Failed ODE does not return the expected retcode

[gustavoheiden]

Although my ODE is failing by "Error test failures occurred too many times during one internal time step", the integrator.retcode still returned "Success".

After some debugging I found that in if block (line 1477) the integrator.flag is changing from -3 to 0 after evaluating line

Sundials.jl/src/common_interface/solve.jl

Line 1482 in 3fa20aa

integrator(integrator.u, integrator.t, Val{1})

In my local code, I performed this change in order to get the correct retcode from my failed ODE:
if integrator.opts.save_end && (isempty(integrator.sol.t) || integrator.sol.t[end] != integrator.t) && integrator.flag == 0

I hope that helps returning a more accurate error.

[ChrisRackauckas]

I'm not quite sure what you mean here. Could you share a bit more?

[gustavoheiden]

I'll try to explain.
Let's assume the solver returned "error test failed too many times", so we get integrator.flag = -3

Even though we have a failed EDO, the algorithm access this if block

Sundials.jl/src/common_interface/solve.jl

Lines 1477 to 1485 in 3fa20aa

	if integrator.opts.save_end &&
	(isempty(integrator.sol.t) || integrator.sol.t[end] != integrator.t)
	save_value!(integrator.sol.u, integrator.u, uType, integrator.sizeu)
	push!(integrator.sol.t, integrator.t)
	if integrator.opts.dense
	integrator(integrator.u, integrator.t, Val{1})
	save_value!(integrator.sol.interp.du, integrator.u, uType, integrator.sizeu)
	end
	end

Moreover, if you opt for a dense save, the evaluation of line 1482 integrator(integrator.u, integrator.t, Val{1}) changes the state of integrator.flag to 0 (which means success)

That's why I opened this issue, we shouldn't evaluate integrator after its flag < 0, because it can change it to a success state.

[ChrisRackauckas]

Ahh, indeed that's odd and should get fixed.

[gustavoheiden]

I added && integrator.flag == 0 to assure only a success is able to access that if block