Add support for creating an empty Simulation

[rjplevin]

Made it possible to call sim = Simulation() and use the new API to populate sim.

[davidanthoff]

I like the idea, but will let @lrennels do a proper review of the code :)

[lrennels]

What’s the use case here? Doesn’t the new API work about the same but by deep copying the existing sim and then adding a few characteristics like models and N?

[rjplevin]

This was in response to James' issue. It occurred to me that he could allocated an empty sim and use the functional API to add to it in each component's file. We might streamline the API as we discussed.

Alternatively, we could modify @defsim so it also works on a pre-allocated Simulation. This would provide the "incremental defsim" we talked about today. It would basically do what I suggested above, but it would be consistent with the existing approach. In fact, it would be a handy way to modify an existing Simulation, say, returned from Mimi.get_sim(). Maybe call it @update_sim instead.

[lrennels]

Ah ok so this is in line with our conversation at the end today, not just the API we had on the board.

[davidanthoff]

We could also generally think about support for merging simulations? Either via merge(s1, s2), or maybe even something like s1 + s2? Although I kind of like the first option better, because it is more explicit that this operation is not commutative.

[rjplevin]

Yes, merge(s1, s2, ...) seems good. We could also support merge!(s1, s2, ...), as for Dicts. Or We define the merge operation as a constructor Simulation(sim, ...) that returns a new sim.

Maybe a better pattern for James' case: each file with a @defcomp defines a new Simulation instance and the main program merges them.

[lrennels]

Looks like some tests are failing so I will look into that!

[lrennels]

Looks good to me!