"""Drawer close-up with pyrender: cabinet mesh + VAE/SIN drawer trajectories."""
import os
import sys

sys.path.insert(0, "/home/zhanghao/scene/DragMesh")
os.environ["PYOPENGL_PLATFORM"] = "egl"

import json
import numpy as np
import trimesh
import pyrender
import imageio

INPUT_DIR = "/home/zhanghao/scene/integration/dragmesh_input"
TRAJ_JSON = "/home/zhanghao/scene/integration/dragmesh_drawer_traj/cabinet_drawer_trajectory.json"
OUT_DIR = "/home/zhanghao/scene/integration/dragmesh_drawer_demo"
os.makedirs(OUT_DIR, exist_ok=True)

# Load cabinet mesh + labels (body=0, left_door=1, right_door=2, left_drawer=3, right_drawer=4)
mesh = trimesh.load(f"{INPUT_DIR}/cabinet_mesh.obj", process=False)
labels = np.load(f"{INPUT_DIR}/cabinet_labels.npy")
traj = json.load(open(TRAJ_JSON))
vae_trans = np.array(traj["translations"])
axis = np.array(traj["axis"])

# Drawer parts
drawer_mask = np.isin(labels, [3, 4])
body_mask = ~drawer_mask

print(f"Drawer verts: {drawer_mask.sum()}, Body verts: {body_mask.sum()}")

# Split mesh
faces = mesh.faces
drawer_faces = np.where(drawer_mask[faces].all(axis=1))[0]
body_faces = np.where(body_mask[faces].all(axis=1))[0]
# Mixed faces (drawer + body): assign to body
mixed = np.where(~(drawer_mask[faces].all(axis=1) | body_mask[faces].all(axis=1)))[0]
if len(mixed):
    # Check which part has majority
    for fid in mixed:
        v = faces[fid]
        if drawer_mask[v].mean() > 0.5:
            drawer_faces = np.append(drawer_faces, fid)
        else:
            body_faces = np.append(body_faces, fid)

body_mesh = mesh.submesh([body_faces], append=True) if len(body_faces) else None
drawer_mesh = mesh.submesh([drawer_faces], append=True) if len(drawer_faces) else None
print(f"Body faces: {len(body_faces)}, Drawer faces: {len(drawer_faces)}")

if body_mesh is not None:
    body_mesh.visual.face_colors = [160, 150, 140, 255]
if drawer_mesh is not None:
    drawer_mesh.visual.face_colors = [210, 170, 130, 255]

# Build trajectories
N = 60
vae_full = np.interp(np.linspace(0, len(vae_trans) - 1, N), np.arange(len(vae_trans)), vae_trans)
if np.abs(vae_full).max() > 0:
    vae_full = vae_full / np.abs(vae_full).max() * 0.25
vae_pp = np.concatenate([vae_full, vae_full[-2:0:-1]])

sin_full = 0.25 * np.sin(np.linspace(0, np.pi / 2, N))
sin_pp = np.concatenate([sin_full, sin_full[-2:0:-1]])

# PyRender scene
scene = pyrender.Scene(ambient_light=[0.4, 0.4, 0.45], bg_color=[250, 248, 245, 255])

# Floor
floor = trimesh.creation.box([2.5, 2.5, 0.02])
floor.apply_translation([0, 0, -0.01])
floor.visual.face_colors = [190, 175, 160, 255]
scene.add(pyrender.Mesh.from_trimesh(floor, smooth=False))

# Camera looking at cabinet front (drawer slides along -Y, camera at +Y)
cam = pyrender.PerspectiveCamera(yfov=np.pi / 3.0)
cam_pose = np.eye(4)
cam_pos = np.array([0.0, -1.1, 0.75])
cam_target = np.array([0.0, 0.0, 0.45])
cd = cam_target - cam_pos
cd = cd / np.linalg.norm(cd)
cr = np.cross(cd, [0, 0, 1])
cr = cr / np.linalg.norm(cr)
cu = np.cross(cr, cd)
cam_pose[:3, 0] = cr
cam_pose[:3, 1] = cu
cam_pose[:3, 2] = -cd
cam_pose[:3, 3] = cam_pos
scene.add(cam, pose=cam_pose)
scene.add(pyrender.DirectionalLight(color=[1, 1, 0.95], intensity=5.0), pose=cam_pose)

renderer = pyrender.OffscreenRenderer(800, 600)

for name, seq in [("vae", vae_pp), ("sin", sin_pp)]:
    frames = []
    body_node = scene.add(pyrender.Mesh.from_trimesh(body_mesh, smooth=False)) if body_mesh else None
    for d in seq:
        if drawer_mesh is not None:
            # Move drawer vertices along axis
            dm = drawer_mesh.copy()
            dm.vertices = drawer_mesh.vertices + axis * d
            dm.visual.face_colors = [210, 170, 130, 255]
            drawer_node = scene.add(pyrender.Mesh.from_trimesh(dm, smooth=False))
            color, _ = renderer.render(scene)
            frames.append(color)
            scene.remove_node(drawer_node)
    if body_node:
        scene.remove_node(body_node)
    imageio.mimsave(f"{OUT_DIR}/{name}_drawer.mp4", frames, fps=15)
    mid = len(frames) // 2
    imageio.imwrite(f"{OUT_DIR}/{name}_closed.png", frames[0])
    imageio.imwrite(f"{OUT_DIR}/{name}_open.png", frames[mid])
    motions = [np.abs(frames[i].astype(float) - frames[i - 1].astype(float)).mean() for i in range(1, len(frames))]
    print(f"{name}: peak motion {max(motions):.3f}, saved {len(frames)} frames")

renderer.delete()
print(f"\nOutputs in {OUT_DIR}")