions run in the eventloop and should never do blocking I/O. Before 0.32+ these functions ran in a select() loop and should not have been doing any blocking I/O, but it was not clear to implementors that I/O would block the loop.

• Pass both the new and old records to async_update_records (#792) @bdraco

  Pass the old_record (cached) as the value and the new_record (wire) to async_update_records instead of forcing each consumer to check the cache since we will always have the old_record when generating the async_update_records call. This avoids the overhead of multiple cache lookups for each listener.

• Separated cache loading from I/O in ServiceInfo and fixed cache lookup (#356), thanks to J. Nick Koston.

  The ServiceInfo class gained a load_from_cache() method to only fetch information from Zeroconf cache (if it exists) with no IO performed. Additionally this should reduce IO in cases where cache lookups were previously incorrectly failing.

• Some nice refactoring work including removal of the Reaper thread, thanks to J. Nick Koston.
• Fixed a Windows-specific The requested address is not valid in its context regression, thanks to Timothee ‘TTimo’ Besset and J. Nick Koston.
• Provided an asyncio-compatible service registration layer (in the zeroconf.asyncio module), thanks to J. Nick Koston.

• A single socket is used for listening on responding when InterfaceChoice.Default is chosen. Thanks to J. Nick Koston.

Backwards incompatible:

• Dropped Python 3.5 support

• Fixed the packet generation when multiple packets are necessary, previously invalid packets were generated sometimes. Patch thanks to J. Nick Koston.

• Fixed the IPv6 address rendering in the browser example, thanks to Alexey Vazhnov.
• Fixed a crash happening when a service is added or removed during handle_response and improved exception handling, thanks to J. Nick Koston.

• Loosened service name validation when receiving from the network this lets us handle some real world devices previously causing errors, thanks to J. Nick Koston.

• Enabled ignoring duplicated messages which decreases CPU usage, thanks to J. Nick Koston.
• Fixed spurious AttributeError: module 'unittest' has no attribute 'mock' in tests.

• Improved cache reaper performance significantly, thanks to J. Nick Koston.
• Added ServiceListener to __all__ as it's part of the public API, thanks to Justin Nesselrotte.

• Reduced a time an internal lock is held which should eliminate deadlocks in high-traffic networks, thanks to J. Nick Koston.

• Stopped asking questions we already have answers for in cache, thanks to Paul Daumlechner.
• Removed initial delay before querying for service info, thanks to Erik Montnemery.

• Fixed a resource leak connected to using ServiceBrowser with multiple types, thanks to J. Nick Koston.

• Improved Windows support when using socket errno checks, thanks to Sandy Patterson.
• Added support for passing text addresses to ServiceInfo.
• Improved logging (includes fixing an incorrect logging call)
• Improved Windows compatibility by using Adapter.index from ifaddr, thanks to PhilippSelenium.
• Improved Windows compatibility by stopping using socket.if_nameindex.
• Fixed an OS X edge case which should also eliminate a memory leak, thanks to Emil Styrke.

Technically backwards incompatible:

• ifaddr 0.1.7 or newer is required now.

• Improved the logging situation (includes fixing a false-positive "packets() made no progress adding records", thanks to Greg Badros)

• Large multi-resource responses are now split into separate packets which fixes a bad mdns-repeater/ChromeCast Audio interaction ending with ChromeCast Audio crash (and possibly some others) and improves RFC 6762 compliance, thanks to Greg Badros
• Added a warning presented when the listener passed to ServiceBrowser lacks update_service() callback
• Added support for finding all services available in the browser example, thanks to Perry Kunder

Backwards incompatible:

• Removed previously deprecated ServiceInfo address constructor parameter and property

• Improved readability of logged incoming data, thanks to Erik Montnemery
• Threads are given unique names now to aid debugging, thanks to Erik Montnemery
• Fixed a regression where get_service_info() called within a listener add_service method would deadlock, timeout and incorrectly return None, fix thanks to Erik Montnemery, but Matt Saxon and Hmmbob were also involved in debugging it.

• Added support for multiple types to ServiceBrowser, thanks to J. Nick Koston
• Fixed a race condition where a listener gets a message before the lock is created, thanks to J. Nick Koston

• Fixed a performance regression introduced in 0.26.0, thanks to J. Nick Koston (this is close in spirit to an optimization made in 0.24.5 by the same author)

• Fixed a regression where service update listener wasn't called on IP address change (it's called on SRV/A/AAAA record changes now), thanks to Matt Saxon

Technically backwards incompatible:

• Service update hook is no longer called on service addition (service added hook is still called), this is related to the fix above

• Eliminated 5s hangup when calling Zeroconf.close(), thanks to Erik Montnemery

• Reverted uniqueness assertions when browsing, they caused a regression

Backwards incompatible:

• Rationalized handling of TXT records. Non-bytes values are converted to str and encoded to bytes using UTF-8 now, None values mean value-less attributes. When receiving TXT records no decoding is performed now, keys are always bytes and values are either bytes or None in value-less attributes.

• Fixed issues with shared records being used where they shouldn't be (TXT, SRV, A records are unique now), thanks to Matt Saxon
• Stopped unnecessarily excluding host-only interfaces from InterfaceChoice.all as they don't forbid multicast, thanks to Andreas Oberritter
• Fixed repr() of IPv6 DNSAddress, thanks to Aldo Hoeben
• Removed duplicate update messages sent to listeners, thanks to Matt Saxon
• Added support for cooperating responders, thanks to Matt Saxon
• Optimized handle_response cache check, thanks to J. Nick Koston
• Fixed memory leak in DNSCache, thanks to J. Nick Koston

• Fixed resetting TTL in DNSRecord.reset_ttl(), thanks to Matt Saxon
• Improved various DNS class' string representations, thanks to Jay Hogg

• Fixed import-time "TypeError: 'ellipsis' object is not iterable." on CPython 3.5.2

• Added support for AWDL interface on macOS (needed and used by the opendrop project but should be useful in general), thanks to Milan Stute
• Added missing type hints

• Applied some significant performance optimizations, thanks to Jaime van Kessel for the patch and to Ghostkeeper for performance measurements
• Fixed flushing outdated cache entries when incoming record is unique, thanks to Michael Hu
• Fixed handling updates of TXT records (they'd not get recorded previously), thanks to Michael Hu

• Added IPv6 support, thanks to Dmitry Tantsur
• Added additional recommended records to PTR responses, thanks to Scott Mertz
• Added handling of ENOTCONN being raised during shutdown when using Eventlet, thanks to Tamás Nepusz
• Included the py.typed marker in the package so that type checkers know to use type hints from the source code, thanks to Dmitry Tantsur

• Added support for MyListener call getting updates to service TXT records, thanks to Matt Saxon
• Added support for multiple addresses when publishing a service, getting/setting single address has become deprecated. Change thanks to Dmitry Tantsur

Backwards incompatible:

• Dropped Python 3.4 support

• A lot of maintenance work (tooling, typing coverage and improvements, spelling) done, thanks to Ville Skyttä
• Provided saner defaults in ServiceInfo's constructor, thanks to Jorge Miranda
• Fixed service removal packets not being sent on shutdown, thanks to Andrew Bonney
• Added a way to define TTL-s through ServiceInfo contructor parameters, thanks to Andrew Bonney

Technically backwards incompatible:

• Adjusted query intervals to match RFC 6762, thanks to Andrew Bonney
• Made default TTL-s match RFC 6762, thanks to Andrew Bonney

• This time really allowed incoming service names to contain underscores (patch released as part of 0.21.0 was defective)

• Fixed import-time typing-related TypeError when older typing version is used

• Fixed installation on Python 3.4 (we use typing now but there was no explicit dependency on it)

• Added an error message when importing the package using unsupported Python version
• Fixed TTL handling for published service
• Implemented unicast support
• Fixed WSL (Windows Subsystem for Linux) compatibility
• Fixed occasional UnboundLocalError issue
• Fixed UTF-8 multibyte name compression
• Switched from netifaces to ifaddr (pure Python)
• Allowed incoming service names to contain underscores

• Dropped support for Python 2 (this includes PyPy) and 3.3
• Fixed some class' equality operators
• ServiceBrowser entries are being refreshed when 'stale' now
• Cache returns new records first now instead of last

• Allowed installation with netifaces >= 0.10.6 (a bug that was concerning us got fixed)

• Technically backwards incompatible - restricted netifaces dependency version to work around a bug, see python-zeroconf#84 for details

• Dropped Python 2.6 support
• Improved error handling inside code executed when Zeroconf object is being closed

• Better Handling of DNS Incoming Packets parsing exceptions
• Many exceptions will now log a warning the first time they are seen
• Catch and log sendto() errors
• Fix/Implement duplicate name change
• Fix overly strict name validation introduced in 0.17.6
• Greatly improve handling of oversized packets including:
  • Implement name compression per RFC1035
  • Limit size of generated packets to 9000 bytes as per RFC6762
  • Better handle over sized incoming packets
• Increased test coverage to 95%

• Many improvements to address race conditions and exceptions during ZC() startup and shutdown, thanks to: morpav, veawor, justingiorgi, herczy, stephenrauch
• Added more test coverage: strahlex, stephenrauch
• Stephen Rauch contributed:
  • Speed up browser startup
  • Add ZeroconfServiceTypes() query class to discover all advertised service types
  • Add full validation for service names, types and subtypes
  • Fix for subtype browsing
  • Fix DNSHInfo support

• Fixed OpenBSD compatibility, thanks to Alessio Sergi
• Fixed race condition on ServiceBrowser startup, thanks to gbiddison
• Fixed installation on some Python 3 systems, thanks to Per Sandström
• Fixed "size change during iteration" bug on Python 3, thanks to gbiddison

• Fixed support for Linux kernel versions < 3.9 (thanks to Giovanni Harting and Luckydonald, GitHub pull request #26)

• Fixed DNSText repr on Python 3 (it'd crash when the text was longer than 10 bytes), thanks to Paulus Schoutsen for the patch, GitHub pull request #24

• Fixed installation on Python 3.4.3+ (was failing because of enum34 dependency which fails to install on 3.4.3+, changed to depend on enum-compat instead; thanks to Michael Brennan for the original patch, GitHub pull request #22)

• Fixed EADDRNOTAVAIL when attempting to use dummy network interfaces on Windows, thanks to daid

• Added some Python dependencies so it's not zero-dependencies anymore
• Improved exception handling (it'll be quieter now)
• Messages are listened to and sent using all available network interfaces by default (configurable); thanks to Marcus Müller
• Started using logging more freely
• Fixed a bug with binary strings as property values being converted to False (python-zeroconf#10); thanks to Dr. Seuss
• Added new ServiceBrowser event handler interface (see the examples)
• PyPy3 now officially supported
• Fixed ServiceInfo repr on Python 3, thanks to Yordan Miladinov

• Set up Python logging and started using it
• Cleaned up code style (includes migrating from camel case to snake case)

• Fixed handling closed socket (GitHub #4)

• Forked by Jakub Stasiak
• Made Python 3 compatible
• Added setup script, made installable by pip and uploaded to PyPI
• Set up Travis build
• Reformatted the code and moved files around
• Stopped catching BaseException in several places, that could hide errors
• Marked threads as daemonic, they won't keep application alive now

• Fix for SOL_IP undefined on some systems - thanks Mike Erdely.
• Cleaned up examples.
• Lowercased module name.

• Various minor changes; see git for details.
• No longer compatible with Python 2.2. Only tested with 2.5-2.7.
• Fork by William McBrine.

• allow selection of binding interface
• typo fix - Thanks A. M. Kuchlingi
• removed all use of word 'Rendezvous' - this is an API change

• correction to comments for addListener method
• support for new record types seen from OS X - IPv6 address - hostinfo
• ignore unknown DNS record types
• fixes to name decoding
• works alongside other processes using port 5353 (e.g. on Mac OS X)
• tested against Mac OS X 10.3.2's mDNSResponder
• corrections to removal of list entries for service browser

• Jonathon Paisley contributed these corrections:
  • always multicast replies, even when query is unicast
  • correct a pointer encoding problem
  • can now write records in any order
  • traceback shown on failure
  • better TXT record parsing
  • server is now separate from name
  • can cancel a service browser
• modified some unit tests to accommodate these changes

• remove all records on service unregistration
• fix DOS security problem with readName

• changed licensing to LGPL

• faster shutdown on engine
• pointer encoding of outgoing names
• ServiceBrowser now works
• new unit tests

• small improvements with unit tests
• added defined exception types
• new style objects
• fixed hostname/interface problem
• fixed socket timeout problem
• fixed add_service_listener() typo bug
• using select() for socket reads
• tested on Debian unstable with Python 2.2.2

• ensure case insensitivty on domain names
• support for unicast DNS queries

• added some unit tests
• added __ne__ adjuncts where required
• ensure names end in '.local.'
• timeout on receiving socket for clean shutdown

LGPL, see COPYING file for details.