import gradio as gr
import requests
from bs4 import BeautifulSoup
import os
import json
import logging
import pandas as pd
import numpy as np # Added for mean calculation
import matplotlib.pyplot as plt # Added for plotting
from typing import Optional, List, Dict, Any

# ------------------------
# Configuration
# ------------------------
WORDLIFT_API_URL = "https://api.wordlift.io/content-evaluations"
WORDLIFT_API_KEY = os.getenv("WORDLIFT_API_KEY") # Get API key from environment variable

# Set up logging
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__)

# ------------------------
# Custom CSS & Theme
# ------------------------

css = """
@import url('https://fonts.googleapis.com/css2?family=Open+Sans:wght@300;400;600;700&display=swap');
body {
    font-family: 'Open Sans', sans-serif !important;
}
.primary-btn {
    background-color: #3452db !important;
    color: white !important;
}
.primary-btn:hover {
    background-color: #2a41af !important;
}
.gradio-container {
    max-width: 1200px; /* Limit width for better readability */
    margin: auto;
}
.plot-container {
    min-height: 400px; /* Ensure plot area is visible */
}
"""

theme = gr.themes.Soft(
    primary_hue=gr.themes.colors.Color(
        name="blue",
        c50="#eef1ff",
        c100="#e0e5ff",
        c200="#c3cbff",
        c300="#a5b2ff",
        c400="#8798ff",
        c500="#6a7eff",
        c600="#3452db",
        c700="#2a41af",
        c800="#1f3183",
        c900="#152156",
        c950="#0a102b",
    )
)

# ------------------------
# Content Fetching Logic
# ------------------------

def fetch_content_from_url(url: str, timeout: int = 15) -> str:
    """Fetches main text content from a URL."""
    logger.info(f"Fetching content from: {url}")
    try:
        headers = {
            'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36'
        }
        # Use stream=True and then process content to handle large files efficiently,
        # though BeautifulSoup will load it all eventually. Timeout is for connection.
        with requests.get(url, headers=headers, timeout=timeout, stream=True) as response:
             response.raise_for_status() # Raise an exception for bad status codes

             # Limit the amount of data read to avoid excessive memory usage
             max_bytes_to_read = 2 * 1024 * 1024 # 2MB limit for initial read
             content = response.content[:max_bytes_to_read]
             if len(response.content) > max_bytes_to_read:
                 logger.warning(f"Content for {url} is larger than {max_bytes_to_read} bytes, reading truncated content.")

        soup = BeautifulSoup(content, 'html.parser')

        # Attempt to find main content block
        # Prioritize more specific semantic tags
        main_content = soup.find('article') or soup.find('main') or soup.find(class_=lambda x: x and ('content' in x.lower() or 'article' in x.lower()))


        if main_content:
             # Extract text from common text-containing tags within the main block
            text_elements = main_content.find_all(['p', 'h1', 'h2', 'h3', 'h4', 'h5', 'h6', 'li', 'blockquote', 'figcaption'])
            text = ' '.join([elem.get_text() for elem in text_elements])
        else:
            # Fallback to extracting text from body if no main block found
            text_elements = soup.body.find_all(['p', 'h1', 'h2', 'h3', 'h4', 'h5', 'h6', 'li', 'blockquote', 'figcaption'])
            text = ' '.join([elem.get_text() for elem in text_elements])
            logger.warning(f"No specific content tags (<article>, <main>, etc.) found for {url}, extracting from body.")

        # Clean up extra whitespace
        text = ' '.join(text.split())

        # Limit text length *after* extraction and cleaning
        # Adjust based on API limits/cost. WordLift's typical text APIs handle up to ~1M chars.
        max_text_length = 1000000
        if len(text) > max_text_length:
            logger.warning(f"Extracted text for {url} is too long ({len(text)} chars), truncating to {max_text_length} chars.")
            text = text[:max_text_length]

        return text.strip() if text else None # Return None if text is empty after processing

    except requests.exceptions.RequestException as e:
        logger.error(f"Failed to fetch content from {url}: {e}")
        return None
    except Exception as e:
        logger.error(f"Error processing content from {url}: {e}")
        return None

# ------------------------
# WordLift API Call Logic
# ------------------------

def call_wordlift_api(text: str, keywords: Optional[List[str]] = None) -> Optional[Dict[str, Any]]:
    """Calls the WordLift Content Evaluation API."""
    if not WORDLIFT_API_KEY:
        logger.error("WORDLIFT_API_KEY environment variable not set.")
        return {"error": "API key not configured."}

    if not text or not text.strip():
        return {"error": "No significant content to evaluate."}

    payload = {
        "text": text,
        "keywords": keywords if keywords else []
    }

    headers = {
        'Authorization': f'Key {WORDLIFT_API_KEY}',
        'Content-Type': 'application/json',
        'Accept': 'application/json'
    }

    logger.info(f"Calling WordLift API with text length {len(text)} and {len(keywords or [])} keywords.")

    try:
        response = requests.post(WORDLIFT_API_URL, headers=headers, json=payload, timeout=90) # Increased timeout again
        response.raise_for_status() # Raise HTTPError for bad responses (4xx or 5xx)
        return response.json()

    except requests.exceptions.HTTPError as e:
        logger.error(f"WordLift API HTTP error for {e.request.url}: {e.response.status_code} - {e.response.text}")
        try:
            error_detail = e.response.json()
        except json.JSONDecodeError:
            error_detail = e.response.text
        return {"error": f"API returned status code {e.response.status_code}", "details": error_detail}
    except requests.exceptions.Timeout as e:
         logger.error(f"WordLift API request timed out for {e.request.url}: {e}")
         return {"error": f"API request timed out."}
    except requests.exceptions.RequestException as e:
        logger.error(f"WordLift API request error for {e.request.url}: {e}")
        return {"error": f"API request failed: {e}"}
    except Exception as e:
        logger.error(f"Unexpected error during API call: {e}")
        return {"error": f"An unexpected error occurred: {e}"}


# ------------------------
# Plotting Logic
# ------------------------

def plot_average_radar(average_scores: Dict[str, float], avg_overall: Optional[float]) -> Any:
    """Return a radar (spider) plot as a Matplotlib figure showing average scores."""

    if not average_scores or all(v is None for v in average_scores.values()):
        # Return a placeholder figure if no valid data is available
        fig, ax = plt.subplots(figsize=(6, 6))
        ax.text(0.5, 0.5, "No successful evaluations to plot.", horizontalalignment='center', verticalalignment='center', transform=ax.transAxes, fontsize=12)
        ax.axis('off') # Hide axes
        plt.title("Average Content Quality Scores", size=16, y=1.05)
        plt.tight_layout()
        return fig


    categories = list(average_scores.keys())
    values = [average_scores[cat] for cat in categories]

    # Ensure values are floats, replace None with 0 for plotting
    values = [float(v) if v is not None else 0 for v in values]

    num_vars = len(categories)
    # Calculate angles for the radar chart
    angles = [n / float(num_vars) * 2 * np.pi for n in range(num_vars)]
    angles += angles[:1] # Complete the circle
    values += values[:1] # Complete the circle for values

    fig, ax = plt.subplots(figsize=(6, 6), subplot_kw=dict(projection='polar'))

    line_color = '#3452DB'
    fill_color = '#A1A7AF'
    background_color = '#F6F6F7'
    annotation_color = '#191919'

    # Plot data
    ax.plot(angles, values, 'o-', linewidth=2, color=line_color, label='Average Scores')
    ax.fill(angles, values, alpha=0.4, color=fill_color)

    # Set tick locations and labels
    ax.set_xticks(angles[:-1])
    ax.set_xticklabels(categories, color=line_color, fontsize=10)

    # Set y-axis limits. Max score is 100.
    ax.set_ylim(0, 100)

    # Draw grid lines and axes
    ax.grid(True, alpha=0.5, color=fill_color)
    ax.set_facecolor(background_color)

    # Add score annotations next to points
    for angle, value, category in zip(angles[:-1], values[:-1], categories):
        # Adjust position slightly so text doesn't overlap the point/line
        # Radius adjustment can be tricky; let's just add text at the point for simplicity
        ax.text(angle, value + 5, f'{value:.1f}', color=annotation_color,
                horizontalalignment='center', verticalalignment='bottom' if value > 50 else 'top', fontsize=9)


    # Add title
    overall_title = f'Average Content Quality Scores\nOverall: {avg_overall:.1f}/100' if avg_overall is not None else 'Average Content Quality Scores'
    plt.title(overall_title, size=16, y=1.1, color=annotation_color)

    plt.tight_layout()
    return fig

# ------------------------
# Main Evaluation Batch Function
# ------------------------

def evaluate_urls_batch(url_data: pd.DataFrame):
    """
    Evaluates a batch of URLs using the WordLift API.

    Args:
        url_data: A pandas DataFrame with columns ['URL', 'Target Keywords (comma-separated)'].

    Returns:
        A tuple containing:
        - A pandas DataFrame with the summary results.
        - A dictionary containing the full results (including errors) keyed by URL.
        - A Matplotlib figure for the average radar chart.
    """
    # Check if the DataFrame has any rows (correct way using .empty)
    if url_data.empty:
        logger.info("Input DataFrame is empty. Returning empty results.")
        # Return empty summary DF, empty full results, and an empty placeholder plot
        empty_summary_df = pd.DataFrame(columns=[
             'URL', 'Status', 'Overall Score', 'Content Purpose',
             'Content Accuracy', 'Content Depth', 'Readability Score (API)',
             'Readability Grade Level', 'SEO Score', 'Word Count', 'Error/Details'
        ])
        return empty_summary_df, {}, plot_average_radar(None, None) # Pass None to plotting function

    summary_results = []
    full_results = {}

    # Lists to store scores for calculating averages
    purpose_scores = []
    accuracy_scores = []
    depth_scores = []
    readability_scores = []
    seo_scores = []
    overall_scores = []

    # Ensure columns exist, add them if not (though Dataframe component should enforce this)
    # Using .get() with default None is safer if columns might sometimes be missing
    urls = url_data.get('URL', pd.Series(dtype=str))
    keywords_col = url_data.get('Target Keywords (comma-separated)', pd.Series(dtype=str))


    for index, url in enumerate(urls):
        url = url.strip() if pd.notna(url) else ""
        keywords_str = keywords_col.iloc[index].strip() if pd.notna(keywords_col.iloc[index]) else ""
        keywords = [kw.strip() for kw in keywords_str.split(',') if kw.strip()]

        # Generate a unique key for full_results, especially if URL is empty or duplicate
        result_key = url if url else f"Row_{index}"
        # Ensure unique key in case of duplicate empty URLs, maybe use index always?
        result_key = f"Row_{index}_{url}" if url else f"Row_{index}"


        if not url:
            summary_results.append(["", "Skipped", "-", "-", "-", "-", "-", "-", "-", "-", "Empty URL"])
            full_results[result_key] = {"status": "Skipped", "error": "Empty URL input."}
            logger.warning(f"Skipping evaluation for row {index}: Empty URL")
            continue # Move to next URL

        logger.info(f"Processing URL: {url} (Row {index}) with keywords: {keywords}")

        # 1. Fetch Content
        content = fetch_content_from_url(url)

        if content is None or not content.strip():
            status = "Failed"
            error_msg = "Failed to fetch or extract content."
            summary_results.append([url, status, "-", "-", "-", "-", "-", "-", "-", "-", error_msg])
            full_results[result_key] = {"status": status, "error": error_msg}
            logger.error(f"Processing failed for {url} (Row {index}): {error_msg}")
            continue # Move to next URL

        # 2. Call WordLift API
        api_result = call_wordlift_api(content, keywords)

        # 3. Process API Result
        summary_row = [url]
        if api_result and "error" not in api_result:
            status = "Success"
            qs = api_result.get('quality_score', {})
            breakdown = qs.get('breakdown', {})
            content_breakdown = breakdown.get('content', {})
            readability_breakdown = breakdown.get('readability', {})
            seo_breakdown = breakdown.get('seo', {})
            metadata = api_result.get('metadata', {})

            # Append scores for average calculation (only for successful calls)
            purpose_scores.append(content_breakdown.get('purpose'))
            accuracy_scores.append(content_breakdown.get('accuracy'))
            depth_scores.append(content_breakdown.get('depth'))
            readability_scores.append(readability_breakdown.get('score')) # API's readability score (e.g. 2.5)
            seo_scores.append(seo_breakdown.get('score'))
            overall_scores.append(qs.get('overall'))


            # Append data for the summary table row
            summary_row.extend([
                status,
                f'{qs.get("overall", "-"): .1f}',
                f'{content_breakdown.get("purpose", "-"): .0f}', # Assuming integer scores
                f'{content_breakdown.get("accuracy", "-"): .0f}', # Assuming integer scores
                f'{content_breakdown.get("depth", "-"): .0f}', # Assuming integer scores
                f'{readability_breakdown.get("score", "-"): .1f}',
                f'{readability_breakdown.get("grade_level", "-"): .0f}', # Assuming integer grade
                f'{seo_breakdown.get("score", "-"): .1f}',
                f'{metadata.get("word_count", "-"): .0f}', # Assuming integer word count
                None # No error
            ])
            full_results[result_key] = api_result # Store full API result

        else:
            status = "Failed"
            error_msg = api_result.get("error", "Unknown API error.") if api_result else "API call failed."
            details = api_result.get("details", "") if api_result else ""
            summary_row.extend([
                status,
                "-", "-", "-", "-", "-", "-", "-", "-",
                f"{error_msg} {details}"
            ])
            full_results[result_key] = {"status": status, "error": error_msg, "details": details}
            logger.error(f"API call failed for {url} (Row {index}): {error_msg} {details}")

        summary_results.append(summary_row)

    # Calculate Averages *after* processing all URLs
    avg_purpose = np.nanmean(purpose_scores) if purpose_scores else None # Use nanmean to ignore None/NaN
    avg_accuracy = np.nanmean(accuracy_scores) if accuracy_scores else None
    avg_depth = np.nanmean(depth_scores) if depth_scores else None
    avg_readability = np.nanmean(readability_scores) if readability_scores else None
    avg_seo = np.nanmean(seo_scores) if seo_scores else None
    avg_overall = np.nanmean(overall_scores) if overall_scores else None

    # Prepare scores for the radar plot function
    average_scores_dict = {
        'Purpose': avg_purpose,
        'Accuracy': avg_accuracy,
        'Depth': avg_depth,
        'Readability': avg_readability,
        'SEO': avg_seo
    }

    # Generate the average radar plot
    average_radar_fig = plot_average_radar(average_scores_dict, avg_overall)


    # Create pandas DataFrame for summary output
    summary_df = pd.DataFrame(summary_results, columns=[
        'URL', 'Status', 'Overall Score', 'Content Purpose',
        'Content Accuracy', 'Content Depth', 'Readability Score (API)',
        'Readability Grade Level', 'SEO Score', 'Word Count', 'Error/Details'
    ])

    # Note: Formatting is already done when creating the summary_row list above
    # using f-strings like f'{value: .1f}' or f'{value: .0f}', and setting '-' for None.
    # This ensures that pandas DataFrame displays formatted strings directly.


    return summary_df, full_results, average_radar_fig # Return the plot too

# ------------------------
# Gradio Blocks Interface Setup
# ------------------------

with gr.Blocks(css=css, theme=theme) as demo:
    gr.Markdown("# WordLift Multi-URL Content Evaluator")
    gr.Markdown(
        "Enter up to 30 URLs in the table below. "
        "Optionally, provide comma-separated target keywords for each URL. "
        "The app will fetch content from each URL and evaluate it using the WordLift API."
    )

    with gr.Row():
        with gr.Column(scale=1):
            url_input_df = gr.Dataframe(
                headers=["URL", "Target Keywords (comma-separated)"],
                datatype=["str", "str"],
                row_count=(1, 30), # Allow adding rows up to 30
                col_count=(2, "fixed"),
                value=[
                    ["https://www.wordlift.io/blog/google-helpful-content-update-2023/", "helpful content, google update"],
                    ["https://www.wordlift.io/blog/what-is-a-knowledge-graph/", "knowledge graph, semantic web"],
                    ["https://www.example.com/non-existent-page", ""], # Example of a failing URL
                    ["", ""] # Example of an empty row
                ], # Default examples
                label="URLs and Keywords"
            )
            submit_button = gr.Button("Evaluate All URLs", elem_classes=["primary-btn"])

        with gr.Column(scale=1, elem_classes="plot-container"):
             # New component for the average radar plot
             average_radar_output = gr.Plot(label="Average Content Quality Scores Radar")


    gr.Markdown("## Detailed Results")

    with gr.Column():
        summary_output_df = gr.DataFrame(
            label="Summary Results",
            # Data types are all string now because we formatted them with f-strings to include '-'
            headers=['URL', 'Status', 'Overall Score', 'Content Purpose',
                     'Content Accuracy', 'Content Depth', 'Readability Score (API)',
                     'Readability Grade Level', 'SEO Score', 'Word Count', 'Error/Details'],
            datatype=["str"] * 11,
            wrap=True # Wrap text in columns
        )
        with gr.Accordion("Full JSON Results", open=False):
             # Changed the output type to gr.JSON
             full_results_json = gr.JSON(label="Raw API Results per URL (or Error)")

    submit_button.click(
        fn=evaluate_urls_batch,
        inputs=[url_input_df],
        # Updated outputs to include the average radar plot
        outputs=[summary_output_df, full_results_json, average_radar_output]
    )

# Launch the app
if __name__ == "__main__":
    if not WORDLIFT_API_KEY:
        logger.error("\n----------------------------------------------------------")
        logger.error("WORDLIFT_API_KEY environment variable is not set.")
        logger.error("Please set it before running the script:")
        logger.error("  export WORDLIFT_API_KEY='YOUR_API_KEY'")
        logger.error("Or if using a .env file and python-dotenv:")
        logger.error("  pip install python-dotenv")
        logger.error("  # Add WORDLIFT_API_KEY=YOUR_API_KEY to a .env file")
        logger.error("  # import dotenv; dotenv.load_dotenv()")
        logger.error("  # in your script before getting the key.")
        logger.error("----------------------------------------------------------\n")
        # You might want to sys.exit(1) here if the API key is mandatory

    logger.info("Launching Gradio app...")
    # Consider using share=True for easy sharing, but be mindful of security/costs
    # demo.launch(share=True)
    demo.launch()