A discussion in application design and event propagation

[Fred Hebert]

I'm starting this thread to spark a discussion about Erlang software design.
I've been writing a few applications here and there and I've noticed a few
common patterns occurring over time, but with not clear solution to them.
That seems to be due to the general "there is more than one way to do it"
attitude that exists when organizing processes, and not  because of a lack
of activism.

Anyway, the design question comes from designs having to do with multi-user
activities, might it be chat rooms, games of cards over the web, voting
system, a pub-sub mechanism, etc. The general idea is that you have a group
of users and a sequence of events that has to be forwarded to every user in
real time. To make this clearer, I'll take the example of a blackjack table
where every player has a GUI showing the state of the game.

You'd have (I imagine) one process representing the dealer and the table
(D). Then you have 3 players, P1, P2 and P3, each represented as a process.
A hand distribution goes like this:


   1. Player 1 registers to the table
   2. Player 2 registers. Player 1 is notified of this.
   3. Player 3 registers, P1 and P2 are notified.
   4. The game starts, All players are notified of this.
   5. Each player (and the dealer) is handed a card
   6. Players bet. All players are notified of this;
   7. ...
   8. etc.

Then you could also have some messages private (imagine having a private
hand as in a poker game or whatever).

There are two obvious ways I can think of to model this. The first one is as
a gen_event process and handlers, where all players register an event
handler. When an event occurs, either the dealer or the players notify the
event manager which then forwards the event to subscribers through their
respective event handlers. The second way I can think of are process groups,
where a message sent to the leader is automatically forwarded to all
processes who are members of the group.

It seems to me these are equivalent solutions, as they allow for the same
propagation of events in correct order. However, the former runs the
filtering and forwarding functions in the event manager's process and the
latter does it in each of the group's processes. This makes it look like the
process groups would make it easier to scale (you can forward the message at
once and then the evaluation takes place wherever it needs to be), although
it would have a higher amount of messages being sent around because no
filtering could take place beforehand.

So here's to launch the discussion. What would you think would be the best
choice in this case? and otherwise, what other options do you see? What are
the pros and the cons of each? Both options seem valid, but I don't remember
seeing them being discussed too much.