Re: [hybi] Is there a traffic jam?

[Maciej Stachowiak <[e-mail]>]

On Apr 14, 2009, at 2:53 AM, Anne van Kesteren wrote:

> On Tue, 14 Apr 2009 10:28:31 +0200, Maciej Stachowiak  
> <[e-mail]> wrote:
>> In my opinion, optional but well-defined subprotocol identification  
>> is another way, since it enabled building more structured protocols  
>> on top in a clear way.
>
> It seems to me that the path can be used for that, but that is not  
> formal enough for your taste currently as I understand it (it's not  
> well-defined). Any ideas as to what would work better? Or maybe as  
> to how to make the path fit the criteria?


Here's some possibilities that I think are better than leaving it  
completely undefined.

1) Optional HTTP response header for protocol type in the server  
handshake response.

2) Define that part of the path defines the type - this could be (a)  
the first path segment, (b) everything after the last dot (the  
"extension", though that is weird to use for protocol type since for  
server side scripts the extension often just reflects the  
implementation technology) or (c) a semicolon-delimited "parameter"  
which is otherwise not much used in the URI syntax.

3) A dedicated initial pair of messages over the channel with a  
specific format.

I think of these solutions, (1) is the cleanest. There's probably  
other possibilities I didn't think of.

Regards,
Maciej