Summary

Replace the generic AttributeDict helper class with a dedicated frozen dataclass Message for improved type safety and developer experience when working with Mesa signals.

Bug / Issue

The current signal implementation uses AttributeDict, a generic dictionary with attribute-like access. This has several limitations:

• No IDE autocomplete support when writing signal handlers
• No type safety - typos in field names fail at runtime
• Less clear intent - the structure of signal data is implicit
• Inconsistent field access patterns

Part of #3131 (addresses item 1)

Implementation

Replaced AttributeDict with a frozen dataclass Message in the mesa_signals module:

1. New Message dataclass in signals_util.py with explicit fields: name, old, new, owner, signal_type, additional_kwargs
2. Updated type hints throughout mesa_signal.py to accept str | SignalType | All where appropriate
3. Changed signal.type → signal.signal_type (more explicit, avoids Python keyword)
4. Additional kwargs now stored in additional_kwargs dict rather than spread into the signal object

Local benchmarking shows no measurable performance difference.

Testing

All existing tests pass with updated field names. Type checking and IDE autocomplete now work correctly for signal handlers:

def on_change(signal: Message):  # Type hint enables autocomplete
    assert signal.signal_type == SignalType.CHANGE
    assert signal.old == 10

Additional Notes

This is a breaking change for experimental code:

• signal.type → signal.signal_type
• Custom kwargs now in signal.additional_kwargs dict

Since mesa_signals is experimental, breaking changes are allowed per Mesa's policy.