It appears that we observe a partion GC run on some systems,
especially ones with few resources. Hopefully this increases the
chances to observe the expected GC affects so that the unit test
passes also on those systems.
Because an failing test on Travis was observered using a timeout of
60s:
org.jivesoftware.smackx.muc.MucMemoryLeakTest > mucMemoryLeakTest FAILED
java.lang.AssertionError: No reference found after 60000ms
at org.junit.Assert.fail(Assert.java:88)
at org.junit.Assert.assertTrue(Assert.java:41)
at org.junit.Assert.assertNotNull(Assert.java:712)
at org.jivesoftware.smack.util.MemoryLeakTestUtil.assertReferencesQueueSize(MemoryLeakTestUtil.java:110)
at org.jivesoftware.smack.util.MemoryLeakTestUtil.noResourceLeakTest(MemoryLeakTestUtil.java:103)
at org.jivesoftware.smackx.muc.MucMemoryLeakTest.mucMemoryLeakTest(MucMemoryLeakTest.java:29)
Introducing Smack's own XmlPullParser interface which tries to stay as
compatible as possible to XPP3. The interface is used to either wrap
StAX's XMLStreamReader if Smack is used on Java SE, and XPP3's
XmlPullParser if Smack is used on on Android.
Fixes SMACK-591.
Also introduce JUnit 5 and non-strict javadoc projects.
and also call notifyConnectionError() on exception thrown by
openStream().
In hindsight I wonder why openStream() was ever called in the writer
thread, as it only caused unnecessary synchronization overhead, as can
be seen by the initialOpenStreamSend synchronization point.
Smack will throw an IllegalArguementException if extremely
long reply timeouts are tried to be set. I assumed currentTimeMilli()
to be the boundary condition as per SMACK-718
XEP-0060 prohibits publishing a request that contains an item to a node that
is both 'notification-only' and 'transient' (section 7.1.3.6)
In commit 8ed872ca63 the existing pubsub publication
test was modified (to resolve a different issue) to operate on a node that's both
'notification-only' and 'transient'. This resulted in a test that should return
an error, even though the test implementation didn't expect one.
This commit explicitly verifies that publishing an item to such a node causes
an error to be returned. A new test is added that verifies that publishing an
event notification does succeed.
