fix(hex_ant): fix relative imports and PropertyLayer API to make example fully runnable

[dietcoke-17]

Summary

Closes #357 — hex_ant was completely non-functional. Running solara run app.py
crashed immediately and the model never reached the browser. This PR works through
every crash in sequence until the example fully runs.

Bug / Issue

Bug 1 — Relative imports in app.py and model.py

app.py — before
from .agent import AntState
from .model import AntForaging
removed the . before the module

why this change
Relative imports (.module) only work when Python treats the directory as a package with a known parent. When Solara executes app.py directly, __package__ is None, so there is no parent to be relative to

Result: Model loads past the import stage.

Bug 2 — Direct PropertyLayer item assignment in _init_environment
model.py — before
self.grid.pheromone_home[center] = 1.0
self.grid.home[center] = 1

model.py — after
self.grid.pheromone_home.data[center] = 1.0
self.grid.home.data[center] = 1

Why
In Mesa 3.x, PropertyLayer is a wrapper object, not a numpy array. Direct item assignment (layer[x, y] = val) is not supported. The underlying numpy array is accessed through the .data attribute, which does support index assignment.

Result: Model initialises, nest and food sources are placed,
visualization opens in browser.

Bug 3 — Wrong dynamic layer lookup in _update_pheromone_layer
model.py — before
np_layer = self.grid.property_layers[layer_name]

model.py — after
np_layer = getattr(self.grid, layer_name).data

Why:

self.grid.property_layers does not exist as an attribute. HexGrid overrides __getattr__ and routes ALL attribute access through the internal property layer dictionary — so accessing .property_layers is itself treated as a layer name lookup and fails. The correct pattern for dynamic layer access by name is getattr(self.grid, layer_name). The .data is also needed here for the same reason as Bug 2 — numpy operations like *= and boolean masking must be performed on the raw
array, not the wrapper object.

Testing

Additional Notes

Relation to existing PRs

• Fix hex_ant and hex_snowflake: migrate to Mesa 3.x visualization API #377 and Fix shape_example: migrate from ModularServer to SolaraViz (Mesa 3.x) #380 — attempted import fixes only, both closed
• fix: resolve relative import errors in emperor_dilemma, hex_ant, hex_… #427 — correctly fixes relative imports; the copy of the code worked on here already had clean absolute imports, so that was not the blocker. The PropertyLayer bugs are entirely separate and were not addressed by any existing PR.
• Bug 3 specifically would have crashed the model on step 1 even if every other fix had been applied first — it was the last hidden blocker.

Verified

Tested on Windows, Mesa 3.5.0, Python 3.14:

• solara run app.py starts without error
• Visualization opens at http://localhost:8765
• Ants spawn at nest and begin foraging
• Pheromone trails build up and evaporate each step
• Returning ants (gold) follow home pheromone correctly
• Food clusters deplete as ants pick up food

After Codex Suggestion
In model.py and app.py: used try/except import pattern to support both solara run app.py (direct execution) and package import via test runner and it worked

Also CI Failing Issue
Also fixed a pre-existing lint error in rl/wolf_sheep/agents.py — two variables x, y were unpacked but never actually used anywhere. Renamed them to _x, _y to tell the linter they are intentionally unused. This was flagged by pre-commit CI and is unrelated to the hex_ant fix.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: fa7a20f19e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".