A sequence of connect(), disconnect(), and connect() could cause the
connection to get disconnected again by the old reader thread, which was
blocking on disconnect because a closing stream tag was read. If the
second connect() was processed before the disconnect(), then the
connectin would get disconnected right after the second connect().
This showed up as a "strange" sequence of stanzas in the XMPP servers
log. Note that the stanza ID of the unavailable presence has a lower
number then the previous stanzas:
Jun 11 23:11:11 c2s18c2370 debug Resource bound: smack-inttest-two-93t70@geekplace.eu/two-93t70
Jun 11 23:11:11 c2s18c2370 debug Received[c2s]: <iq id='qn03S-26' type='get'>
Jun 11 23:11:11 c2s18c2370 debug Received[c2s]: <presence id='qn03S-27'>
Jun 11 23:11:11 c2s18c2370 debug Received[c2s]: <presence id='qn03S-23' type='unavailable'>
Jun 11 23:11:11 c2s18c2370 debug Received </stream:stream>
This is because the disconnect() of the first reader thread could
generate the unavailable presence, but was blocked afterwards when
entering the synchronized disconnect(Presence unavailablePresence).
SMACK-633.
If such an exception ever happes when running a integration, then the
framework should not abort, but instead log the exception so that other
information is also logged.
Make Roster entries Map from BareJid to RosterEntry, since only
bare JIDs are allowed as roster items as per RFC 6121 § 3.1.1
When a user sends a presence subscription request to a potential
instant messaging and presence contact, the value of the 'to'
attribute MUST be a bare JID <contact@domainpart> rather than a full
JID <contact@domainpart/resourcepart>,…
Also some further Roster API changes regarding JIDs.
and use weak references.
Disabling the Socks5Manager every time the connection is terminated, and
re-enabling it when it got connected again causes unwanted side
effects. Like adding a new feature to the ServiceDiscoveryManager causes
an update of the entity's capabilities, which then triggers a new outgoing
presence (announcing the new caps version).
SMACK-671
Also don't override eventually send presences on
updateLocalEntityCaps(), instead save the last sent Presence stanza and
re-send that stanza.
SMACK-669.
Instead of allocating one big continuous memory block before "writing"
the XMPP stream element to the socket, write the single CharSequences of
LazyStringBuilder/XmlStringBuilder.
Also change Obserable writer to only notify the listeners after a
flush *or* if a certain limit has been reached since the last
notification. Otherwise the debugger would "print" every single XML part
as result of this change.
