Python 3.4 started raising ValueError instead of AttributeError when writing to
a disconnected SSL socket. When the client writes some stray messages following
a call to disconnect(), they are not caught in Python 3.4 or Python 3.5.
Signed-off-by: NJason Swails <jason.swails@gmail.com>

Closes #42. Thanks to Akash Goswami.

Change-Id: I1e409a140e0505ae5fd699352d78ffe9b3e38f5b

Change-Id: I1b239bbe8f607e4795108dc469c878d946e67216

Change-Id: I61e27ac2496b731afa2b70996aee816c030b3fbf

This error is not catched and stops loop_forever etc., that is very
unexpected. So we catch the error and raise a error that is actually
catched (and better represents the situation)

See also: http://bugs.python.org/issue13721

Change-Id: I495c08ec079290a97caeaf0e7f88116041d8cec6
Signed-off-by: NHmvp <hmvp@hmvp.nl>

If the connection in loop_start() does not complete before loop_stop()
is called the network thread never ends.

Thanks to Kees Bakker.

Bug: https://bugs.eclipse.org/bugs/show_bug.cgi?id=448428
Change-Id: If44b313cd0e642b4bf98182d8e50c4da6a8d975c

Change-Id: Ic81bcceed68303216402ce755dbc38380174965a

Publishes with qos > 0 which are sent before a CONACK is received are still counted as inflight
even though they don't really get sent.  This means they get sent after the retry period (20 sec default)

But this won't work when max_inflight is 1 since this means no retry will get sent and we wait untill keepalive is exhausted
This then forces a reconnect and will reset stuff so messages can be sent.

Change-Id: I516a6cce8e5bbfbd3030711a80044af4e5bf6733
Signed-off-by: NHmvp <hmvp@hmvp.nl>

Change-Id: I4770c18969456ea28ef648494b0a1c9bac3bd86b

Handle exceptions from select() in client loop() function. Closes #443881.
Thanks to Jeff Jasper.

Bug: https://bugs.eclipse.org/bugs/show_bug.cgi?id=443881
Change-Id: I4bbee80554917b61a0d413dc490370391a6fe2b5

Document the "protocol" argument to the client constructor. Closes
 #440169. Thanks to Andy Piper.

Bug: https://bugs.eclipse.org/bugs/show_bug.cgi?id=440169
Change-Id: I46467156237235adff41eda0bf98c26bfba083bf

Fix possible race condition when connecting with TLS and publishing at
the same time, which could lead to PUBLISH data being sent before any
other messages and unencrypted. Closes #443964. Thanks to Hiram van
Paassen.

Bug: https://bugs.eclipse.org/bugs/show_bug.cgi?id=443964
Change-Id: I3945ed7d470b69fc939343fc21beffcd7315c502

Fix reconnecting after sending more QoS>0 messages than inflight messages is
set to, whilst connecting.  Closes #443935. Thanks to Hiram van Paassen.

Bug: https://bugs.eclipse.org/bugs/show_bug.cgi?id=443935
Change-Id: I31ddf2ffcbd6a7efa1e5b51c163d112ce3cf61ae

Change-Id: Idcea1dcd2b22b757d467b0ad5478b48a515afc73

Thanks to Luc Milland.

Change-Id: I1f3bf49ed3845f61cbd31773fcf11a58943f8108

Thanks to Luc Milland.

Change-Id: I954db79a6cf332350337c5147ac716cb3e6c4e2a

Bug: https://bugs.eclipse.org/bugs/show_bug.cgi?id=439277
Change-Id: I05d268e4fbaba56116ea35bc8b2dc2c7f31ca2a9

Change-Id: I4225b09eec7586deed80c2ddada314b647154096

Fix incorrect reconnect that occurred if calling loop_stop() before
disconnect().

Change-Id: I50c0880b1773a0570511ed6473f81f3a6d13d4c8

Change-Id: I661f3ef2d54c6f5a25111cb0de4d8a4c128a4c80

Change-Id: I736892bb099592fc226b2ebc889169f4a149e06a

Host certificates that have the subjectAltName needs this to be parsed correctly when used with tls_set(cert)

Change-Id: Iacd70e81278b7cbb6eba69076fb6ec8dbcc36084
Signed-off-by: NHmvp <hmvp@hmvp.nl>
Bug: 436379

Change-Id: Icbcb603f4e5f42ff2fa35fb780ba5410248c1db0

Bug: 436615
Change-Id: Ia1219f72e090fc8ed0cb87863c56c9ee8e713b5d

Change-Id: I9e8d8a58d32339760ecd0e56be496e8753a38522

Change-Id: I5890f79f7ecfc06be9d959456545d7733c604aa2

Connecting from both client.py and publish.py will use MQTT v3.1.1 by
default.

Client will reconnect using MQTT v3.1 if a v3.1.1 connection fails due to
the incorrect protocol version number.

Bug: https://bugs.eclipse.org/bugs/show_bug.cgi?id=434143
Change-Id: I34c4288535b99c8a4f05d92f9e3c99d1fa07b5c1

Change-Id: Ic7b095c00581e2ec7dc7fc89b227a38f3491ce01

Change-Id: I340301dab8f9171fbb209b71e360ac0d3b114048

Change-Id: Id8833f29c2054baece19ad42be6c9dd4cfcacb4c

Thanks to Tobias Assarsson.

Change-Id: If98891597ac47488a072389d483dba15cf54c520

Change-Id: I5445a4b2677f39dd6f5873f4d740abd4fa9a6665

Simplify overall socket error handling.

Fixes bug #431698.

Change-Id: Ia291959b8abb701355d651c4ace442d03c3b0808

Change-Id: I7a3f73037cc11333316cfd8503aad23ce70e8661

Change-Id: I376a0df26196a7c377652c101bc13381008190af

Change-Id: Id0a08ab798a9675ce18da4962d56172522248b5b

Change-Id: Ic6a39e870f6ff1db89c78096f8629397bfeb08da

When a call to publish()/subscribe()/unsubscribe()/disconnect() is made
and the client is using the threaded mode, the outgoing command will be
delayed by 0.5 seconds on average. This commit breaks out of select() to
remove the delay.

Change-Id: I7a669604c767abb0bcc399bdf1fe3b82c7ef9539