Sim with load state does not produce the same edge data output

[gladysmcgan]

Hi, I am getting different results for a run with load state and a run without.

1. Simulation, saving simulation state at interval 55800.
   sumo-gui -c test.sumocfg --save-state.rng --save-state.times 55800

2. Simulation, loading simulation state from above, starting at 55800.
   sumo-gui -c test_load_state.sumocfg

Comparing from interval 55800, the edge data from the two simulations gave slightly different outputs, though I would expect them to be the same. I'm looking at the edge data of one edge (-695666936)

sim files are here: issue_test_save_state.zip

Would be grateful for your help!

[namdre]

identified causes so far:

• queue entryBlockTime is not preserved when saving/loading state #16770
• rerouting device must save history of estimated speeds #16772
• timing of simulation rerouting differs after loading stime #16775
• stochastic device assignment is not reproduced when loading state #16784
• state-saving must preserve loading/insertion order for vehicles that depart in the same step #16870
• probabilistic device assignment differs when loading state #16871
• behavior after loading state differs with --meso-overtaking #16874
• behavior after loading state differs with --weights.random-factor #16876
• edgeData must restore myLastVehicleUpdateValues when loading state #16894

[namdre]

there seem to be at least three more issues

• one related to --meso-overtaking
• one related to --weights.random-factor
• one more

[gladysmcgan]

Thank you for the fixes so far! Is there any configuration we could try now that can reproduce save states?

[namdre]

No. I haven't figured out all of the remaining issues and thus cannot recommend a workaround yet.

[namdre]

Thank you for the fixes so far! Is there any configuration we could try now that can reproduce save states?

Todays nightly build should reproduce from saved state when running without option --weights.random-factor.

[namdre]

after adfc9dc, your simulation replicates from loaded state.

[gladysmcgan]

Thanks for the fixes! I tried the above test case with its configs, also without weights.random-factor, but unfortunately I'm still getting some differences in the edge data output.. I used Version v1_23_1+1041-39398899bf9.

Though they start to show the same output at the last few intervals in the simulation