On 2/5/2010 4:35 PM, Graham Leggett wrote: > > ideally there should be a value r->remote_ip, populated initially from > connection->remote_ip, which a request can change at will, and that will > go away when the request is finished. Modules that want to do access > control, etc should rather look in r->remote_ip instead of > r->connection->remote_ip.
I can almost guarantee that porters will not catch this delta between 2.2 and 2.4. If anything, rather than growing either request_rec or conn_rec, these structures need some fundamental simplification and rethinking. They are 20 year dumping grounds of whatever struck someone's fancy at the time, irrespective of other per-module options for storing data that is not universally interesting. I'd wholeheartedly agree with your concern if http was a bidi protocol requiring connections from the server back to the client, but that isn't the case. remote_ip is informative, not canonical. Do we want an indication of whether remote_host has been transformed to a name, or inferred from context? That might be a good addition to the remote_* members of this tracking class.
