For certain use cases, this provides a performance improvement,
probably due better cache locality. However, it comes with the cost of
additional memory consumption.
This was initially suggested by Boris Grozev, who also reported a
significant performance problem of
XmlStringBuilder/LazyStringBuilder. However, the main cause of the
performance probelm was the missing caching of LazyStringBuilder. The
length of the lazy string is now cached by LazyStringBuidler since
70e48300a6 ("[core] Cache length in LazyStringBuilder"), which
accounts for large performance improvement. A significantly smaller
improvement is achieved by this commit and setting
XmlStringBuilder.FLAT_APPEND to 'true'.
Suggested-by: Boris Grozev <boris@jitsi.org>
With 92f253cc74 ("[core] Replace
AbstractXMPPConnection.inOrderListeners") we changes the behavior of
listeners. This commit documents the expectations to the user.
The default local address is often just "the first address found in the list of addresses read from the OS" and this might mean an internal IP address that cannot reach external servers. So wherever possible use the same IP address being used to connect to the XMPP server because this local address has a better chance of being suitable.
This MR adds the above behaviour, and two UTs to test that we use the local XMPP connection IP when connected, and the previous behaviour when not.
Also delete StanzaIdTest since the test was fundamentally weak and
flawed. It does not work anyway, since TestIQ has a fixed stanza ID.
Fixes SMACK-931.
While markdown is easier to write, Smack's markdown documentation was
never tightly coupled with the source. For example, the markdown
documentation never provided links to the actual Java classes and
methods. This poses the risk that the documentation and the code
diverge over time. Furthermore, javadoc is constantly improving (for
example @snippet annotations) and I expect that one will be able to
write javadoc in markdown.
Fixes SMACK-928.
This option is meant a quick and dirty hack until
XMPPConnection.sendStanza() throws a dedicated Exception in case the
connection's outgoing queue is full.
Using any of the two methods is error prone, see 0e0c0a4093 ("[tcp]
Fix handling in connection exceptions when resuming a stream"), as one
can easily forget to check for connection exceptions after it
returned. There are also no longer any call sites of those methods.
Let's inline both variants of this method.
Smack historically provided fine-grained options for the
SSLContext. This is however not flexible enough, as an option to
specifiy the KeyManager(s) was missing.
This deprecated the options for keystore path, keystore type, and
PKCS#11 library, in favor of an option to set the KeyManager, which
could be aware of the keystore path and type, and the PKCS#11
library. At some point, Smack may provide some high-level methods to
create a KeyManager from provided keystore path, keytsore type and
PKCS#11 library.
Previously, if a SmackFuture both was successful and unsuccessful, it
was possible that the onSuccess() callback was invoked twice.
Reported-by: Boris Grozev <boris@jitsi.org>
All other enum-using methods of XmlStringBuilder already use
Enum.toString(), as opposed to Enum.name(), this was the only left. I
do not remember why I did not to change this method too, probably
because of its plenty call sites.
But since this method already broke Jingle XML serializaton,
JingleAction was e.g., 'session_accept' when it should be
'session-accept', we change it now.
Fixes SMACK-921.
Connections like XMPPTCPConnection may still reported connected, if
they are, for example disconnected but resumable. This is already
accounted for in throwAlreadyConnectedExceptionIfAppropriate(), hence
the assert is unnecessary and leads to false negatives for
XMPPTCPConnection.
For the sake of completeness, the right condition, assuming
XMPPTCPConnection is used, for the assert would be:
assert !connected || disconnectedButResumable;
