ph4 stopping conditions and reproducibility

[rieder]

I recently encountered a weird issue with the gravity_collision.py example: it turned out that the result of this script is not reproducible - even though initial conditions and random seed are identical.
After a lot of searching, it looks like this originates from very small roundoff errors in ph4.
To fix it, I replaced the following (summarised) code:

        gravity.evolve_model(time + time_step)
        time = gravity.model_time
        if stopping_condition.is_set():
            handle_stopping_conditions()

with the following:

        gravity.evolve_model(time + time_step)
        while stopping_condition.is_set():
            handle_stopping_conditions()
            gravity.evolve_model(time + time_step)
        time = gravity.model_time

The problem with the former is that ph4 stops at a "weird" time - when it detects a collision would happen - and then is instructed by AMUSE to take that time as a starting point for the next step. The fixed code instead ignores this moment (though it handles the collision), and continues to the original end time.
I haven't checked, but I expect the time of collision is determined via the position and velocities of the particles, which may introduce roundoff errors which would then propagate.
So, be warned if you use stopping conditions: preferably use these in a while loop rather than with an if statement!

[rieder]

The fix is not complete yet, unfortunately. Maybe more is needed on the ph4 side.