Correct way to pass jump times to step size controllers if interpolation evaluated as part of vector field

[johannahaffner]

What is the correct way to pass information about knots ts of a linear interpolation to the step size controller? I'm using Kvaerno5 for a stiff ODE and specify experimental input $u(t)$ with a linear interpolation.

I noticed that the FSAL solver must be informed about the knots (small wiggles around them), and included the information about the knot times as a jump_ts argument to the step size controller, as specified here in the warning.
However, I have not put my linear interpolation into a control term - I simply pass it to the vector field as an argument and then evaluate it.

I then get ValueError: A fixed step size controller is being used alongside an implicit solver, but the tolerances for the implicit solver have not been specified. (Being unspecified is the default in Diffrax.) ... (no traceback).

This seems like a bit of a footgun? I don't intend to make the PID controller implement a fixed step size just because it now has to respect jump times, but that does seem to be the net effect (?).

[patrick-kidger]

Ah, I think this is a known bug in v0.7.0, and one of the reasons I really need to (find the time for) a new release already. The error message is spurious. Try the latest main / latest dev, it should be fixed on either.

So indeed jump_ts would be the appropriate way to inform the stepsize controller about the jumps in the vector field. As this is with a PIDController then the step size controller is already adaptive, and as such it should be suitable for use alongside an implicit solver.