DiscreteCallback finalize not getting called at end of solve().

[staticfloat]

I would expect that at the end of solve(), all finalize callback methods would be getting called, however that does not appear to be the case:

using OrdinaryDiffEq

f(u, p, t) = 1.01*u
u0 = 0.5
cb = DiscreteCallback(
    Returns(false),
    Returns(nothing);
    initialize=(args...)->@info("initialized"),
    finalize=(args...)->@info("finalized"),
)
prob = ODEProblem(f, u0, (0.0, 1.0); callback=cb)
sol = solve(prob, Rodas5P())

Only ever prints initialized

[ChrisRackauckas]

I don't think finalize ever got finalized (pun not intended). It would be good to actually get that completed and documented.

[ChrisRackauckas]

@oscardssmith could you help with this?

[oscardssmith]

Do we have a good example of a use-case for this to add to the tests?

[staticfloat]

It's pretty easy, just create a DiscreteCallback with a finalize, like in the OP

[oscardssmith]

right, but testing for the presence of an @info isn't especially easy. I can always add a test that does something silly (like mutate a global in the finalizer) but a test where the finalization is actually doing something somewhat useful would likely be better at making sure errors aren't introduced (like running multiple times/running at the wrong point in time etc). Also, do we have a strong idea of where in the process this finalize call should happen? The place I put it seems reasonable, but I don't have a great sense of the semantic here other than "after the time-stepping but before the return from solve". Should the finalize be able to affect the final u values before they are saved? Should it be able to restart integration (e.g. by adding an extra tstop)? etc.

[staticfloat]

The docs say:

finalize: This is a function (c,u,t,integrator) which can be used to finalize the state of the callback c. It can modify the argument c and the return is ignored.

So my intuition says that it should not be allowed to affect the final u values, and it should not be allowed to restart integration.

[staticfloat]

I can always add a test that does something silly (like mutate a global in the finalizer)

That's exactly what I would do; just increment a global and check that it was incremented the correct number of times.