In no particular order, from person-to-person discussions over the past week;

to those who are against the state structure, let me offer that I've suggested
the alternative of a sub-connection.  Now that may not be the answer, but realize 
that, just as we _must_ have sub-requests to do anything useful in the apache 
http server, other protocols will require sub-connections.  It's a stateful mess.

About request_rec comments;

The answer isn't to deny all of the very useful information in connection and
request records to other protocols.  FTP maps quite well to a bunch of extra
information we collect in the context of http.  The problem is mapping the info
we collect into feedback that is meaningful to the protocol at hand.  I'd hate
for module authors to reinvent the wheel for every protocol they wish to support.

About the security aspect;

For the limited security scope, I've suggested we have a chain of auth structs,
that each type/value that is negotiated is tacked on, and torn down as the user
leaves a given connection or sub-connection.  E.g. their IP address becomes an
identity record, their SSL sig becomes the next identity record, and finally the
basic auth that the resource required becomes the final identity record.  This
would allow us to have very stateful information, in a very dense format, across
a number of negotiations.  [http suffers from this as well, if using NTLM auth.]

And in general;

If we get this all right, https over http [upgrade connection] will fall right
into place.  If not, we know we blew it :-(  Let's try to get resource-oriented 
requests (e.g. scheme://server/resource) into the same general schema so that 
the ftp/http/https details can be as transparent as authors desire.

I'm not writing to the non-resource requests (e.g. smtp and that ilk) since those
will follow an entirely different state and service model.  I'm only arguing that
similar service models should be interchangable, even if the state model changes.

Bill

Reply via email to