Fix AgentSet iteration fails on Python 3.14 with "dictionary changed size during iteration- Issue #2896

[madhavik-2005]

Summary

Fixes #2896

This PR fixes the GroupBy.__iter__ method that was causing test_agentset_groupby to fail with AttributeError: 'GroupBy' object has no attribute '_agents'.

Problem

The GroupBy.__iter__ method was incorrectly trying to access self._agents.keyrefs(), but GroupBy doesn't have an _agents attribute. The GroupBy class stores its data in self.groups (a dictionary), not _agents.

Solution

Changed GroupBy.__iter__ to correctly iterate over the groups dictionary:

Before:

def __iter__(self):
    for ref in self._agents.keyrefs():
        if (agent := ref()) is not None:
            yield agent

After:

def __iter__(self) -> Iterator[tuple[Hashable, list | AgentSet]]:
    """Iterate over (group_name, group) tuples."""
    return iter(self.groups.items())

This now correctly returns (group_name, group) tuples as expected by the test suite and matches how GroupBy is intended to be used.

Testing

Added tests/test_iteration_fix.py with a test that forces garbage collection during iteration to ensure robustness:

def test_agentset_iteration_with_gc():
    """Test that AgentSet iteration doesn't fail when GC runs during iteration."""
    model = Model()
    agents = [TestAgent(model) for _ in range(100)]
    agentset = AgentSet(agents, random=model.random)
    
    collected_ids = []
    for i, agent in enumerate(agentset):
        collected_ids.append(agent.unique_id)
        if i % 10 == 0:
            gc.collect()
    
    assert len(collected_ids) == 100

Test Results:

• ✅ New GC iteration test passes
• ✅ All 22 tests in test_agent.py pass (including the previously failing test_agentset_groupby)

Additional Context

This fix ensures that iterating over GroupBy objects works as intended and prevents the AttributeError that was blocking the test suite.

[github-actions]

Performance benchmarks:

Model	Size	Init time [95% CI]	Run time [95% CI]
BoltzmannWealth	small	🔵 +1.6% [+0.8%, +2.6%]	🔵 -1.2% [-1.6%, -0.9%]
BoltzmannWealth	large	🔵 -1.2% [-2.0%, -0.5%]	🔵 -2.4% [-3.9%, -1.2%]
Schelling	small	🔵 -0.9% [-1.2%, -0.6%]	🔵 -0.8% [-1.0%, -0.7%]
Schelling	large	🔵 -0.9% [-1.3%, -0.4%]	🔵 -3.3% [-4.3%, -2.4%]
WolfSheep	small	🔵 -2.7% [-3.0%, -2.4%]	🔵 -1.6% [-2.2%, -1.1%]
WolfSheep	large	🔵 -3.4% [-4.4%, -2.4%]	🔵 -3.7% [-5.8%, -1.6%]
BoidFlockers	small	🔵 -1.1% [-1.8%, -0.5%]	🔵 +1.9% [+1.6%, +2.1%]
BoidFlockers	large	🔵 -1.4% [-2.1%, -0.7%]	🔵 +0.0% [-0.9%, +1.0%]

[tpike3]

@madhavik-2005 Thanks!

• Could you please add the docstring you put n the PR """Iterate over (group_name, group) tuples.""" into the code. That should stop the pre-commit error.
• The readthedocs failure looks like its an issue with solara since it is failing on import of mesa.visualization. That one will take a little more digging.

[madhavik-2005]

Hi @tpike3,

I've fixed the docstring issue. I had accidentally removed part of the docstring that pre-commit had added. The full docstring is now back in place:

"""Iterate over (group_name, group) tuples.

Yields:
tuple[Hashable, list | AgentSet]: pairs of group name and group.
"""

The pre-commit checks should pass now. Please let me know if there are any other changes needed - I'm still learning the contribution workflow and appreciate your guidance!

[EwoutH]

I think I made a mistake, and filled this bug wrongly. I think it's actually not at our end, but Solara's: widgetti/solara#1107. See also #2843.

Sorry, appreciate the effort.

[madhavik-2005]

Thanks for the update! Since this issue is actually on Solara’s side, I’ll close this PR and follow the Solara issue (widgetti/solara#1107). Appreciate the guidance!