Actor Framework Discussions
Actor Tree vs Network Routing [Branch of "Classifying Actor Messages by their Intent" thread]
[Note to Officer Staab: I know this whole tangent discussion on control/benign messages is better suited to the other thread. But this is where people are talking about it so I'm just going with the flow, you know? Please don't give me a ticket for a topic violation. 
]
fabric wrote: Perhaps the best way to categorise messages as Benign is simply: that the system does not care whether they arrive or not. Obviously that is quite a restrictive definition but at least there is no grey!
The only messages the system "doesn't care whether or not they arrive" are those messages that are filtered out by the receiving actor. However, the receiver's filter (assuming it has one) changes depending on the actor's state. No messages are always filtered out. If the definition is so restrictive very few messages can qualify as benign, there's no value in making the distinction in the first place.
Strictly speaking, I do believe the set of benign messages in a system can change during execution. However, if we're utilizing a secondary messaging routing system for benign messages the messages we send on it have to *always* be benign. We can't risk sending a benign message through a central dispatcher if that message could change into a control message during transit.
fabric wrote:
The best example I can think of are UI update messages, or at least the subset of them that are non-interdependent.
UI updates are control messages in those applications where humans play a critical role in the overall system. Without feedback from the UI the human will do the wrong thing and the system will break down.
fabric wrote:
Surely any messages which have other actions depending on their arrival or ordering are not, by definition, benign?
No, although any message that does not cause an action is, by definition, benign. Except for those messages that are filtered out, all meaningful messages do something when they arrive. The difference between control and benign is determined by the importance of the action that results from the message, not whether or not an action occurs at all.
I see I neglected to continue with this conversation, sorry. I was in the middle of writing my first app using Actors (my own framework) that I had the chance to do in over a year and I was waiting to see if it produced any insight. Unfortunately, the main conclusion is that the apps I work on are too small to throw up any issues of message order!
But hey, here's a screenshot anyway, of the top-level "Slug Main" actor with a schematic diagram of some of the interactions between the five actors involved (actually, 11 actors are running, including a debug logger and five clones of a reusable helper actor that allows the main actors to "publish" messages):
"Slugs", BTW, are short sections of fluid in a tiny capillary that need to be created, measured, and photographed. "Slug Main" delegates most automation to a "Slug Maker" actor, which needs to create slugs through control of a "Linear Slide" actor, and also needs to receive sensor information from a DAQ actor. Note that these messages are "cross-tree" and definitely not "benign". However, the messaging setup is specified in the parent "Slug Main"; see the case shown in the bottom part of the picture where Slug Main registers the Linear Slide to receive Slug Maker's published "Requested Depth" messages (relabelled as "Set Position" to match the accepted command of the Linear Slide). To understand the cross-tree interaction, and if desired to change it to a strict tree topology, I only have to consider/change the code of Slug Main. So in a way, I'm not really breaking the tree topology, as the parent actor is still acting as an intermediary between it's subactors. I imagine a similar thing can be done in the Actor Framework using self-addressed messages created by the parent actor.
-- James
