core.py

import os
import json
from groq import Groq
from openai import OpenAI

# Load environment variables only if .env file exists (local development)
try:
    from dotenv import load_dotenv

    load_dotenv()
except ImportError:
    # dotenv not needed in production (Hugging Face Spaces)
    pass

# Initialize Groq client
# Security Note: API Key is loaded from environment variable GROQ_API_KEY
# In Hugging Face Spaces, this will be loaded from Repository Secrets
# client = Groq(
#     api_key=os.environ.get("GROQ_API_KEY"),
# )

MODEL_NAME = "llama-3.3-70b-versatile"
GLM_MODEL_NAME = "glm-4.7"
GLM_BASE_URL = "https://api.z.ai/api/coding/paas/v4"

SYSTEM_PROMPT = """
You are an expert English language tutor for Spanish speakers. Your task is to analyze user's English input and provide a "mirror" response that includes:
1.  **Grammatical Error Correction (GEC)**: Correct any morphosyntactic, lexical, or spelling errors.
2.  **Reverse Translation**: Translate the *corrected* English text back into Spanish to verify the semantic intention.
3.  **Pedagogical Explanation**: Explain the grammatical rules violated in Spanish.

Output valid JSON strictly following this schema:
{
  "corrected_text": "Complete corrected English text",
  "spanish_translation": "Spanish translation of the corrected text",
  "errors": [
    {
      "original": "Error word/phrase",
      "correction": "Correction",
      "explanation": "Brief explanation in Spanish",
      "type": "Grammar/Spelling/Style"
    }
  ],
  "general_feedback": "Short motivational message in Spanish"
}

If there are no errors, the "errors" list should be empty, and "corrected_text" should be the same as the input.
Do not include any text outside of JSON object.
"""


def analyze_text_glm(input_text: str, api_key: str) -> dict:
    """
    Analyzes the input English text using GLM-4.7 API and returns a structured response.

    Args:
        input_text (str): The English text to analyze.
        api_key (str): The GLM API key to use.

    Returns:
        dict: The structured analysis result.
    """
    if not input_text.strip():
        return {
            "corrected_text": "",
            "spanish_translation": "",
            "errors": [],
            "general_feedback": "Por favor, ingresa un texto para analizar.",
        }

    if not api_key:
        return {
            "corrected_text": input_text,
            "spanish_translation": "Error de configuración",
            "errors": [],
            "general_feedback": "No se proporcionó una API Key válida.",
        }

    try:
        client = OpenAI(api_key=api_key, base_url=GLM_BASE_URL)

        response = client.chat.completions.create(
            model=GLM_MODEL_NAME,
            messages=[
                {"role": "system", "content": SYSTEM_PROMPT},
                {"role": "user", "content": input_text},
            ],
            temperature=0,
        )

        content = response.choices[0].message.content

        # Log raw response for debugging
        raw_response = content
        if not raw_response.strip():
            return {
                "corrected_text": input_text,
                "spanish_translation": "Error al procesar",
                "errors": [],
                "general_feedback": f"Ocurrió un error: GLM devolvió una respuesta vacía o con solo espacios. Longitud: {len(raw_response)} caracteres. Respuesta cruda: '{raw_response}'",
            }

        # Clean markdown code blocks from response
        cleaned_response = raw_response.strip()

        # Remove ```json and ``` if present
        if cleaned_response.startswith("```json"):
            cleaned_response = cleaned_response[7:]  # Remove ```json
        if cleaned_response.startswith("```"):
            cleaned_response = cleaned_response[3:]  # Remove ```
        if cleaned_response.endswith("```"):
            cleaned_response = cleaned_response[:-3]  # Remove trailing ```

        # Remove any system messages or content after JSON
        # Find closing brace of the JSON object
        try:
            json_end = cleaned_response.rfind("}") + 1
            if json_end > 0:
                cleaned_response = cleaned_response[:json_end]
        except:
            pass

        try:
            return json.loads(cleaned_response)
        except json.JSONDecodeError as json_error:
            # Return actual response content for debugging
            preview = (
                cleaned_response[:500]
                if len(cleaned_response) > 500
                else cleaned_response
            )
            return {
                "corrected_text": input_text,
                "spanish_translation": "Error al procesar",
                "errors": [],
                "general_feedback": f"Error de JSON: {str(json_error)}. Respuesta cruda del modelo (primeros 500 caracteres): {preview}",
            }

    except Exception as e:
        return {
            "corrected_text": input_text,
            "spanish_translation": "Error al procesar",
            "errors": [],
            "general_feedback": f"Ocurrió un error al conectar con GLM: {str(e)}",
        }


def analyze_text(input_text: str, api_key: str, provider: str = "groq") -> dict:
    """
    Analyzes the input English text using the specified provider and returns a structured response.

    Args:
        input_text (str): The English text to analyze.
        api_key (str): The API key to use for the provider.
        provider (str): The provider to use - "groq" or "glm". Defaults to "groq".

    Returns:
        dict: The structured analysis result.
    """
    if provider == "glm":
        return analyze_text_glm(input_text, api_key)

    # Default Groq provider logic
    if not input_text.strip():
        return {
            "corrected_text": "",
            "spanish_translation": "",
            "errors": [],
            "general_feedback": "Por favor, ingresa un texto para analizar.",
        }

    if not api_key:
        return {
            "corrected_text": input_text,
            "spanish_translation": "Error de configuración",
            "errors": [],
            "general_feedback": "No se proporcionó una API Key válida.",
        }

    try:
        client = Groq(api_key=api_key)

        completion = client.chat.completions.create(
            model=MODEL_NAME,
            messages=[
                {"role": "system", "content": SYSTEM_PROMPT},
                {"role": "user", "content": input_text},
            ],
            temperature=0,
            response_format={"type": "json_object"},
        )

        response_content = completion.choices[0].message.content
        return json.loads(response_content)

    except Exception as e:
        # Fallback error handling
        return {
            "corrected_text": input_text,
            "spanish_translation": "Error al procesar",
            "errors": [],
            "general_feedback": f"Ocurrió un error al conectar con el asistente: {str(e)}",
        }


if __name__ == "__main__":
    # Simple test
    test_text = "She dont like play football."
    print("Analyzing:", test_text)
    result = analyze_text(test_text)
    print(json.dumps(result, indent=2, ensure_ascii=False))


def validate_glm_api_key(api_key: str) -> bool:
    """
    Validates the GLM API Key by making a simple request.
    """
    if not api_key:
        return False

    try:
        client = OpenAI(
            model=GLM_MODEL_NAME,
            base_url=GLM_BASE_URL,
            api_key=api_key,
            temperature=0,
            max_tokens=1,
        )
        client.chat.completions.create(
            model=GLM_MODEL_NAME,
            messages=[{"role": "user", "content": "test"}],
            temperature=0,
        )
        return True
    except Exception:
        return False


def validate_api_key(api_key: str) -> bool:
    """
    Validates the Groq API Key by making a simple request (listing models).
    """
    if not api_key:
        return False

    try:
        client = Groq(api_key=api_key)
        # Attempt to list models to verify the key
        client.models.list()
        return True
    except Exception:
        return False