Order of PresetTimeCallback and an ODEFunction is non-intuitive

[Vilin97]

I am trying to understand the order of evaluation of a callback and an ode function. It seems somewhat cryptic. The callback seems to go before the ode function at first but then switches to going after, in effect skipping one step in-between.

using OrdinaryDiffEq, DiffEqCallbacks
tspan = (0.0, 2.0)
dt = 1.
ts = collect(tspan[1]:dt:tspan[2])
g!(integrator) = (u_modified!(integrator, false); println("$(integrator.t) g "))
f!(du,u,p,t) = (du .= u; println("$t f "))
cb = PresetTimeCallback(ts, g!, save_positions=(false, false))
solve(ODEProblem(f!, [1.0], tspan), Euler(), dt=dt, callback=cb)

Output:

0.0 g 
0.0 f 
1.0 f
1.0 g
2.0 f
2.0 g

[ChrisRackauckas]

Described on Slack, this is because tspan[1] is in the time choice and thus initialize! is setup so that it calls the affect! at time 0. This is a special behavior to ensure correctness, given that the user in this instance told us of the special case that there should be an event applied before the start of the simulation.