Use a reference of problem for set_problem()

[mhovd]

During simulation it may be necessary to update the state, which also requires the problem. Previously set_problem() consumed problem, but this change instead calls for a reference. The problem itself is small, and as such the clone should not be especially expensive.

[mhovd]

Alternatively a different solution focused on updating the state without requiring a new call to set_problem()? It feels like an XY problem.

[martinjrobins]

thanks @mhovd, few minor changes below to reduce some lines of unnessessary code:

[martinjrobins]

Alternatively a different solution focused on updating the state without requiring a new call to set_problem()? It feels like an XY problem.

As I see it, the old solver method needs to know if the state has been altered, so it can reset its own internal state. You could either:

• have the ode solver method own the state, so to alter the state you need to call a method on the ode solver method
• have a reset() function on the ode solver method so it knows to reset its internal state (but keep the current problem). This will be pretty similar to the current set_problem solution, just with one less arg

[Siel]

I understand that sometimes it is necessary to update the state (and if we're planning to support events this will be happening all the time). But I'm not sure about the approach.

Wouldn't it be better if we have a sort of higher level orchestrator that sub divides the time vector to stop the solver at the event's time, and then modifies the state by changing the initial condition of the next simulation?

I know there is some overhead to this approach, namely the need to interpolate for every event time, but this approach (I think) can be more easily extrapolated when new solvers are implemented.

[martinjrobins]

Wouldn't it be better if we have a sort of higher level orchestrator that sub divides the time vector to stop the solver at the event's time, and then modifies the state by changing the initial condition of the next simulation?

Agreed, I think this PR is to sort out the api on the solver methods that this orchestrator could use to implement this.

[martinjrobins]

looks great, thanks @mhovd