On Wed, Jun 02, 2010 at 04:36:58PM -0700, [email protected] wrote:
> On Wed, Jun 02, 2010 at 06:29:16PM -0500, Nicolas Williams wrote:
> > More seriously, how would you write HTTP URIs over Unix domain sockets?
> > http:///path-to-socket/resource/path?  But how would you distinguish the
> > path to the socket from the path to the resource??  Or maybe just
> > http:///resource/path (the path to the socket would be implied, and
> > there could be only one).
> 
> The URI and path to the socket are kept separately, since the underlying
> framework (libcurl) doesn't understand that unix sockets are present,
> and the two are really separate namespaces anyway.  The depot listening
> on the other end of the socket may be proxying for multiple publishers,
> so the URI needs to name something that the depot can use to distinguish
> the requests.  In HTTP 1.1 the Host header is required in all requests.
> This allows the server to disambiguate in situations where it may be
> serving requests for many virtual domains.

Ah, so normal URIs, and the server is actually a proxy.  Got it.  Thanks.
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss

Reply via email to