Complex routing protocols

[ecpeterson]

In network.lisp, one finds the following comment:

;; IMPORTANT NOTE: On reflection, a nondeterministic routing strategy will mean
;;     that it's possible for violate message delivery order, which is no bueno.
;;     Either this function should always be deterministic, or we we will have
;;     to put up with out-of-order behavior, or we will have to reject out-of-
;;     order messages.  For now, let's stick with the first option.

We actually don't need to put up with out-of-order behavior: if a message originator also transmits the ID of the message it previously sent to a given address, the receiving courier can wait until that message is also received before installing this message in the appropriate inbox. At present, send-message is implemented too minimally to support this, and I think making it less minimal would affect one's ability to use it (with this guarantee) outside of the context of a process—but it is possible to make this modification.

This is worth doing, as it opens the door to more complicated routing protocols and the alleviation of network pressure.