Re: [hybi] Is there a traffic jam?

["Sylvain Hellegouarch" <[e-mail]>]

> On Tue, 14 Apr 2009 08:10:37 +0200, Sylvain Hellegouarch <[e-mail]>
> wrote:
>> Indeed. I actually wonder in fact if the problem hasn't been addressed
>> the wrong way around. It seems to me that we've had to discuss so far of
>> a protocol that could basically work out for everything. But could it? I
>> guess most people here have at some point implemented various protocols
>> and feel like WS will not be good at supporting any of them without
>> major headaches and hacks.
>
> This hasn't really been demonstrated. Some people feel that what goes over
> the wire should be identifiable as an intrinsic part of the protocol, but
> I'm still out as to why this is important. (I.e. something more than just
> "string" vs "bytes" vs "...".)

I would assume for multiplexing over one single WS connection. Mind you
the protocol could also impose connections can't be shared.

>
> One concern that was raised is that library authors might not work
> together to do things in common. However, library authors already work
> together on common things, such as supporting the same query language,
> ways of allowing multiple libraries to be used by a single page, et
> cetera. There's a whole "alliance" based around this.
>
>
>> The question I have now is, what does precisely stop browsers from
>> offering a raw TCP socket API at the Javascript level?
>
> The same-origin security policy.

I can appreciate this being a serious issue but somehow I'm fuzzy as to
why the response to that is a new base protocol other protocols should
interface with. From the browser's perspective, I can see why this is so
simple when I go through the WS specification but from a server's
perspective this means adding another layer which means:

* Having to understand for each protocol (XMPP, IMAP, etc.) the best way
to channel said protocols into the WS constraints.
* Having to gauge how to scale such infrastructure.

I find it odd to throw years of experience from the server side to fix one
security issue (even such a serious one).

> The closest to raw TCP socket access we can offer is something that looks
> like Web Sockets.
>

I find that seriously limitating.

- Sylvain
-- 
Sylvain Hellegouarch
http://www.defuze.org