from __future__ import annotations

import argparse
from datetime import datetime
import sys
from pathlib import Path

ROOT = Path(__file__).resolve().parents[1]
sys.path.insert(0, str(ROOT / "src"))

from stable_baselines3 import DQN, PPO, SAC

from solarchain_eval.agent.llm_client import make_llm_client
from solarchain_eval.agent.wrappers import AgenticConfig
from solarchain_eval.config import load_config
from solarchain_eval.evaluate import SB3Policy, evaluate_policies
from solarchain_eval.policies import make_builtin_policy
from solarchain_eval.run_metadata import write_run_metadata


def main() -> None:
    parser = argparse.ArgumentParser(description="Evaluate SolarChain-Eval policies")
    parser.add_argument("--config", default="configs/default.yaml")
    parser.add_argument("--policies", default="static,random,myopic")
    parser.add_argument("--episodes", type=int, default=None)
    parser.add_argument("--output-dir", default=None)
    parser.add_argument("--ppo-model", default=None)
    parser.add_argument("--sac-model", default=None)
    parser.add_argument("--dqn-model", default=None)
    parser.add_argument("--run-name", default="eval")
    parser.add_argument("--no-timestamp", action="store_true")
    parser.add_argument("--no-physics-penalty", action="store_true")
    parser.add_argument("--agentic-mode", choices=["none", "planner", "planner_auditor"], default="none")
    parser.add_argument("--planner", choices=["none", "rule", "llm"], default="none")
    parser.add_argument("--auditor", choices=["none", "rule", "llm"], default="none")
    parser.add_argument("--audit-trigger", choices=["event", "always"], default="event")
    parser.add_argument("--llm-provider", default=None)
    parser.add_argument("--llm-model", default=None)
    parser.add_argument("--llm-base-url", default=None)
    parser.add_argument("--save-agentic-logs", action="store_true")
    parser.add_argument("--no-progress", action="store_true")
    args = parser.parse_args()

    config = load_config(args.config)
    if args.no_physics_penalty:
        config.no_physics_penalty = True
    policies = []
    for name in [item.strip() for item in args.policies.split(",") if item.strip()]:
        if name == "ppo":
            config.action_mode = "continuous"
            policies.append(SB3Policy(PPO.load(args.ppo_model), "ppo"))
        elif name == "sac":
            config.action_mode = "continuous"
            policies.append(SB3Policy(SAC.load(args.sac_model), "sac"))
        elif name == "dqn":
            config.action_mode = "discrete"
            policies.append(SB3Policy(DQN.load(args.dqn_model), "dqn"))
        else:
            config.action_mode = "continuous"
            policies.append(make_builtin_policy(name, config))

    base_output = Path(args.output_dir or config.output_dir)
    if args.no_timestamp:
        output_dir = base_output / args.run_name
    else:
        stamp = datetime.now().strftime("%Y%m%d_%H%M%S")
        output_dir = base_output / f"{stamp}_{args.run_name}"
    episodes = args.episodes or config.evaluation.episodes
    agentic_config = AgenticConfig(
        agentic_mode=args.agentic_mode,
        planner=args.planner,
        auditor=args.auditor,
        audit_trigger=args.audit_trigger,
        save_agentic_logs=args.save_agentic_logs,
    )
    llm_client = None
    if args.agentic_mode != "none" and (args.planner == "llm" or args.auditor == "llm"):
        llm_client = make_llm_client(
            provider=args.llm_provider,
            model=args.llm_model,
            base_url=args.llm_base_url,
        )
    evaluate_policies(
        policies,
        config,
        episodes,
        output_dir,
        agentic_config=agentic_config if args.agentic_mode != "none" else None,
        llm_client=llm_client,
        show_progress=not args.no_progress,
    )
    write_run_metadata(
        output_dir,
        run_type="evaluate",
        args={**vars(args), "resolved_episodes": episodes, "output_dir": str(output_dir)},
        config=config,
        extra={
            "policies": [getattr(policy, "name", "policy") for policy in policies],
            "agentic": agentic_config.__dict__,
        },
    )
    print(f"Wrote evaluation outputs to {output_dir}")


if __name__ == "__main__":
    main()