The lambda we schedule in 25ms captures a strong reference to the
XMPPConnection. However the lambda is part of the scheduled action,
which we save in the renewEntityCapsScheduledAction field. This causes
a memory leak, since the ServiceDiscoveryManager now holds a strong
reference to its XMPPConnection.
Fix this by obtaining the strong reference to the XMPPConnection, if
one still exists, within the lambda.
Fixes SMACK-926.
Reported-by: Damian Minkov <damencho@jitsi.org>
Presence interceptors where hooked into stanza sending listeners,
which are called *after* the stanza has been put on the wire, i.e., to
late for interceptors that any modifications, they may perform, to
take effect.
Fixes SMACK-925.
We now also dynamically add the MUC's main presence interceptor to the
connection.
Reported-by: Damian Minkov <damencho@jitsi.org>
This method is meant to provide 'raw' access to what has been sent
over the wire. It is assumed to be not of much use in the typically
case, but provided for completeness.
According to XEP-0004 § 3.3, the default value of a boolean form field
is 'false'. And since users are typically interested in getting the
value, and not potentially 'null' as result, we adjust the behavior of
the getValueAsBoolean() method consider the default value.
Since notify() is a rather expensive operation, we should only invoke
it once. Especially since some servers include 110 in all self
presences, not just the initially reflected one on MUC join.
The change in 52a49769f9 ("[muc] Check for self-presence first in
presence listener") caused all self-presences (MUC user status 110) to
be ignored in the further processing chain eventually invoking
checkRoleModifications() and checkAffiliationModifications(). However,
some servers (e.g., ejabberd) include 110 not only in the initial
presence but in all following, and we ant to process these.
Fixes SMACK-918
Fixes: 52a49769f9
Eventually JingleTransportManager should be a subclass of Manager (or
be replaced by Manager), as JingleTransportManager holds a strong
reference to the XMPPConnection. This could cause memory leaks. But
for now, we mimic the Manager API in JingleTransportManger to make a
future transition to Manager easier.
This allows to avoid redundant XML namespaces within IQs, like for
example here:
<iq xmlns='jabber:client' id='EKP8I-1' type='set'>
<jingle xmlns='urn:xmpp:jingle:1' action='content-accept' sid='MySession'>
<content xmlns='urn:xmpp:jingle:1' creator='initiator' name='Hello world'>
</content>
</jingle>
</iq>
Fixes SMACK-917
Reported-by: Jonathan Lennox
Calling userHasLeft before sending the leave presence may result in
invalid state as the MUC presence lister may modify the MUCs local
state, e.g., occupantsMap, conurrently with the leave operation.
If we reset it after the leave presence was send and acknowledged,
then this can not happen as the server will not longer send any MUC
related presences to us.
Also fixes SMACK-914. In theory 52a49769f9 ("[muc] Check for
self-presence first in presence listener") alone would fix SMACK-914,
but this also fixes it indepentendly of 52a49769f9. Both commits are
sensible, so both are applied.
In 9352225f44 ("Rework SOCKS5 unit tests so that they can be run in
parallel") the call to getSocks5Proxy() in
Socks5BytestreamManager.getLocalStreamHost() was removed. Since
getSocks5Proxy() does also start the local proxy, if it is not already
running, this caused Smack to no longer automatically start the local
proxy.
This commit re-adds the call to getSocks5Proxy() and fixes SMACK-912.
