IE7 with node server on same domain, different port: access denied

[gygelly]

I'm using the socket.io-node server 6.1 for testing and as an example so you can easily recreate the issue.
Everything works great when the browser goes directly to mydomain:8080/chat.html.
However, I'd like to host the files via nginx on port 80 (with the node server still running on port 8080).
Everything works great when I use any browser besides IE7, and go to mydomain/chat.html. This includes IE6 and IE8.
However, when I use IE7, I get the following script error:

Line 338 (I think it's actually line 337, when req.open is called).

Char 3

Error: access is denied

code:0

IE7 is defaulting to htmlfile. It gives the same error when I try to force xhr-multipart, or xhr-polling, and fails without an error when I force flashsocket.
Both IE6 and IE8 also use htmlfile, but work perfectly. All 3 browsers are using the same default security settings of "medium-high".

[rauchg]

Correct, we're only testing host and we should also test port (and possibly protocol?). Will fix asap.

[gygelly]

awesome.

[gygelly]

Just a heads up:
htmlfile works great on IE6 and IE8 in this situation (same host, different port).
If you start testing port, that means that these 2 browsers will also fall back to jsonp-polling, when they could have kept using htmlfile.
Maybe instead use a conditional to test for IE7. Here is my code:

Before chat.html script tags:
<!--[if IE 7]><script>var ie7 = true;</script><![endif]-->

In chat.html script tags:
if (typeof(ie7) !== 'undefined') { var socket = new io.Socket(null, {port: 8080, rememberTransport: false, transports:['jsonp-polling']}); } else { var socket = new io.Socket(null, {port: 8080, rememberTransport: false }); }

[gygelly]

There is one problem with what I just posted above: if for some reason htmlfile fails in ie8, it falls back to xhr-polling, which also fails due to permission errors. So you may want to add another check for IE8, similar to the above, and set it's transports to ['htmlfile','jsonp-polling'] in this situation.

It would help my code if I understood which transports were "better" (more stable, etc.) than others.
Am I correct in thinking that htmlfile is better than jsonp and xhr-polling?
Why is xhr-polling the fallback before jsonp?

[rauchg]

htmlfile definitely has lower overhead (headers are not sent for every message)
If it fallbacks to a non-xdomain transport, then we have a bug in the fallback mechanism.

[rauchg]

But also, keep in mind IE8 supports XDomainRequest, which is xhr-polling over xdomain.

[gygelly]

You are right, I was confused in one of my tests:
IE8 does fall back to xhr-polling before jsonp-polling.
This is also not a problem, as xhr-polling works just fine in IE8. It only gave me problems in IE7.

Basically, you just want to put jsonp before htmlfile (and xhr) in IE7 only.

[rauchg]

We also need to move (and improve) this logic
https://github.com/LearnBoost/Socket.IO/blob/master/lib/socket.js#L147

to the different transports.
For htmlfile, different port is not xdomain, according to what you said.
So basically, Socket should check that if the transport has a isXDomain method it should use that, otherwise use Socket#isXDomain

[gygelly]

actually, it's the browser, host, and port, that you need to look at.
In IE6, IE8, Chrome, and Firefox, same host + different port is not considered cross domain, so everything works, including htmlfile and xhr-polling.
In IE7, same host + different port IS considered cross domain, and so htmlfile and xhr-polling result in an error.

[rauchg]

Fixed