ci on everything, including performance tests

[MaxOstrowski]

No description provided.

[MaxOstrowski]

This probably can't be merged as it will trigger CI with performance check which is currently degraded.

@AbdallahS @kstrauch94 current performance regression is in

E       AssertionError: propagator benchmark large_int_domain_performance average runtime 198.377s exceeded 95.000s over 1 measured runs (durations=198.377s)
E       assert 198.37650261400086 <= 95.0
E        +  where 95.0 = PerformanceBenchmark(name='large_int_domain_performance', engine='propagator', max_average_seconds=95.0, measured_runs=1, warmup_runs=0, check_mode=False, constants={'int_domain_size': 3000}).max_average_seconds

can we fix this fast or should we set a slower bound in the performance tests.

[kstrauch94]

I am not quite sure why it would be so much slower. I don't see any obvious places where performance would be affected so much. I would say that you can set slower bound for now.

[MaxOstrowski]

Not very satisfying to half the performance without reason, but feel free to adjust the timings.
Have you checked why a single constraint, adding 1 to a number takes 200seconds for the propagator?

[kstrauch94]

Basically, clingo can solve it during grounding/preprocessing. There is no search at all. The propagator doesn't, so it goes through a lot of choices before finding the right one. If the propagator could do inference it would be much faster, probably.

Also, the amount of choices for the propagator seems to be huge since the constraints that enforce at most 1 value for a variable are lazily added. So, it tries to have x have mutiple values constantly.