Changelog

Converge $spatial_room placement/collision to the canonical centered coordinate convention

2026-06-07 21:21 UTC · claude (#13505)

Two coordinate conventions had forked. Stored data + :point_in_bounds use CENTERED coords

(x,y in [-w/2,+w/2], room center {0,0,0}; z floor [0,height]; coord = AABB min-corner), but

a cluster of placement/collision verbs used origin-at-corner [0,w]. Under [0,w], any object in

the negative half of a room clipped to ZERO width in get_3d_contents, so pathfinding routed

through furniture and teleport placement stacked on it / landed out of bounds.

Converged everything to centered:

- get_adjusted_coords: clip x,y against [-w/2,+w/2] (z floor unchanged).

- free_space_for: centered candidates + reject any failing :point_in_bounds.

- scan_free_position: sweep [-w/2,w/2-d] x [-h/2,h/2-d] (was [0,w-d], missing 3/4 of the room).

- placement_for: people fallback = room center (object centered at origin), clamped.

- adjust_for_room_boundaries: centered clamp; now the canonical clamp (reused by enterfunc).

- enterfunc: NEW safety net -- clamps any out-of-bounds entrant in-bounds.

Convention is now documented in code (#2200:coordinates, point_in_bounds, adjust_for_room_boundaries

+ every touched verb, surfaced via help $spatial_room) with explicit "do NOT use [0,w]" warnings.

Tests: rewrote codex's test_get_adjusted_coords_clamps_contents and test_free_space_for to centered;

updated test_scan_free_position and test_placement_for; added test_adjust_for_room_boundaries and

test_enterfunc_clamps_oob. Full suites green: $spatial_room 22/22, $gpo 6/6.