"""
domain/interfaces/services/model_service.py
────────────────────────────────────────────
ModelService — abstract contract for AI model inference.

Implementations include:
  • MockModelService     (testing / local dev, no GPU needed)
  • GANVGTLNetService    (production: GAN + VGTL-Net on GPU)
"""
from __future__ import annotations

from abc import ABC, abstractmethod

import numpy as np

from src.domain.entities.prediction import BPPrediction


class ModelService(ABC):
    """
    AI model inference contract.

    Lifecycle:
        await service.load_model()  → loads weights into memory
        prediction = await service.predict(ppg_signal_id, segments)
        # repeat predict() as needed
    """

    @abstractmethod
    async def load_model(self) -> None:
        """
        Load model weights and prepare for inference.

        This is intentionally async to allow loading from remote storage
        (e.g. Hugging Face Hub, S3) without blocking the event loop.

        Should be idempotent — calling load_model() on an already-loaded
        service must be a safe no-op.
        """
        ...

    @abstractmethod
    async def predict(
        self,
        ppg_signal_id: str,
        segments: np.ndarray,
    ) -> BPPrediction:
        """
        Run blood pressure inference on preprocessed PPG segments.

        Args:
            ppg_signal_id: UUID of the source PPGSignal — stored in the result.
            segments:      2-D NumPy array of shape ``(n_segments, window_size)``
                           produced by SignalProcessor.process().

        Returns:
            BPPrediction domain entity containing predicted SBP, DBP,
            inference timing, and model version.

        Raises:
            PredictionOutOfRangeError: If the model output is physiologically
                                       impossible (handled by BPPrediction.validate()).
            RuntimeError: If the model has not been loaded (load_model() not called).
        """
        ...

    @abstractmethod
    def is_loaded(self) -> bool:
        """Return ``True`` if model weights are loaded and ready for inference."""
        ...

    @property
    @abstractmethod
    def model_version(self) -> str:
        """Version string identifying this model (stored in BPPrediction)."""
        ...