"""
HuggingFace Space entry point (Gradio SDK)
Mounts Gradio onto the existing FastAPI backend.

HF Gradio SDK pre-starts a server on port 7860 before executing app.py.
So instead of starting our own uvicorn (port conflict), we monkey-patch
blocks.launch to pass app=combined — injecting all FastAPI routes into
the server that HF already started.
"""
import os
import types
import gradio as gr

from app.main import app as fastapi_app

try:
    import spaces
    import torch

    @spaces.GPU
    def gpu_warmup():
        try:
            if torch.cuda.is_available():
                _ = torch.zeros(1, device="cuda")
                return "ZeroGPU CUDA Tensor Allocated"
            return "CUDA unavailable"
        except Exception as e:
            return f"ZeroGPU CUDA Warmup Error: {e}"
except Exception:
    def gpu_warmup():
        return "CPU Active"

# Run warmup immediately during container startup to satisfy the ZeroGPU supervisor check
try:
    gpu_warmup()
except Exception as e:
    print(f"Startup GPU Warmup failed (possibly transient HF hardware error): {e}")







# ── Blocks UI ─────────────────────────────────────────────────────────
# Gradio 6: title & theme moved from Blocks() constructor to launch()
with gr.Blocks(title="Othaim ALPR API", theme=gr.themes.Soft()) as blocks:
    gr.Markdown("""
    # 🅿️ Othaim Parking ALPR System

    Backend is running. All API endpoints are available at their normal paths.

    **Endpoints:**
    - [`/docs`](/docs) → Swagger Interactive API Docs
    - [`/health`](/health) → Health Check
    - [`/predict`](/predict) → Plate Recognition
    - [`/auth/login`](/auth/login) → Authentication
    """)

    with gr.Tab("System Status"):
        status_btn = gr.Button("Check API Status")
        status_out = gr.JSON()

        def check_health():
            import requests
            port = os.getenv("PORT", "7860")
            try:
                r = requests.get(f"http://localhost:{port}/health", timeout=10)
                return r.json()
            except Exception as e:
                return {"status": "error", "detail": str(e)}

        status_btn.click(check_health, outputs=status_out)
    blocks.load(gpu_warmup)


# ── Combine FastAPI + Gradio so the server serves both ────────────────
combined = gr.mount_gradio_app(fastapi_app, blocks, path="/gradio")

# ── Monkey-patch: inject combined app when HF calls demo.launch() ─────
# HF's Gradio wrapper calls demo.launch(server_name=..., server_port=...).
# We intercept to pass app=combined so the FastAPI routes are included.
_blocks_launch_original = blocks.launch


def _launch_with_app(self, **kwargs):
    import uvicorn
    server_name = kwargs.get("server_name", "0.0.0.0")
    server_port = int(kwargs.get("server_port", os.getenv("PORT", "7860")))
    uvicorn.run(combined, host=server_name, port=server_port)



blocks.launch = types.MethodType(_launch_with_app, blocks)
demo = blocks

if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0", server_port=7860)