"""Gradio frontend for the Küchenpass-Agent.

Run locally:
    python -m space.app

Deployment: this file is uploaded as-is to a HuggingFace Space (see the
metadata header in space/README.md).
"""

from __future__ import annotations

import json
import os
import sys
from datetime import datetime
from pathlib import Path

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

from src.runtime import configure_runtime, ensure_model_artifacts  # noqa: E402

configure_runtime()
ensure_model_artifacts(
    required_files=(
        "prep_time_pipeline.joblib",
        "food_classifier.pth",
    )
)

# Avoid noisy/proxy-sensitive Gradio analytics requests during local demos.
os.environ.setdefault("GRADIO_ANALYTICS_ENABLED", "False")

import gradio as gr
from PIL import Image

from src.pipeline.orchestrator import KuechenpassPipeline  # noqa: E402

PIPELINE = KuechenpassPipeline()

EXAMPLE_ORDERS = [
    "Two cheeseburgers medium-rare and a side of fries please",
    "Eine Pizza Margherita ohne Knoblauch und einen Caesar Salad",
    "drei spaghetti bolognese extra parmesan",
    "I'd like the steak well done and a tiramisu for dessert",
    "ramen mit extra ei, dazu sushi mix 8 stück",
]


def _format_ticket(ticket: dict) -> str:
    items = ticket.get("items", [])
    if not items:
        return "_(keine Items erkannt)_"
    lines = ["| Menge | Gericht | Station | Modifier |", "|---|---|---|---|"]
    for it in items:
        mods = ", ".join(it.get("modifiers", []) or []) or "-"
        lines.append(
            f"| {it.get('quantity', 1)} | {it.get('dish','?')} | {it.get('station','?')} | {mods} |"
        )
    return "\n".join(lines)


def _format_prep_times(items: list[dict], total: float | None) -> str:
    if not items:
        return "_ML-Modell nicht verfügbar — Pipeline läuft im Demo-Modus._"
    lines = ["| Gericht | Menge | Station | Geschätzte Zeit |", "|---|---|---|---|"]
    for p in items:
        lines.append(
            f"| {p['dish']} | {p['quantity']} | {p['station']} | {p['minutes']:.1f} min |"
        )
    if total is not None:
        lines.append(f"\n**Gesamt-ETA (parallelisiert):** {total:.1f} min")
    return "\n".join(lines)


def _format_cv(preds: list[dict] | None) -> str:
    if not preds:
        return "_(kein Foto hochgeladen)_"
    lines = ["| Rang | Label | Confidence |", "|---|---|---|"]
    for i, p in enumerate(preds, 1):
        lines.append(f"| {i} | {p['label']} | {p['confidence']*100:.1f} % |")
    return "\n".join(lines)


def _format_decision(decision: dict | None) -> str:
    if not decision:
        return "_(noch keine Pass-Entscheidung)_"
    verdict = decision["verdict"]
    emoji = "✅" if verdict == "to_guest" else "🚫"
    label = "KANN ZUM GAST" if verdict == "to_guest" else "GANG NEU MACHEN"
    return (
        f"### {emoji} **{label}**\n\n"
        f"- Grund: {decision['reason']}\n"
        f"- CV-Label: `{decision['predicted_label']}` ({decision['confidence']*100:.1f} %)\n"
        f"- Match: `{decision.get('matched_item') or '—'}`"
    )


def run_pipeline(
    order_text: str,
    kitchen_load: int,
    photo: Image.Image | None,
    prompt_version: str,
) -> tuple[str, str, str, str, str]:
    result = PIPELINE.run(
        order_text=order_text or "",
        kitchen_load=int(kitchen_load),
        photo=photo,
        prompt_version=prompt_version,
        order_datetime=datetime.now(),
    )
    ticket_md = _format_ticket(result.ticket)
    prep_md = _format_prep_times(
        result.estimated_minutes_per_item, result.estimated_minutes_total
    )
    cv_md = _format_cv(result.cv_predictions)
    decision_md = _format_decision(result.pass_decision)
    raw_json = json.dumps(
        {
            "ticket": result.ticket,
            "estimated_minutes_per_item": result.estimated_minutes_per_item,
            "estimated_minutes_total": result.estimated_minutes_total,
            "cv_predictions": result.cv_predictions,
            "pass_decision": result.pass_decision,
        },
        indent=2,
        ensure_ascii=False,
    )
    return ticket_md, prep_md, cv_md, decision_md, raw_json


def build_ui() -> gr.Blocks:
    with gr.Blocks(title="Küchenpass-Agent", theme=gr.themes.Soft()) as demo:
        gr.Markdown(
            "# Küchenpass-Agent\n"
            "Free-Text-Bestellung → strukturiertes Ticket (NLP) → ETA-Prognose (ML)"
            " → Pass-Prüfung anhand Foto (CV).\n"
            "_Semesterprojekt ZHAW · KI Anwendungen FS 2026._"
        )
        with gr.Row():
            with gr.Column(scale=1):
                order_text = gr.Textbox(
                    label="Bestellung (Free Text, DE/EN)",
                    lines=4,
                    placeholder="z.B. Zwei Burger medium und eine Pizza Margherita",
                )
                kitchen_load = gr.Slider(
                    minimum=0,
                    maximum=40,
                    step=1,
                    value=5,
                    label="Aktuelle Küchen-Auslastung (offene Bestellungen)",
                )
                prompt_version = gr.Radio(
                    choices=["v1_basic", "v2_fewshot", "v3_constrained"],
                    value="v3_constrained",
                    label="Prompt-Strategie",
                )
                photo = gr.Image(
                    label="Foto vom fertigen Gang (optional)",
                    type="pil",
                    height=240,
                )
                submit = gr.Button("An den Pass schicken", variant="primary")
                gr.Examples(
                    examples=[[x] for x in EXAMPLE_ORDERS],
                    inputs=[order_text],
                    label="Beispiel-Bestellungen",
                    cache_examples=False,
                )

            with gr.Column(scale=2):
                gr.Markdown("### 1. Geparstes Ticket (NLP)")
                ticket_out = gr.Markdown()
                gr.Markdown("### 2. Geschätzte Zubereitungszeiten (ML)")
                prep_out = gr.Markdown()
                gr.Markdown("### 3. CV-Predictions am Pass")
                cv_out = gr.Markdown()
                gr.Markdown("### 4. Pass-Entscheid")
                decision_out = gr.Markdown()
                with gr.Accordion("Rohdaten (JSON)", open=False):
                    raw_out = gr.Code(language="json")

        submit.click(
            fn=run_pipeline,
            inputs=[order_text, kitchen_load, photo, prompt_version],
            outputs=[ticket_out, prep_out, cv_out, decision_out, raw_out],
        )

    return demo


def main() -> None:
    demo = build_ui()
    demo.launch()


if __name__ == "__main__":
    main()