Message

Messages are the means of passing Control in an ObjectOriented system. Objects communicate by sending messages. When code in one object sends a message to another object, control passes to the second object, which must then figure out how to process the message.

Typically, ObjectOrientedProgramming languages have special syntax for defining UnitsOfCode that handle specific messages—these handlers being traditionally referred to as Methods (though note that my definition of a Method is slightly broader). However, it's possible to do OO without special method syntax, as AnjanaVakil demonstrates (TODO link to talk)

Strictly speaking, messages are Values, though object-oriented languages like Java and Ruby lack first-class values and represent messages as Object references. The problem with using objects as messages is that it breaks encapsulation, defeating the object-oriented principle that objects may only communicate by message-passing. As an example of the problem: if two objects A and B are communicating, A may send a reference to a third object C to B. Since both A and B then have a reference to C, B might communicate back to A by changing C's State. Another thing to note is that implementing literally all messages as objects leads to an absurd infinite regress: if messages are objects, then the only way to extract information from a message is to send a message, meaning that every message received causes at least one other message to be sent.

An open question I need to explore: if all messages are values, how does the object graph get built? It seems like sometimes you want an object to be shared (e.g. the ObserverPattern and VisitorPattern).