"""1080×1920 실제 카메라 투영으로 T5 주석 위치와 필요한 매 프레임 키를 내보낸다."""
import argparse
import hashlib
import json
from pathlib import Path
import sys
import bpy
import numpy as np
from mathutils import Vector
from bpy_extras.object_utils import world_to_camera_view
ROOT=Path(__file__).resolve().parents[3]
PROJECT=ROOT/'projects/water-hydrogen-bond-surface-tension'
parser=argparse.ArgumentParser()
parser.add_argument('--source',required=True)
parser.add_argument('--out',required=True)
args=parser.parse_args(sys.argv[sys.argv.index('--')+1:])
source=Path(args.source).resolve()
bpy.ops.wm.open_mainfile(filepath=str(source))
cuts=json.loads((PROJECT/'stages/07-cuts/r1/artifact.json').read_text(encoding='utf-8'))['payload']['cuts']
plans=json.loads((PROJECT/'stages/08-scenes/r3/artifact.json').read_text(encoding='utf-8'))['payload']['scenes']
cache={}


def bbox_objects(obj):
    return [obj] if obj.type=='MESH' else [child for child in obj.children if child.type=='MESH']


def coordinates(obj,deps):
    if obj.name not in cache:
        assert not obj.modifiers and not obj.data.shape_keys, '정점 캐시가 불가능한 변형 mesh'
        evaluated=obj.evaluated_get(deps)
        mesh=evaluated.to_mesh()
        cache[obj.name]=np.array([tuple(v.co) for v in mesh.vertices])
        evaluated.to_mesh_clear()
    matrix=np.array(obj.evaluated_get(deps).matrix_world)
    return cache[obj.name] @ matrix[:3,:3].T + matrix[:3,3]


result=[]
for cut,plan in zip(cuts,plans):
    scene=bpy.data.scenes[plan['start_state'].split(';')[0].split('=')[1]]
    bpy.context.window.scene=scene
    assert scene.camera.data.type=='ORTHO'
    origin=world_to_camera_view(scene,scene.camera,Vector((0,0,0)))
    projection=np.array([tuple(world_to_camera_view(scene,scene.camera,Vector(b))-origin)[:2] for b in [(1,0,0),(0,1,0),(0,0,1)]])
    offset=np.array(tuple(origin)[:2])
    objects=[obj for obj in scene.objects if obj.get('anchor')]
    start,end=cut['start_ms']//20,cut['end_ms']//20
    tracks={obj['anchor']:[] for obj in objects}
    states=[]
    for frame in range(start,end+1):
        scene.frame_set(frame+1)
        deps=bpy.context.evaluated_depsgraph_get()
        labels={}
        for obj in objects:
            center=np.array(tuple(obj.matrix_world.translation)) @ projection + offset
            target=[round(float(center[0]*1080),2),round(float((1-center[1])*1920),2)]
            parts=bbox_objects(obj)
            box=None
            if parts:
                vertices=np.vstack([coordinates(part,deps) for part in parts])
                projected=vertices @ projection + offset
                px=projected[:,0]*1080
                py=(1-projected[:,1])*1920
                box=[round(float(px.min()),2),round(float(py.min()),2),round(float(px.max()),2),round(float(py.max()),2)]
                label=[round((box[0]+box[2])/2,2),round(box[1]-32,2)]
            else:
                label=target
            assert 0<label[0]<1080 and 0<label[1]<1920, '주석 제안점 화면 밖'
            labels[obj['anchor']]=dict(object_name=obj.name,target_pixel=target,label_pixel=label,bbox_px=box)
            tracks[obj['anchor']].append([frame,*label,*target])
        if frame in [start,(start+end)//2,end]:
            states.append(dict(logical_frame=frame,blender_frame=frame+1,anchors=labels))
    motion=[]
    for label,values in tracks.items():
        changing=any(v[1:]!=values[0][1:] for v in values[1:])
        keys=[]
        for index,value in enumerate(values):
            if index in [0,len(values)-1] or value[1:]!=values[index-1][1:]:
                keys.append(value)
        motion.append(dict(anchor_id=label,animated=changing,columns=['logical_frame','label_x','label_y','target_x','target_y'],keys=keys if changing else [values[0],values[-1]]))
    result.append(dict(cut_id=cut['id'],scene_name=scene.name,camera_name=scene.camera.name,start_frame=start,end_frame=end,states=states,motion=motion))
value=dict(source_path=source.relative_to(ROOT).as_posix(),source_sha256=hashlib.sha256(source.read_bytes()).hexdigest(),width=1080,height=1920,coordinate_system='top_left_pixel',logical_to_blender='logical_frame + 1',label_note='label_pixel은 실제 mesh 위 32px의 배치 시작점이다. 합성 단계에서 글자 크기·충돌·자막 여백을 실제 합성 화면으로 검수한다. bbox_px와 target_pixel은 실제 카메라 투영이다.',cuts=result)
Path(args.out).write_text(json.dumps(value,ensure_ascii=False,separators=(',',':'))+'\n',encoding='utf-8')
print('실제 카메라 주석 앵커 GREEN',len(result))