Skip to content

Add missing mesa-internal attributes to mesa_primitives blocklist in create_meta_agent#3538

Open
codebyNJ wants to merge 8 commits intomesa:mainfrom
codebyNJ:fix/meta-agent-mesa-primitives-blocklist
Open

Add missing mesa-internal attributes to mesa_primitives blocklist in create_meta_agent#3538
codebyNJ wants to merge 8 commits intomesa:mainfrom
codebyNJ:fix/meta-agent-mesa-primitives-blocklist

Conversation

@codebyNJ
Copy link
Copy Markdown
Contributor

@codebyNJ codebyNJ commented Mar 14, 2026 •
edited
Loading

Pre-PR Checklist

  • This PR is a bug fix, not a new feature or enhancement.

Summary

Fixes add_attributes in create_meta_agent so existing meta-agent attributes are not overwritten by values inferred from constituting agents. Uses hasattr() checks to preserve existing attributes—both internal Mesa attributes and explicitly-provided ones.

Bug / Issue

When assume_constituting_agent_attributes=True, the previous implementation unconditionally copied all public non-callable attributes from constituent agents onto the meta-agent, silently overwriting:

  1. Mesa internal attributes (cell, current_action, meta_agent, meta_agents)
  2. Explicitly-passed meta_attributes (as reported in preserve explicit meta attributes/methods over inferred members #3472)

Expected behavior: don't overwrite attributes that already exist on the meta-agent.
Actual behavior: any existing attribute could be silently overwritten without warning.

Related context: issues #3433, #3506, #3472 all identified meta-agent attribute problems.

Implementation

Updated add_attributes helper inside create_meta_agent in mesa/experimental/meta_agents/meta_agent.py:

  • When building meta_attributes from agents: Check not hasattr(meta_agent_instance, name) before copying
  • When applying meta_attributes to meta-agent: Check if not hasattr(meta_agent_instance, key) before calling setattr()

Why this is better than a blocklist:

Testing

Added test_assume_attributes_does_not_overwrite_existing_attributes:

  • Verifies explicit meta_attributes are preserved when adding agents with assume_constituting_agent_attributes=True
  • Verifies missing attributes are still copied from constituents

Additional Notes

This change keeps behavior predictable: existing attributes are preserved, inferred values only fill missing keys.

No API changes are required for current users.

This complements determinism-related fixes (#3433, #3506) and does not replace them.

@github-actions
Copy link
Copy Markdown

github-actions bot commented Mar 14, 2026

Performance benchmarks:

Model Size Init time [95% CI] Run time [95% CI]
BoltzmannWealth small 🔵 -1.0% [-1.4%, -0.7%] 🔵 -0.1% [-0.3%, +0.0%]
BoltzmannWealth large 🔵 -0.3% [-0.8%, +0.0%] 🔵 -0.9% [-1.7%, -0.2%]
Schelling small 🔵 +0.3% [-0.2%, +0.8%] 🔵 +0.3% [-0.4%, +1.0%]
Schelling large 🔵 -1.4% [-2.0%, -0.8%] 🔵 -2.2% [-3.2%, -1.1%]
WolfSheep small 🔵 +0.1% [-0.5%, +0.7%] 🔵 +2.3% [-1.5%, +6.1%]
WolfSheep large 🔵 +0.1% [-0.8%, +1.2%] 🔵 +1.8% [+0.7%, +2.9%]
SugarscapeG1mt small 🔵 +0.1% [-0.3%, +0.5%] 🔵 +2.7% [+1.1%, +4.4%]
SugarscapeG1mt large 🔵 +0.9% [-0.5%, +2.4%] 🔵 +1.3% [-0.9%, +3.4%]
BoidFlockers small 🔵 +1.3% [+0.6%, +2.1%] 🔵 +2.0% [+1.6%, +2.5%]
BoidFlockers large 🔵 +2.7% [+1.8%, +3.9%] 🔵 +2.0% [+1.2%, +3.1%]

@codebyNJ codebyNJ mentioned this pull request Mar 16, 2026
Merged
6 tasks
@tpike3 tpike3 self-assigned this Mar 18, 2026
@tpike3
Copy link
Copy Markdown
Member

tpike3 commented Mar 18, 2026

As this is clearly a bug and there are some related issues as identified in #3472, instead of both #3538 and #3472 why dont we just ensure the meta-agent doesn't overwrite attributes it already has.

This approach

 mesa_primitives = [
            "unique_id",
            "model",
            "pos",
            "name",
            "random",
            "rng",
            "cell",
            "current_action",
            "meta_agent",
            "meta_agents",
        ]

is very brittle, and so issues identified in #3472 and #3538, plus the current brittleness can all be solved with not overwriting existing meta-agent attributes.

@tpike3 tpike3 mentioned this pull request Mar 18, 2026
Merged
1 task
@falloficarus22
Copy link
Copy Markdown
Contributor

falloficarus22 commented Mar 18, 2026 •
edited
Loading

As this is clearly a bug and there are some related issues as identified in #3472, instead of both #3538 and #3472 why dont we just ensure the meta-agent doesn't overwrite attributes it already has.

This approach

 mesa_primitives = [
            "unique_id",
            "model",
            "pos",
            "name",
            "random",
            "rng",
            "cell",
            "current_action",
            "meta_agent",
            "meta_agents",
        ]

is very brittle, and so issues identified in #3472 and #3538, plus the current brittleness can all be solved with not overwriting existing meta-agent attributes.

This is a better direction for the attribute side. It seems more robust than extending the mesa_primitives blocklist, and it also looks like it would cover both the internal-attribute corruption in this PR and the explicit-attribute overwrite case in #3472. I do think methods are still a separate question, since inferred method-name collisions are not addressed by the same rule, but for attributes this feels like the cleaner approach.

@codebyNJ
Copy link
Copy Markdown
Contributor Author

codebyNJ commented Mar 19, 2026

@tpike3 @falloficarus22 I have updated the PR.

tpike3
tpike3 requested changes Mar 25, 2026
View reviewed changes
Copy link
Copy Markdown
Member

@tpike3 tpike3 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks @codebyNJ , this looks good to me, can you fix the test conflicts and then I will merge.

@codebyNJ
Copy link
Copy Markdown
Contributor Author

codebyNJ commented Mar 28, 2026

Thanks @codebyNJ , this looks good to me, can you fix the test conflicts and then I will merge.

Welcome @tpike3. I have fixed the merge conflict issues in the test files.

@tpike3
Copy link
Copy Markdown
Member

tpike3 commented Mar 29, 2026

@codebyNJ Could you look at the Warehouse Model failure. The warehouse model is the other meta-agent example so we need to understand why this code broke it. https://github.com/mesa/mesa-examples/tree/main/examples/warehouse

@codebyNJ
Copy link
Copy Markdown
Contributor Author

codebyNJ commented Mar 30, 2026

sure @tpike3 i will look into warehouse model and find out the reason why this code broke.

@codebyNJ
Copy link
Copy Markdown
Contributor Author

codebyNJ commented Mar 30, 2026

@tpike3 Fixed the warehouse failure. The issue: hasattr() check prevented explicit meta_attributes (like cell) from overriding class defaults.

Solution: Always apply explicit meta_attributes (user intent takes precedence).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@tpike3 tpike3 tpike3 approved these changes

Assignees

@tpike3 tpike3

Labels

GSoC: Not ready

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

4 participants

@codebyNJ @tpike3 @falloficarus22 @EwoutH