timesolver.solve often breaks everything

[lamoreel]

@ilateur : by making stupidly small duration steps

[lamoreel]

What's this about? solve() assures a sensible timestep at the beginning of the first run().

[ilateur]

Sometimes you can call solve twice for a single system, for example from t=0 to t=t1, then from t=t1 to t=t2. (I think) after the first solve, time is not exactly at t1, so then it tries to reach t1, but the timesolver thinks it's already good enough, and it gets stuck in an infinite while loop.
I should have made an example, because sometimes it's completely fine too and it might be hard to recreate.

[lamoreel]

For future reference, this minimal working example reproduces this getting stuck:

from mumaxplus import Ferromagnet, Grid, World
import numpy as np


w = World(cellsize=(1e-9, 1e-9, 1e-9))
magnet = Ferromagnet(w, Grid((5, 5, 5)), "fm")

eps = 1e-17
tp1 = np.linspace(0, 5e-10, 5)
tp2 = np.linspace(5e-10 + eps, 10e-10, 5)

outputquantities = {"m": lambda: magnet.magnetization.average()}

output1 = w.timesolver.solve(tp1, outputquantities)
output2 = w.timesolver.solve(tp2, outputquantities) # fails to compute

[ilateur]

Semi-related: Jonathan Leliaert recommended to keep the time steps fixed/adaptive, and if you overshoot your desired data point it is numerically better to linearly interpolate between the previous and next step.