[Bug]: Error when pickling simulation

[brosaplanella]

PyBaMM Version

23.4.1

Python Version

3.9.16

Describe the bug

When trying to pickle a simulation with experiment I get an error related to a SwigPy object that can't be pickled. I suspect it is a problem with Casadi, but haven't been able to trace the origin of such object.

Steps to Reproduce

Run

import pybamm
model = pybamm.lithium_ion.SPMe()

experiment = pybamm.Experiment(
    [
        (
            "Discharge at 1C until 2.5 V",
            "Charge at C/3 until 4.2 V",
            "Hold at 4.2 V until C/20",
        )
    ]
)

sim = pybamm.Simulation(
    model,
    experiment=experiment,
)
sim.solve()
sim.save("test.pkl")

If instead you replace the last few lines by

sim = pybamm.Simulation(
    model,
)
sim.solve([0, 3600])
sim.save("test.pkl")

it works fine.

Relevant log output

Traceback (most recent call last):
  File "/home/brosaplanella/support-pybamm/mwe_swigpy.py", line 19, in <module>
    sim.save("test.pkl")
  File "/home/brosaplanella/support-pybamm/env/lib/python3.9/site-packages/pybamm/simulation.py", line 1096, in save
    pickle.dump(self, f, pickle.HIGHEST_PROTOCOL)
TypeError: cannot pickle 'SwigPyObject' object

[valentinsulzer]

Pickling simulations is a pain and simulations are super cheap to set up. Do we want to continue allowing this, or suggest that people set up configuration files instead?

[valentinsulzer]

This also kind of relates to #2787

[brosaplanella]

I think that some optimisers pickle (or do something similar to pickling) when running in parallel. It was not where this issues came up, but had similar issues in the past with pybamm-param. I think for that it would be enough to be able to access an unsolved version.

[brosaplanella]

Ok, I found the issue. It is with op_conds_to_build_solvers which keeps a copy of the solver which is not processed for pickling. It should be an easy fix, we only need to process the solvers in op_conds_to_build_solvers similarly as we do in

PyBaMM/pybamm/simulation.py

Lines 1090 to 1094 in a1e64c7

	if (
	isinstance(self._solver, pybamm.CasadiSolver)
	and self._solver.integrator_specs != {}
	):
	self._solver.integrator_specs = {}

Note that this issue appears only after merging #2960. I labelled it as high priority as ideally it would be good to fix that jointly with #2960.