Consider multi-actor sequences

[MattWindsor91]

Eventually, we're going to want sequence diagrams between multiple controllers, or multiple STMs. This is key to get the full expressivity of sequence diagrams, but complicates a lot of things:

• Some things always assume two actors.
• Specification of arrows assumes (semantically and syntactically) an implicit source and destination that can be resolved into two actors: target and world.
• Control flows will be significantly more complex, as they won't always apply to all lifelines.
• Time would presumably be a complete novelty here.

Amusingly,

Some possible ideas and pointers:

• Lucas Lima's semantics scales to arbitrary numbers of actors, but requires a degree of complexity I've hitherto tried to avoid (for readability, performance, and simplicity).
• It might be possible to maintain both two-actor and multi-actor semantics, and keep them in lockstep using tests/proofs, but this might be overkill.
• We might be able to generalise the idea of a message's endpoint specification to either implicit or explicit, with implicit being ill-formed in multi-actor sequences.

[MattWindsor91]

We can reuse some of the syntactic ideas we used to have in older versions of the language. Here's a sketch:

sequence group S
with C1: controller Ctrl1
 and C2: controller Ctrl2
 and C3: controller Ctrl3 {
    sequence S1 {
             [C1->C2]event Evt1
        then [C2->C3]event Evt2
        then [C3->C1]event Evt3
    }
}

In other words, -> and <- would become one kind of endpoint specification, and [X->Y] another. (The square brackets here are just a clue to the parser so that it knows it's dealing with an arrow, but I'm not particularly attached to them.)

[MattWindsor91]

While this change will likely need a large (but not total) change in the semantics, it'll hopefully not massively impact the metamodel.

[MattWindsor91]

This has now been drafted for v0.1.