Bugfixes to DEVS

[quaquel]

Summary

This PR fixes several bugs in the experimental DEVS (Discrete Event Simulation) module and its usage in the wolf-sheep benchmark model. The changes improve initialization safety and timing consistency in event scheduling.

Motive

Several subtle bugs were discovered while working on #2470:

1. The simulator's setup process could silently overwrite already scheduled events
2. Models could attempt to run without proper initialization
3. The ABMSimulator had incorrect timing for the first step event

These issues needed to be addressed to ensure reliable event scheduling and simulation execution.

Implementation

The changes include:

1. Enhanced setup validation:

   • Added checks to prevent setup when time != starttime
   • Added checks to prevent setup when events are already scheduled
   • Added validation in run methods to ensure setup has been called

2. Moved simulator setup responsibility:

   • Models that schedule events during initialization must now call simulator.setup() inside their __init__
   • Updated wolf-sheep model to follow this pattern

3. Fixed ABMSimulator timing:

   • Changed initial step scheduling to use schedule_event_next_tick() instead of schedule_event_now()
   • Ensures first step occurs at t=1 in line with model: Automatically increase steps counter #2223

4. Added run_next_event() method to support step-by-step simulation execution

Usage Examples

Example of proper model initialization with DEVS:

class MyModel(Model):
    def __init__(self, simulator, seed=None):
        super().__init__(seed=seed)
        self.simulator = simulator
        self.simulator.setup(self)  # Must be called before any event scheduling
        
        # Now safe to schedule events
        self.schedule_initial_events()

Additional Notes

• This PR isolates DEVS-related changes from Support simulators in SolaraViz #2470
• The changes are backward-incompatible for any code that relied on the old setup behavior
• All existing tests have been updated and new tests added to verify the fixes
• This affects experimental features only, no stable Mesa features are impacted

[github-actions]

Performance benchmarks:

Model	Size	Init time [95% CI]	Run time [95% CI]
BoltzmannWealth	small	🔵 -0.5% [-0.9%, -0.1%]	🔵 +0.2% [+0.1%, +0.3%]
BoltzmannWealth	large	🔵 +0.2% [-0.3%, +0.6%]	🔵 +1.0% [+0.6%, +1.4%]
Schelling	small	🔵 -0.0% [-0.2%, +0.2%]	🔵 +0.6% [+0.5%, +0.8%]
Schelling	large	🔵 -0.3% [-0.6%, +0.0%]	🔵 +0.3% [-0.1%, +0.7%]
WolfSheep	small	🔵 +0.7% [+0.3%, +1.1%]	🔴 +78.2% [+75.4%, +80.8%]
WolfSheep	large	🔵 -1.3% [-1.6%, -1.0%]	🔴 +55.1% [+53.0%, +57.2%]
BoidFlockers	small	🔵 +1.5% [+0.9%, +2.0%]	🔵 +2.7% [+1.9%, +3.6%]
BoidFlockers	large	🔵 +0.9% [+0.5%, +1.4%]	🔵 +0.9% [+0.3%, +1.5%]

[EwoutH]

Thanks, really appreciated. Generally, it looks good. Let's wait for the benchmarks to confirm performance.

Unrelated, but to confirm, the best practice is currently to define the simulator.setup(model) insode the init of the Model class, so self.simulator.setup(self)?

[github-actions]

Performance benchmarks:

Model	Size	Init time [95% CI]	Run time [95% CI]
BoltzmannWealth	small	🔵 +1.2% [+0.3%, +2.0%]	🔵 +1.7% [+1.5%, +1.9%]
BoltzmannWealth	large	🔵 +2.0% [+1.4%, +2.7%]	🔵 +0.7% [+0.1%, +1.3%]
Schelling	small	🔵 +3.3% [+2.9%, +3.6%]	🔵 +1.7% [+1.5%, +1.9%]
Schelling	large	🔵 +2.9% [+2.5%, +3.3%]	🔵 +0.9% [+0.2%, +1.5%]
WolfSheep	small	🔵 +0.9% [+0.5%, +1.3%]	🔴 +74.6% [+71.8%, +77.2%]
WolfSheep	large	🔵 +0.4% [-0.1%, +0.9%]	🔴 +52.6% [+49.3%, +55.9%]
BoidFlockers	small	🔵 +0.4% [-0.2%, +0.9%]	🔵 -2.2% [-2.8%, -1.6%]
BoidFlockers	large	🔵 -0.2% [-1.0%, +0.9%]	🔵 -2.3% [-3.0%, -1.5%]

[EwoutH]

Unfortunately, the performance degradation looks reproducible and significant.

Any idea what causing it?

[quaquel]

Unrelated, but to confirm, the best practice is currently to define the simulator.setup(model) insode the init of the Model class, so self.simulator.setup(self)?

Yes. If setup is called after events have been scheduled, you get an exception. So if __init__ results in events being scheduled, you have to do it in model.__init__. I actually think I like this over the old situation.

[quaquel]

Unfortunately, the performance degradation looks reproducible and significant.

Any idea what causing it?

yes: inccorrect behavior before. What happend was the following

1. grass schedules regrowth events if not fully grown
2. simulator.setup is called, canceling all events
3. Only full grown grass is available to eat and to reschedule itself
4. sheep die very quickly (first 10 turns or so)
5. wolfs die shortly after

Once I had the animation in #2470, the error was obvious.

[EwoutH]

Good to know the root cause!

Also shows how valuable visual debugging can be.

[quaquel]

Also shows how valuable visual debugging can be.

absolutely