Commentary

edls.rkt

@@ -84,6 +84,26 @@ Are these containery things the conversation/chatrooms of racket-ssh?
 ;; way, no value is forbidden to the userland, including descriptions
 ;; of actions.
 
+;; Can we say that the *communications facility* in EDLS is
+;; underpowered? It's point-to-point and relies on noninterference
+;; between action-handling extensions. Since each extension is
+;; logically managing its own resources, it feels like we have a kind
+;; of network-layer here IF we enforce separation of the managed
+;; resources so there are no cross-channels.
+;;
+;; The reason I'm interested in this approach is to get some kind of
+;; objcap-oriented interface not only to the substrate but to the
+;; extensions embedded in the substrate. Consider the DNS-server case,
+;; where a socket needs to be opened and then straight-line
+;; interactions with the socket take place. Now consider the DNS-proxy
+;; case, where not only does a socket (or more than one!) need to be
+;; created, but complex conversional contexts are built up as each
+;; query arrives and is processed. As DJB observes, a single query can
+;; in principle result in unbounded recursion as "glue" records are
+;; looked up. It kind of makes sense to have each conversational
+;; context as a separate entity, managing its own resources, embedded
+;; in the substrate.
+
 (require racket/match)
 (require racket/class)