Erlang and the OCP

Erlang and the OCP

Eric Smith
Eric Smith

May 24, 2008

OCP was defined in 1988 in Bertrand Meyer’s book “Object Oriented Software Construction” as follows: “Modules should be both open (for extension and adaptation) and closed (to avoid modification that affect clients).”

Recently we had an issue in my office with our system which uses Rinda (a Ruby implentation of Linda) as a blackboard architecture. The architecture looks a little like this crude drawing.

Rinda (a Ruby implentation of Linda) as a blackboard architecture.

This hides the OCP violation. The rindlet, our informal name for these subprocesses, actually removes the message from the blackboard. So if you send a message that looked like [1, 2, 3] and one rindlet took it then no other rindlets can receive it.

To repeat myself, this meant all messages had to have as their first data member the name of the rindlet they were sending the message too. This is bad. Adding another rindlet meant changing the client sending the message.

Now to quote Agile Software Development: Principles, Patterns, and Practices [Martin 2003] An axis of change is an axis of change only if the changes actually occur. That is to say if only one rindlet cared about a message than this wouldn’t be an issue.

When this became a problem was when more than one rindlet wanted the same message. Suddenly we have to send many otherwise identical messages to different rindlets.

This is the heart of the OCP, the Server (in this case the one sending the messages), should be closed for implementation, but is not. In our case we changed the architecture to shield the server from those changes, and fixed the issue.

This brings me to Erlang and its message passing system. Last year I began playing with it, but lost interest as the difficulty of the syntax simply ruined the fun for me.

I do know enough about it that when this issue came up in our ruby system, I was immediately reminded of Erlang’s message passing system. Specifically it has three methods [Armstrong 2007 pg. 134]:

  1. Spawn - Create a new concurrent process that evaluates a passed in Fun. The new process runs in parallel with the creating process.

  2. Pid ! Message - Send a message to a Pid - returned by the previous spawn.

  3. Receive…end - Receives a message that has been sent to a process.

On a cursory glance it appears we have the same issue since you must know the Pid of the process. Aha, but the command sending a message to the process actually created the process, and therein lies the difference.

The blackboard architecture deliberately decouples the constantly running processes from messages, the messages are sent to the blackboard. The sender has no right to know about the constantly running process, and should not need to change to add another one.

Erlang on the other hand spawns processes as if they were objects, and sends them messages. Sending messages, heck that’s what you do with an object too isn’t it? I don’t think creating an object and sending it messages is a violation of OCP, and neither is this.

The sender can be extended without effecting clients. It turns out that Erlang does not intrinsically violate the OPC like I’d originally surmised, although like any other system I could create an architecture that did.