gecko/services/sync/modules/xmpp/readme.txt
2008-06-04 17:00:02 -07:00

126 lines
5.4 KiB
Plaintext

About the XMPP module
Here is sample code demonstrating how client code can use the XMPP module.
It assumes that a Jabber server capable of HTTP-Polling is running on localhost
on port 5280.
Components.utils.import( "resource://weave/xmpp/xmppClient.js" );
var serverUrl = "http://127.0.0.1:5280/http-poll";
var jabberDomain = "mylaptop.local";
var transport = new HTTPPollingTransport( serverUrl,
false,
4000 );
// "false" tells the transport not to use session keys. 4000 is the number of
// milliseconds to wait between attempts to poll the server.
var auth = new PlainAuthenticator();
var alice = new XmppClient( "alice", jabberDomain, "iamalice",
transport, auth );
// This sets up an XMPP client for the jabber ID
// "alice@jonathan-dicarlos-macbook-pro.local", who has password
// "iamalice".
// Set up callback for incoming messages:
var aliceMessageHandler = {
handle: function( msgText, from ) {
// Your code goes here. It will be called whenever another XMPP client
// sends a message to Alice.
// msgText is the text of the incoming message, and "from" is the
// jabber ID of the sender.
}
};
alice.registerMessageHandler( aliceMessageHandler );
// Connect
alice.connect( jabberDomain );
alice.waitForConnection();
// this will block until our connection attempt has either succeeded or failed.
// Check if connection succeeded:
if ( alice._connectionStatus == alice.FAILED ) {
// handle error
}
// Send a message
alice.sendMessage( "bob@mylaptop.local", "Hello, Bob." );
// Disconnect
alice.disconnect();
Installing an XMPP server to test against:
I'm using ejabberd, which works well and is open source but is implemented in Erlang. (That's bad, because I don't know how to hack Erlang. I'm thinking of moving to jabberd, implemented in C.)
ejabberd: http://www.ejabberd.im/
jabberd: http://jabberd.jabberstudio.org/
Installation of ejabberd was simple. After it's installed, configure it (create an admin account, and enable http-polling) by editing the configuration file:
ejabberd/conf/ejabberd.cfg
Its web admin interface can be used to create accounts (so that the "alice" and "bob" accounts assumed by the testing code will actually exist). This admin interface is at:
http://localhost:5280/admin/
The ejabberd process is started simply by running:
ejabberd/bin/ejabberdctl start
Outstanding Issues -- bugs and things to do.
* The test above is failing with a timeout. How to debug this? Let's start
by making it two processes again and seeing if that makes a difference...
Nope. It doesn't. Bob just gets an HTTP status 0
* Occasionally (by which I mean, randomly) the server will respond to
a POST with an HTTP status 0. Generally things will go back to
normal with the next POST and then keep working, so it's not
actually preventing anything from working, but I don't understand
what status 0 means and that makes me uncomfortable as it could be
hiding other problems.
What if I modify the sending code to repeat the sending of any messages
that result in HTTP status 0?
-- seems to work, but doesn't help. (It's not HTTP status 0 that's preventing
the test from passing...)
* Occasionally (randomly) a message is received that has a raw integer
for a messageElem, which causes XML parsing of the incoming message
to fail. I don't see any pattern for the values of the raw
integers, nor do I understand how we can get an integer in place of
an XML element when requesting the root node of the incoming XML
document.
* Duplicate messages. Occasionally, even though one instance of the
client sends only a single copy of a <message> stanza, the intended
target will receive it multiple times. Sometimes if the recipient
signs off and then signs back on, it will recieve another copy of
the message. This makes me think it's probably a feature of the
server to resend messages that it thinks the recipient might have
missed while offline. Duplicate messages are a problem especially
when using xmpp to synchronize two processes, for instance. Do I
need to tweak the server settings, or change server implementations?
Do I need to have the client receiving messages acknowledge them to the
server somehow?
Do I need to move to using <iq> stanzas for synchronization? Or
start putting GUIDs into <message>s and discarding duplicates
myself?
* Whenever I send an <iq> request, I get a <bad-request/> error back
from the server. Documentation for the standard error messages
indicates that bad-request is supposed to happen when the type of
the iq element is invalid, i.e. something other than 'set', 'get',
'result', or 'error'. But I'm using 'get' or 'set' and still
getting <bad-request/>. I don't understand what's happening here.
* The authentication layer currently doesn't work with MD5-digest auth, only plain
auth.
* The HTTPPolling transport layer gets a "key sequence error" if useKeys is turned on.
(Everything seems to be working OK with useKeys turned off, but that's less
secure.)