Known Limits

This page describes the state of the examined code, not fixed features. Regression tests added to the project make several defects visible.

Area	Current Limit	Practice to Adopt
Agent.decide()	Returns None after state execution	Explicitly use runStateProcessus()
Agent.copy()	Fails with a Tabletop; incorrect assignment of the tabletop copy	Explicitly build a new agent
Agent.perceive() / HackaAgent	Features not implemented	Define perception and transport in the application
Entity.setShapeArrowTip()	Incorrect call to the shape constructor	Use setReferenceShape(Convex().initArrowTip(...))
Entity.setOutline()	Does not reset the effective orientation attribute	Prefer a reference shape and setPose()
Default Land bank	List and models shared between instances	Provide a new list of new entities
Copies and collections	Some internal data remains shared or directly modifiable	Explicitly define object ownership

Tabletop Restoration

Verification of examples revealed that dataTreeCopy() fails on a tabletop containing an entity: Tile.setIndex() calls Entity.setArea(), a missing method. The serialization example is therefore limited to a tabletop without entities.

Movements and Orientations

A reading of tileOrientEntity() and tileRotateEntity() shows that they select the first entity of the tile instead of passing the requested index. tileClockMoveEntity() also performs a rotation call with the target identifier after the movement. These points must be corrected and tested before relying on these methods in scenes with multiple bodies. Tutorials use direct movement and transformations on the entity itself.

Scale and Robustness

connectAllClose() compares pairs of tiles, which represents a quadratic cost in the number of tiles even before the cost of polygonal distances. Measure times on target maps before sizing an experiment.

Several validations rely on assert. The Python -O option disables them; it can even remove calls placed in an assertion, such as a tile creation in fromGridRectangles(). Do not use this mode for this version.

Tests

During the initial audit, before the local correction of perceivedBody(), the suite with the new tests showed: 137 successes, 8 failures, 7 tests excluded by the not long filter. The failures are defects expected by the audit but remain true failures: they are not masked by xfail. This historical balance is not the result of the corrected version. perceivedBody() now returns the body in the local code; rerun the suite to obtain an updated balance.