Generate clean 2D game sprites & animation atlases — component-row pipeline: state rows, alpha cleanup, frame extraction, runtime atlases. Codex/Claude skill.
test(parts): the synthetic walker is drawn shape by shape in a window round each shape — the same frames byte for byte, in a fraction of the time
Soohong Kimc1a99be5aef8
diff --git a/tests/video/test_parts.py b/tests/video/test_parts.pyindex df163ec..3433d25 100644--- a/tests/video/test_parts.py+++ b/tests/video/test_parts.py@@ -51,29 +51,47 @@ _yy, _xx = np.mgrid[0:H * SS, 0:W * SS].astype(np.float32)_yy, _xx = (_yy + 0.5) / SS, (_xx + 0.5) / SS-def _seg(p: tuple[float, float], q: tuple[float, float]) -> np.ndarray:+def _window(points, reach: float) -> tuple[slice, slice]:+ """The whole pixels (rows, columns) of the box round `points` grown by `reach` px and a pixel more: every point of+ the supersampled canvas within `reach` of a segment between them lies inside, so a shape of that radius is drawn+ there alone (the frames are the same bytes as drawn over the whole canvas, in a fraction of the time)."""+ ys, xs = [pt[0] for pt in points], [pt[1] for pt in points]+ return (slice(max(0, math.floor(min(ys) - reach) - 1), min(H, math.ceil(max(ys) + reach) + 1)),+ slice(max(0, math.floor(min(xs) - reach) - 1), min(W, math.ceil(max(xs) + reach) + 1)))+++def _seg(p: tuple[float, float], q: tuple[float, float], win: tuple[slice, slice]) -> np.ndarray:+ rows, cols = win+ fine = slice(rows.start * SS, rows.stop * SS), slice(cols.start * SS, cols.stop * SS)+ yy, xx = _yy[fine], _xx[fine](py, px), (qy, qx) = p, qvy, vx = qy - py, qx - px- t = np.clip(((_yy - py) * vy + (_xx - px) * vx) / max(vy * vy + vx * vx, 1e-9), 0, 1)- return np.hypot(_yy - (py + t * vy), _xx - (px + t * vx))+ t = np.clip(((yy - py) * vy + (xx - px) * vx) / max(vy * vy + vx * vx, 1e-9), 0, 1)+ return np.hypot(yy - (py + t * vy), xx - (px + t * vx))def _cover(inside: np.ndarray) -> np.ndarray:- return inside.astype(np.float32).reshape(H, SS, W, SS).mean(axis=(1, 3))+ return inside.astype(np.float32).reshape(inside.shape[0] // SS, SS, inside.shape[1] // SS, SS).mean(axis=(1, 3))-def _shape(e: np.ndarray, rgb: np.ndarray | tuple, line: tuple | None = None) -> tuple[np.ndarray, np.ndarray]:- """A shape (its signed distance `e`, inside at or below 0) filled with `rgb`, outlined LINE px in `line` (a darker- shade of its colour where none is given): its coverage and colour per pixel."""- a = _cover(e <= 0)- fill = np.clip(_cover(e <= -LINE) / np.maximum(a, 1e-6), 0, 1)[..., None]+def _shape(e: np.ndarray, win: tuple[slice, slice], rgb: np.ndarray | tuple, line: tuple | None = None+ ) -> tuple[np.ndarray, np.ndarray]:+ """A shape (its signed distance `e` over the window `win`, inside at or below 0) filled with `rgb`, outlined LINE px+ in `line` (a darker shade of its colour where none is given): its coverage and colour per pixel of the frame, the+ outline's colour and no coverage outside the window."""+ a = np.zeros((H, W), np.float32)+ a[win] = _cover(e <= 0)+ fill = np.clip(_cover(e <= -LINE) / np.maximum(a[win], 1e-6), 0, 1)[..., None]rgb = np.broadcast_to(np.asarray(rgb, np.float32), (H, W, 3))- line = 0.35 * np.asarray(rgb if line is None else line, np.float32)- return a, fill * rgb + (1 - fill) * line+ line = np.broadcast_to(0.35 * np.asarray(rgb if line is None else line, np.float32), (H, W, 3))+ colour = np.array(line)+ colour[win] = fill * rgb[win] + (1 - fill) * line[win]+ return a, colourdef _capsule(p, q, r, rgb):- return _shape(_seg(p, q) - r, rgb)+ win = _window((p, q), r)+ return _shape(_seg(p, q, win) - r, win, rgb)def _tail(at: float, bob: float, walker: str) -> tuple[np.ndarray, np.ndarray]:@@ -84,17 +102,21 @@ def _tail(at: float, bob: float, walker: str) -> tuple[np.ndarray, np.ndarray]:angles = [shape["base"] + shape["sway"] * math.sin(phase)]for k in range(1, 5):angles.append(angles[-1] + 14 * math.sin(phase - 0.7 * k))- p = (shape["root"][0] + bob, shape["root"][1])- e, tip = np.full(_yy.shape, np.inf, np.float32), None- for k, (angle, r) in enumerate(zip(angles, shape["radii"])):- q = (p[0] - 13 * math.sin(math.radians(angle)), p[1] + 13 * math.cos(math.radians(angle)))- ek = _seg(p, q) - r+ joints = [(shape["root"][0] + bob, shape["root"][1])]+ for angle in angles:+ p = joints[-1]+ joints.append((p[0] - 13 * math.sin(math.radians(angle)), p[1] + 13 * math.cos(math.radians(angle))))+ win = _window(joints, max(shape["radii"]))+ e, tip = np.inf, None+ for k, r in enumerate(shape["radii"]):+ ek = _seg(joints[k], joints[k + 1], win) - rif k == 4:tip = ek < ee = np.minimum(e, ek)- p = qtipped = np.clip(_cover(tip & (e <= 0)) / np.maximum(_cover(e <= 0), 1e-6), 0, 1)[..., None]- return _shape(e, (1 - tipped) * np.array(ORANGE, np.float32) + tipped * np.array(CREAM, np.float32), (0.3, 0.2, 0.1))+ rgb = np.array(np.broadcast_to(np.array(ORANGE, np.float32), (H, W, 3)))+ rgb[win] = (1 - tipped) * np.array(ORANGE, np.float32) + tipped * np.array(CREAM, np.float32)+ return _shape(e, win, rgb, (0.3, 0.2, 0.1))def _bob(body: float) -> float:@@ -102,7 +124,8 @@ def _bob(body: float) -> float:def _arm(shoulder, hand, r, hand_r):- return _shape(np.minimum(_seg(shoulder, hand) - r, _seg(hand, hand) - hand_r), ORANGE)+ win = _window((shoulder, hand), max(r, hand_r))+ return _shape(np.minimum(_seg(shoulder, hand, win) - r, _seg(hand, hand, win) - hand_r), win, ORANGE)def _layers(body: float, tail: float | None, walker: str, tail_bob: float | None = None@@ -195,7 +218,8 @@ def _crossed(t: float) -> bool:"""Whether the near hand and the tail of the walker seen from the side, each drawn alone, overlap at time t."""a = -0.5 * math.sin(2 * math.pi * t / N)hand = (78 + _bob(t) + 52 * math.cos(a), 104 + 52 * math.sin(a))- return bool(((_cover(_seg(hand, hand) <= 6.5) > 0.5) & (_tail(t, _bob(t), "side")[0] > 0.5)).any())+ win = _window((hand, hand), 6.5)+ return bool(((_cover(_seg(hand, hand, win) <= 6.5) > 0.5) & (_tail(t, _bob(t), "side")[0][win] > 0.5)).any())def _bytes(frames) -> list[bytes]: