"""End-to-end Genesis/dexmachina scene demo.

Runs inside the `benchmarking/dexmachina:latest` container. Patches the
installed `dexmachina` package with the missing `envs/hand_cfgs/` submodule
(copied to `outputs/_runtime_patches` on the host, volume-mounted at
`/workspace/outputs/_runtime_patches`), verifies imports, builds a small
Genesis scene (ground plane + xhand URDF), renders 60 frames and muxes them
into an mp4 via `shared.bench.VideoRecorder`.

All output artefacts are written under `/workspace/outputs/`.
"""
from __future__ import annotations

import json
import os
import shutil
import sys
import time
import traceback
from pathlib import Path

OUT_ROOT = Path("/workspace/outputs")
LOG_PATH = OUT_ROOT / "logs" / "dexmachina_scene_demo.log"
MP4_PATH = OUT_ROOT / "videos" / "dexmachina_xhand_demo.mp4"
LOG_PATH.parent.mkdir(parents=True, exist_ok=True)
MP4_PATH.parent.mkdir(parents=True, exist_ok=True)

log_buf = []

def log(msg: str) -> None:
    print(msg, flush=True)
    log_buf.append(str(msg))


def flush_log() -> None:
    LOG_PATH.write_text("\n".join(log_buf) + "\n")


# ---- 1. runtime-patch dexmachina.envs.hand_cfgs (missing in installed wheel)
def patch_hand_cfgs() -> None:
    import dexmachina
    dst = Path(dexmachina.__file__).parent / "envs" / "hand_cfgs"
    src = Path("/workspace/outputs/_runtime_patches/dexmachina/envs/hand_cfgs")
    if not dst.exists():
        if not src.exists():
            raise RuntimeError(f"hand_cfgs staged source missing: {src}")
        shutil.copytree(src, dst)
        log(f"patched: copied {src} -> {dst}")
    else:
        log(f"hand_cfgs already present at {dst}")


# ---- 2. import graph
def do_imports() -> dict:
    import importlib.metadata as im
    import torch, genesis, rl_games, dexmachina  # noqa: F401
    from dexmachina.envs.hand_cfgs.xhand import XHAND_CFGs
    from dexmachina.envs import base_env

    info = {
        "torch": torch.__version__,
        "genesis": getattr(genesis, "__version__", "ok"),
        "rl_games": im.version("rl_games"),
        "dexmachina": im.version("dexmachina"),
        "cuda_available": bool(torch.cuda.is_available()),
        "BaseEnv": base_env.BaseEnv.__name__,
        "XHAND_CFGs_keys": list(XHAND_CFGs.keys()),
    }
    return info


# ---- 3. scene demo
def run_scene_demo() -> dict:
    import numpy as np
    import torch
    import genesis as gs
    sys.path.insert(0, "/workspace/shared")
    from bench.video_recorder import VideoRecorder

    backend = gs.cuda if torch.cuda.is_available() else gs.cpu
    log(f"genesis.init backend={backend}")
    gs.init(backend=backend, logging_level="warning")

    scene = gs.Scene(show_viewer=False)

    # ground plane
    scene.add_entity(gs.morphs.Plane())

    # xhand URDF (right hand is the standard variant)
    # Preference order: staging-backed /workspace/assets symlinks (but in this
    # image those aren't mounted, so the symlink target is missing), then the
    # dexmachina-shipped URDF. We try both so future mount changes "just work".
    candidates = [
        "/workspace/assets/urdf/xhand/xhand_urdf/xhand_right/urdf/xhand_right.urdf",
        "/usr/local/lib/python3.10/dist-packages/dexmachina/assets/xhand/xhand_right.urdf",
    ]
    urdf_path = next((p for p in candidates if Path(p).exists()), None)
    assert urdf_path is not None, f"no xhand urdf found among {candidates}"
    log(f"urdf_path={urdf_path}")
    hand = scene.add_entity(
        gs.morphs.URDF(
            file=urdf_path,
            pos=(0.0, 0.0, 0.25),
            fixed=True,
        )
    )

    W, H = 640, 480
    camera = scene.add_camera(
        res=(W, H),
        pos=(0.6, 0.6, 0.45),
        lookat=(0.0, 0.0, 0.2),
        fov=45,
        GUI=False,
    )

    scene.build()
    log(f"scene built; hand entity: {hand}")

    rec = VideoRecorder(MP4_PATH, fps=30, size=(W, H))
    n_frames = 60
    t0 = time.time()
    for i in range(n_frames):
        scene.step()
        out = camera.render()
        # Genesis camera.render returns a tuple (rgb, depth, seg, normal)
        rgb = out[0] if isinstance(out, tuple) else out
        if hasattr(rgb, "cpu"):
            rgb = rgb.cpu().numpy()
        if rgb.dtype != np.uint8:
            rgb = rgb.astype(np.uint8)
        if rgb.shape[-1] == 4:
            rgb = rgb[..., :3]
        rec.log_frame(np.ascontiguousarray(rgb))
    rec.close()
    dt = time.time() - t0

    size_bytes = MP4_PATH.stat().st_size
    log(f"rendered {n_frames} frames {W}x{H} in {dt:.2f}s; mp4={MP4_PATH} ({size_bytes} bytes)")
    return {
        "mp4_path": str(MP4_PATH),
        "frames": n_frames,
        "resolution": [W, H],
        "mp4_size_bytes": size_bytes,
        "backend": str(backend),
        "elapsed_s": round(dt, 3),
    }


def main() -> int:
    result = {"ok": True, "steps": {}}
    try:
        patch_hand_cfgs()
        imports = do_imports()
        result["steps"]["imports"] = imports
        log("IMPORTS: " + json.dumps(imports))
    except Exception as e:
        result["ok"] = False
        result["steps"]["imports_error"] = repr(e)
        log("IMPORT ERROR: " + repr(e))
        log(traceback.format_exc())
        flush_log()
        (OUT_ROOT / "logs" / "dexmachina_scene_demo.json").write_text(json.dumps(result, indent=2))
        return 2

    try:
        scene_info = run_scene_demo()
        result["steps"]["scene"] = scene_info
        log("SCENE: " + json.dumps(scene_info))
    except Exception as e:
        result["ok"] = False
        result["steps"]["scene_error"] = repr(e)
        log("SCENE ERROR: " + repr(e))
        log(traceback.format_exc())
        flush_log()
        (OUT_ROOT / "logs" / "dexmachina_scene_demo.json").write_text(json.dumps(result, indent=2))
        return 3

    flush_log()
    (OUT_ROOT / "logs" / "dexmachina_scene_demo.json").write_text(json.dumps(result, indent=2))
    log("DONE")
    return 0


if __name__ == "__main__":
    sys.exit(main())