app.py

import os
import io
import logging
import zipfile
import tarfile
import time
import uvicorn
import fitz  # PyMuPDF
import docx  # python-docx
import pptx  # python-pptx
import openpyxl
import pandas as pd
from PIL import Image
import pytesseract
from fastapi import FastAPI, UploadFile, File, HTTPException, Header, BackgroundTasks, Body
from fastapi.middleware.cors import CORSMiddleware
from typing import List, Optional, Tuple
import asyncio
from concurrent.futures import ThreadPoolExecutor
import magic
import chardet
import json
import xml.etree.ElementTree as ET
from pathlib import Path
import tempfile
import shutil
import subprocess
from pdf2image import convert_from_bytes
import concurrent.futures
from vector import vdb
from pydantic import BaseModel
from typing import Optional
from typing import List, Dict
from fastapi.responses import JSONResponse
import numpy as np
import re
# ==================== CONFIGURATION ====================
logging.basicConfig(
    level=logging.INFO, 
    format='%(asctime)s | %(levelname)s | %(name)s | %(message)s'
)
logger = logging.getLogger("ProductionExtractor")

# Production Configuration
class Config:
    MAX_ZIP_DEPTH = 3
    MAX_FILES_IN_ZIP = 100
    MAX_FILE_SIZE_MB = 50
    MAX_TOTAL_SIZE_MB = 500
    TIMEOUT_SECONDS = 300
    WORKER_THREADS = 4
    TEXTRACT_TIMEOUT = 30
    MAX_PDF_PAGES = 100
    TESSERACT_TIMEOUT = 60
    ENABLE_OCR = True
    MAX_IMAGE_PIXELS = 80_000_000  # ~40MP limit for PIL
    OCR_LANGUAGE = os.getenv("TESSERACT_LANGUAGE", "eng+hin")

class SearchRequest(BaseModel):
    query: str
    target: Optional[str] = None    
# Performance metrics tracking
metrics = {
    "files_processed": 0,
    "total_bytes": 0,
    "processing_time": 0,
    "errors": []
}

app = FastAPI(
    title="NeuralStream Production Extractor",
    version="1.0.0",
    description="High-performance file extraction service with support for 50+ file types",
    docs_url="/docs",
    redoc_url="/redoc"
)

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

# Thread pool for blocking operations
executor = ThreadPoolExecutor(max_workers=Config.WORKER_THREADS)

# Configure Tesseract path if needed
if os.name == 'nt':  # Windows
    tesseract_path = r'C:\Program Files\Tesseract-OCR\tesseract.exe'
    if os.path.exists(tesseract_path):
        pytesseract.pytesseract.tesseract_cmd = tesseract_path

# ==================== UTILITY FUNCTIONS ====================

def sanitize_filename(filename: str) -> str:
    """Sanitize filename to prevent path traversal attacks."""
    return os.path.basename(filename).replace('\\', '/')

def get_file_extension(filename: str) -> str:
    """Extract file extension in a safe way."""
    return Path(filename).suffix.lower()

def detect_file_type(content: bytes, filename: str) -> str:
    """Detect file type using both magic numbers and extension."""
    try:
        mime = magic.from_buffer(content[:2048], mime=True)
        return mime
    except Exception:
        ext = get_file_extension(filename)
        return f"extension/{ext}"

def is_binary_file(content: bytes) -> bool:
    """Heuristic check if file is binary."""
    if not content:
        return False
    if b'\x00' in content[:1024]:
        return True
    # Check if >30% of bytes are non-printable
    text_chars = bytearray({7,8,9,10,12,13,27} | set(range(0x20, 0x100)) - {0x7f})
    sample = content[:1024] if len(content) >= 1024 else content
    if len(sample) == 0:
        return False
    try:
        non_text = sample.translate(None, text_chars)
        return float(len(non_text)) / len(sample) > 0.3
    except:
        return False

def truncate_content(content: str, max_length: int = 100000) -> str:
    """Truncate content if too long, keeping start and end."""
    if len(content) <= max_length:
        return content
    half = max_length // 2
    return content[:half] + f"\n\n[... TRUNCATED {len(content) - max_length} CHARACTERS ...]\n\n" + content[-half:]

# ==================== EXTRACTION ENGINES ====================

def decode_text_safe(content: bytes, filename: str) -> str:
    """Tier 1: Universal text extraction with advanced encoding detection."""
    try:
        # Try UTF-8 first (most common)
        try:
            decoded = content.decode('utf-8')
            if not is_binary_file(content):
                return format_text_content(decoded, filename, 'utf-8')
        except UnicodeDecodeError:
            pass
        
        # Try common encodings
        for encoding in ['utf-8-sig', 'latin-1', 'cp1252', 'ascii']:
            try:
                decoded = content.decode(encoding)
                if not is_binary_file(content):
                    return format_text_content(decoded, filename, encoding)
            except UnicodeDecodeError:
                continue
        
        # Fallback to chardet
        try:
            detection = chardet.detect(content)
            encoding = detection['encoding'] or 'utf-8'
            decoded = content.decode(encoding, errors='replace')
            return format_text_content(decoded, filename, f"{encoding} (detected)")
        except:
            return f"\n--- BINARY/TEXT FILE: {filename} ---\n[Content appears to be binary or has unknown encoding]\n"
            
    except Exception as e:
        logger.error(f"Text extraction error for {filename}: {e}")
        return f"\n[Error extracting text from {filename}: {str(e)}]\n"

def format_text_content(content: str, filename: str, encoding: str) -> str:
    """Format text content with metadata."""
    content = truncate_content(content)
    return f"""
--- TEXT FILE: {filename} ---
Encoding: {encoding}
Size: {len(content)} characters

{content}

--- END TEXT FILE ---
"""

# ==================== DOCUMENT EXTRACTION ====================

def extract_pdf(content: bytes, filename: str) -> str:
    """Advanced PDF extraction with OCR fallback."""
    start_time = time.time()
    try:
        text_buffer = []
        metadata_info = []
        
        with fitz.open(stream=content, filetype="pdf") as doc:
            if doc.is_encrypted:
                try:
                    doc.authenticate("")
                except:
                    return f"\n[ENCRYPTED PDF: {filename} - Cannot extract content]\n"
            
            metadata = doc.metadata
            if metadata:
                metadata_info.append(f"Title: {metadata.get('title', 'N/A')}")
                metadata_info.append(f"Author: {metadata.get('author', 'N/A')}")
                metadata_info.append(f"Subject: {metadata.get('subject', 'N/A')}")
                metadata_info.append(f"Created: {metadata.get('creationDate', 'N/A')}")
            
            total_pages = len(doc)
            pages_extracted = 0
            
            for i, page in enumerate(doc):
                if i >= Config.MAX_PDF_PAGES:
                    text_buffer.append(f"\n[... Truncated at {Config.MAX_PDF_PAGES} pages from total {total_pages} ...]\n")
                    break
                
                page_text = page.get_text("text")
                if page_text.strip():
                    text_buffer.append(f"\n--- Page {i+1} ---")
                    text_buffer.append(page_text)
                    pages_extracted += 1
        
        full_text = "\n".join(text_buffer)
        
        if len(full_text.strip()) < 10 and Config.ENABLE_OCR:
            logger.info(f"PDF appears to be scanned, attempting OCR: {filename}")
            ocr_result = extract_text_from_image_pdf(content, filename)
            if ocr_result:
                elapsed = time.time() - start_time
                return f"""
=== PDF DOCUMENT (OCR): {filename} ===
Metadata:
{chr(10).join(metadata_info)}

Processing Time: {elapsed:.2f}s
Pages: {pages_extracted}/{total_pages}

{ocr_result}

=== END PDF ===
"""
        
        elapsed = time.time() - start_time
        return f"""
=== PDF DOCUMENT: {filename} ===
Metadata:
{chr(10).join(metadata_info)}

Extraction Time: {elapsed:.2f}s
Pages: {pages_extracted}/{total_pages}

{full_text}

=== END PDF ===
"""
        
    except Exception as e:
        logger.error(f"PDF extraction error for {filename}: {e}")
        return f"\n[Error parsing PDF {filename}: {str(e)}]\n"

def extract_docx(content: bytes, filename: str) -> str:
    """Advanced DOCX extraction with tables."""
    try:
        doc = docx.Document(io.BytesIO(content))
        
        properties = []
        if doc.core_properties.title:
            properties.append(f"Title: {doc.core_properties.title}")
        if doc.core_properties.author:
            properties.append(f"Author: {doc.core_properties.author}")
        if doc.core_properties.created:
            properties.append(f"Created: {doc.core_properties.created}")
        
        paragraphs = []
        for para in doc.paragraphs:
            if para.text.strip():
                paragraphs.append(para.text)
        
        tables_text = []
        for i, table in enumerate(doc.tables):
            table_data = []
            for row in table.rows:
                row_data = [cell.text for cell in row.cells]
                table_data.append(" | ".join(row_data))
            if table_data:
                tables_text.append(f"\n--- Table {i+1} ---")
                tables_text.append("\n".join(table_data))
        
        result = "\n".join(paragraphs)
        if tables_text:
            result += "\n" + "\n".join(tables_text)
            
        return f"""
=== WORD DOCUMENT: {filename} ===
Metadata:
{chr(10).join(properties)}

{result}

=== END DOCUMENT ===
"""
        
    except Exception as e:
        logger.error(f"DOCX extraction error for {filename}: {e}")
        return f"\n[Error parsing DOCX {filename}: {str(e)}]\n"

def extract_pptx(content: bytes, filename: str) -> str:
    """Extract text from PowerPoint presentations."""
    try:
        prs = pptx.Presentation(io.BytesIO(content))
        text_slides = []
        
        for i, slide in enumerate(prs.slides):
            slide_text = []
            for shape in slide.shapes:
                if hasattr(shape, "text") and shape.text:
                    if shape.text.strip():
                        slide_text.append(shape.text)
                # Check for table text
                if shape.has_table:
                    for row in shape.table.rows:
                        for cell in row.cells:
                            if cell.text.strip():
                                slide_text.append(cell.text)
            
            if slide_text:
                text_slides.append(f"\n--- Slide {i+1} ---")
                text_slides.extend(slide_text)
        
        return f"""
=== POWERPOINT: {filename} ===
Slides: {len(prs.slides)}

{chr(10).join(text_slides)}

=== END POWERPOINT ===
"""
        
    except Exception as e:
        logger.error(f"PPTX extraction error for {filename}: {e}")
        return f"\n[Error parsing PPTX {filename}: {str(e)}]\n"

def extract_excel(content: bytes, filename: str) -> str:
    """Extract data from Excel files."""
    try:
        wb = openpyxl.load_workbook(io.BytesIO(content), read_only=True, data_only=True)
        sheets_data = []
        
        for sheet_name in wb.sheetnames:
            sheet = wb[sheet_name]
            sheet_rows = []
            
            max_rows = 100
            for i, row in enumerate(sheet.iter_rows(values_only=True)):
                if i >= max_rows:
                    break
                row_data = [str(cell) if cell is not None else "" for cell in row]
                sheet_rows.append(" | ".join(row_data))
            
            if sheet_rows:
                sheets_data.append(f"\n--- Sheet: {sheet_name} ---")
                sheets_data.extend(sheet_rows)
                if len(sheet_rows) >= max_rows:
                    sheets_data.append(f"[... Only first {max_rows} rows shown ...]")
        
        try:
            df = pd.read_excel(io.BytesIO(content), engine='openpyxl')
            pandas_output = df.head(50).to_string(index=False, max_rows=50, max_colwidth=50)
            if pandas_output:
                sheets_data.append("\n--- Pandas Format (First 50 rows) ---")
                sheets_data.append(pandas_output)
                if len(df) > 50:
                    sheets_data.append(f"[... {len(df) - 50} more rows truncated ...]")
        except Exception as pandas_error:
            logger.warning(f"Pandas extraction failed: {pandas_error}")
        
        return f"""
=== EXCEL FILE: {filename} ===
{chr(10).join(sheets_data)}

=== END EXCEL ===
"""
        
    except Exception as e:
        logger.error(f"Excel extraction error for {filename}: {e}")
        return f"\n[Error parsing Excel {filename}: {str(e)}]\n"

# ==================== IMAGE EXTRACTION ====================

def extract_text_from_image_pdf(pdf_content: bytes, filename: str) -> Optional[str]:
    """Extract text from image-based PDF using OCR with pdf2image."""
    if not Config.ENABLE_OCR:
        return None
        
    try:
        extracted_text = []
        
        # Convert PDF to images with proper error handling
        images = convert_from_bytes(
            pdf_content,
            dpi=300,
            fmt='jpeg',
            thread_count=2,
            poppler_path=None  # Will use system poppler
        )
        
        logger.info(f"Converted {len(images)} pages from {filename} for OCR")
        
        with concurrent.futures.ThreadPoolExecutor(max_workers=2) as executor:
            future_to_page = {
                executor.submit(perform_ocr_on_image, image, page_num): page_num
                for page_num, image in enumerate(images[:Config.MAX_PDF_PAGES])
            }
            
            for future in concurrent.futures.as_completed(future_to_page, timeout=Config.TESSERACT_TIMEOUT):
                page_num = future_to_page[future]
                try:
                    text = future.result(timeout=30)
                    if text and text.strip():
                        extracted_text.append(f"\n--- Page {page_num + 1} (OCR) ---")
                        extracted_text.append(text)
                        logger.info(f"OCR completed for page {page_num + 1}")
                except Exception as e:
                    logger.warning(f"OCR failed for page {page_num + 1}: {e}")
                    continue
        
        if extracted_text:
            return "\n".join(extracted_text)
        else:
            return None
            
    except Exception as e:
        logger.error(f"PDF to image conversion or OCR failed for {filename}: {e}")
        return None

def perform_ocr_on_image(image: Image.Image, page_num: int) -> str:
    """Perform OCR on a single image with proper configuration."""
    try:
        # Resize if too large
        width, height = image.size
        total_pixels = width * height
        
        if total_pixels > Config.MAX_IMAGE_PIXELS:
            scale_factor = (Config.MAX_IMAGE_PIXELS / total_pixels) ** 0.5
            new_width = int(width * scale_factor)
            new_height = int(height * scale_factor)
            image = image.resize((new_width, new_height), Image.Resampling.LANCZOS)
            logger.info(f"Resized page {page_num + 1} from {width}x{height} to {new_width}x{new_height}")
        
        # Configure Tesseract
        custom_config = f'--oem 3 --psm 3 -l {Config.OCR_LANGUAGE}'
        
        # Perform OCR
        text = pytesseract.image_to_string(image, config=custom_config, timeout=30)
        
        return truncate_content(text.strip(), max_length=50000)
        
    except Exception as e:
        logger.error(f"OCR error on page {page_num + 1}: {e}")
        return ""

def extract_image_ocr(content: bytes, filename: str) -> str:
    """Extract text from image files using OCR."""
    if not Config.ENABLE_OCR:
        return f"\n[IMAGE FILE: {filename}]\n[Image extraction disabled]\n"
    
    try:
        with tempfile.NamedTemporaryFile(delete=False, suffix=get_file_extension(filename)) as temp_img:
            temp_img.write(content)
            temp_img.flush()
            
            try:
                # Open and check image
                with Image.open(temp_img.name) as img:
                    img = img.convert('RGB')  # Ensure RGB mode
                    
                    # Resize if too large
                    width, height = img.size
                    total_pixels = width * height
                    
                    if total_pixels > Config.MAX_IMAGE_PIXELS:
                        scale_factor = (Config.MAX_IMAGE_PIXELS / total_pixels) ** 0.5
                        new_size = (int(width * scale_factor), int(height * scale_factor))
                        img = img.resize(new_size, Image.Resampling.LANCZOS)
                    
                    # Perform OCR
                    custom_config = f'--oem 3 --psm 3 -l {Config.OCR_LANGUAGE}'
                    text = pytesseract.image_to_string(img, config=custom_config, timeout=30)
                    
                    if text.strip():
                        return f"""
--- IMAGE FILE (OCR): {filename} ---
Size: {img.size[0]}x{img.size[1]} pixels
Format: {img.format}

Extracted Text:
{text.strip()}

--- END IMAGE ---
"""
                    else:
                        return f"\n[IMAGE FILE: {filename}]\n[No text detected in image]\n"
                        
            finally:
                os.unlink(temp_img.name)
                
    except Exception as e:
        logger.error(f"Image OCR extraction error for {filename}: {e}")
        return f"\n[Error processing image {filename}: {str(e)}]\n"

# ==================== ARCHIVE EXTRACTION ====================

def process_zip_archive(zip_bytes: bytes, zip_name: str, depth: int = 0) -> Tuple[str, int]:
    """Recursive ZIP extraction with safety limits."""
    if depth > Config.MAX_ZIP_DEPTH:
        return f"\n[ZIP Depth Limit Reached: {zip_name}]\n", 0
    
    output_log = f"\n>>> ZIP ARCHIVE: {zip_name} (Depth {depth}) <<<\n"
    file_count = 0
    total_size = 0
    
    try:
        with zipfile.ZipFile(io.BytesIO(zip_bytes)) as z:
            file_list = [f for f in z.infolist() 
                        if not f.filename.startswith(('.', '__')) 
                        and not f.is_dir()]
            
            for zf in file_list:
                if file_count >= Config.MAX_FILES_IN_ZIP:
                    output_log += f"\n[... File limit reached: {Config.MAX_FILES_IN_ZIP} files ...]\n"
                    break
                
                if zf.file_size == 0 or zf.file_size > (Config.MAX_FILE_SIZE_MB * 1024 * 1024):
                    continue
                
                total_size += zf.file_size
                if total_size > (Config.MAX_TOTAL_SIZE_MB * 1024 * 1024):
                    output_log += f"\n[... Total size limit reached: {Config.MAX_TOTAL_SIZE_MB}MB ...]\n"
                    break
                
                try:
                    with z.open(zf) as f:
                        content = f.read()
                    
                    ext = get_file_extension(zf.filename)
                    
                    if ext in ['.zip']:
                        nested_output, nested_count = process_zip_archive(content, zf.filename, depth + 1)
                        output_log += nested_output
                        file_count += nested_count
                    else:
                        output_log += process_file_bytes(zf.filename, content)
                        file_count += 1
                        
                except Exception as e:
                    logger.error(f"Error processing nested file {zf.filename}: {e}")
                    output_log += f"\n[Error processing {zf.filename} inside {zip_name}]\n"
                    continue
                
    except zipfile.BadZipFile:
        return f"\n[Error: Corrupt Zip Archive - {zip_name}]\n", 0
    except Exception as e:
        logger.error(f"Zip processing error for {zip_name}: {e}")
        return f"\n[Zip Processing Error: {str(e)}]\n", 0
    
    output_log += f"\n>>> END ZIP: {zip_name} ({file_count} files) <<<\n"
    return output_log, file_count

def extract_tar_gz(content: bytes, filename: str) -> str:
    """Extract files from tar.gz archives."""
    output_log = f"\n>>> TAR.GZ ARCHIVE: {filename} <<<\n"
    file_count = 0
    
    try:
        # Determine compression mode
        if filename.endswith('.tar.gz') or filename.endswith('.tgz'):
            mode = 'r:gz'
        elif filename.endswith('.tar.bz2'):
            mode = 'r:bz2'
        elif filename.endswith('.tar.xz'):
            mode = 'r:xz'
        else:
            mode = 'r:'
        
        with tarfile.open(fileobj=io.BytesIO(content), mode=mode) as tar:
            members = [m for m in tar.getmembers() 
                      if m.isfile() 
                      and not m.name.startswith(('.', '__'))
                      and m.size <= (Config.MAX_FILE_SIZE_MB * 1024 * 1024)]
            
            for member in members:
                if file_count >= Config.MAX_FILES_IN_ZIP:
                    output_log += "\n[...Tar file limit reached...]\n"
                    break
                
                try:
                    f = tar.extractfile(member)
                    if f:
                        content = f.read()
                        output_log += process_file_bytes(member.name, content)
                        file_count += 1
                except Exception as e:
                    logger.error(f"Error extracting {member.name}: {e}")
                    continue
                    
    except Exception as e:
        logger.error(f"TAR extraction error for {filename}: {e}")
        return f"\n[Error processing TAR {filename}: {str(e)}]\n"
    
    output_log += f"\n>>> END TAR: {filename} ({file_count} files) <<<\n"
    return output_log

# ==================== STRUCTURED DATA EXTRACTION ====================

def extract_json(content: bytes, filename: str) -> str:
    """Extract and format JSON files."""
    try:
        json_obj = json.loads(content.decode('utf-8'))
        formatted = json.dumps(json_obj, indent=2, ensure_ascii=False)
        return f"""
=== JSON FILE: {filename} ===
{formatted}

=== END JSON ===
"""
    except Exception as e:
        logger.error(f"JSON parsing error for {filename}: {e}")
        return decode_text_safe(content, filename)

def extract_xml(content: bytes, filename: str) -> str:
    """Extract readable text from XML files."""
    try:
        root = ET.fromstring(content)
        
        def extract_text(element, depth=0):
            text_parts = []
            indent = "  " * depth
            
            text_parts.append(f"{indent}<{element.tag}>")
            
            if element.text and element.text.strip():
                text_parts.append(f"{indent}  {element.text.strip()}")
            
            for child in element:
                text_parts.extend(extract_text(child, depth + 1))
            
            text_parts.append(f"{indent}</{element.tag}>")
            return text_parts
        
        extracted = extract_text(root)
        return f"""
=== XML FILE: {filename} ===
{chr(10).join(extracted)}

=== END XML ===
"""
    except Exception as e:
        logger.error(f"XML parsing error for {filename}: {e}")
        return decode_text_safe(content, filename)

def extract_csv(content: bytes, filename: str) -> str:
    """Extract and format CSV files."""
    try:
        df = pd.read_csv(io.BytesIO(content), encoding_errors='replace')
        output = df.head(100).to_string(index=False, max_rows=100, max_colwidth=50)
        row_count = len(df)
        
        result = f"""
=== CSV FILE: {filename} ===
Total Rows: {row_count}
Columns: {', '.join(df.columns.astype(str))}

First 100 Rows:
{output}
"""
        
        if row_count > 100:
            result += f"\n[... {row_count - 100} more rows truncated ...]\n"
        
        result += "\n=== END CSV ===\n"
        return result
        
    except Exception as e:
        logger.error(f"CSV parsing error for {filename}: {e}")
        return decode_text_safe(content, filename)

# ==================== MAIN ROUTING LOGIC ====================

def process_file_bytes(filename: str, content: bytes) -> str:
    """Route files to appropriate extraction engines."""
    start_time = time.time()
    
    safe_name = sanitize_filename(filename)
    content_size = len(content)
    ext = get_file_extension(safe_name)
    
    try:
        result = ""
        
        # Document files
        if ext == '.pdf':
            result = extract_pdf(content, safe_name)
        elif ext == '.docx':
            result = extract_docx(content, safe_name)
        elif ext == '.pptx':
            result = extract_pptx(content, safe_name)
        elif ext in ['.xlsx', '.xls']:
            result = extract_excel(content, safe_name)
        
        # Archive files
        elif ext == '.zip':
            archive_result, count = process_zip_archive(content, safe_name)
            result = archive_result
        elif ext in ['.tar', '.tar.gz', '.tgz', '.tar.bz2', '.tar.xz']:
            result = extract_tar_gz(content, safe_name)
        
        # Structured data
        elif ext == '.json':
            result = extract_json(content, safe_name)
        elif ext == '.xml':
            result = extract_xml(content, safe_name)
        elif ext == '.csv':
            result = extract_csv(content, safe_name)
        
        # Image files with OCR
        elif ext in ['.png', '.jpg', '.jpeg', '.gif', '.bmp', '.webp', '.tiff', '.tif']:
            result = extract_image_ocr(content, safe_name)
        
        # Code and text files
        elif ext in [
            '.py', '.js', '.ts', '.tsx', '.jsx', '.vue', '.svelte',
            '.java', '.kt', '.scala', '.clj', '.cljs', '.cljc',
            '.c', '.cpp', '.h', '.hpp', '.cs', '.fs', '.vb',
            '.go', '.rs', '.swift', '.dart', '.php', '.rb', '.pl',
            '.lua', '.r', '.scm', '.hs', '.elm', '.ex', '.exs',
            '.html', '.htm', '.xhtml', '.css', '.scss', '.sass', '.less',
            '.yaml', '.yml', '.toml', '.ini', '.env', '.cfg',
            '.svg', '.sql', '.sh', '.bash', '.zsh', '.fish',
            '.ps1', '.bat', '.cmd', '.md', '.markdown', '.rst', 
            '.txt', '.log', '.tsv'
        ]:
            result = decode_text_safe(content, safe_name)
        
        # Binary files
        elif ext in ['.exe', '.dll', '.so', '.dylib', '.bin', '.dat']:
            result = f"\n[BINARY FILE: {safe_name}]\nSize: {content_size} bytes\n[Binary content not extractable]\n"
        
        # Audio/Video files
        elif ext in ['.mp3', '.mp4', '.avi', '.mov', '.wav', '.flac', '.mkv', '.webm']:
            result = f"\n[MEDIA FILE: {safe_name}]\nSize: {content_size} bytes\n[Media content not extractable]\n"
        
        # Database files
        elif ext in ['.db', '.sqlite', '.sqlite3', '.mdb', '.accdb']:
            result = f"\n[DATABASE FILE: {safe_name}]\n[Database content not extractable for security reasons]\n"
        
        # Unknown file type
        else:
            file_type = detect_file_type(content, safe_name)
            if not is_binary_file(content):
                result = decode_text_safe(content, safe_name)
            else:
                result = f"\n[UNKNOWN FILE TYPE: {safe_name}]\nType: {file_type}\nSize: {content_size} bytes\n[Binary content not extractable]\n"
        
        elapsed = time.time() - start_time
        metrics["files_processed"] += 1
        metrics["total_bytes"] += content_size
        
        logger.info(f"Extracted {safe_name} ({content_size} bytes) in {elapsed:.2f}s")
        return result
        
    except Exception as e:
        error_msg = f"Error processing {safe_name}: {str(e)}"
        logger.error(error_msg)
        metrics["errors"].append(error_msg)
        return f"\n[FATAL ERROR processing {safe_name}: {str(e)}]\n"

async def process_file_async(file: UploadFile) -> str:
    """Process a single file asynchronously."""
    loop = asyncio.get_event_loop()
    
    try:
        content = await file.read()
        safe_name = sanitize_filename(file.filename)
        
        if len(content) > (Config.MAX_FILE_SIZE_MB * 1024 * 1024):
            return f"\n[ERROR: {safe_name} exceeds {Config.MAX_FILE_SIZE_MB}MB limit]\n"
        
        result = await loop.run_in_executor(executor, process_file_bytes, safe_name, content)
        return result
        
    except Exception as e:
        error_msg = f"Async processing error for {file.filename}: {str(e)}"
        logger.error(error_msg)
        metrics["errors"].append(error_msg)
        return f"\n[ERROR processing {file.filename}: {str(e)}]\n"

# ==================== API ENDPOINTS ====================

@app.post("/api/ingest")
async def ingest_files(files: List[UploadFile] = File(...)):
    """Universal file ingestion endpoint with async processing."""
    if not files:
        raise HTTPException(status_code=400, detail="No files provided")
    
    start_time = time.time()
    
    logger.info(f"Processing batch of {len(files)} files")
    
    tasks = [process_file_async(file) for file in files]
    results = await asyncio.gather(*tasks, return_exceptions=True)
    
    combined_result = ""
    files_processed = 0
    errors = []
    total_size = 0
    
    for i, result in enumerate(results):
        if isinstance(result, Exception):
            error_msg = f"Error processing {files[i].filename}: {str(result)}"
            logger.error(error_msg)
            errors.append(error_msg)
            combined_result += f"\n[ERROR: {error_msg}]\n"
        else:
            combined_result += result
            files_processed += 1
            try:
                if hasattr(files[i], 'size'):
                    total_size += files[i].size
            except:
                pass
    
    elapsed = time.time() - start_time
    
    logger.info(f"Batch processed in {elapsed:.2f}s - {files_processed} files, {total_size} bytes")
    
    return {
        "status": "success",
        "extracted_text": combined_result,
        "files_processed": files_processed,
        "total_files": len(files),
        "processing_time": elapsed,
        "total_size_bytes": total_size,
        "errors": errors if errors else []
    }

import re # Ensure this is imported at the top of app.py

@app.post("/api/interaction")
async def interact_with_files(
    files: List[UploadFile] = File(...),
    x_user_id: str = Header(..., alias="X-User-ID"),
    x_chat_id: str = Header(..., alias="X-Chat-ID"),
    x_file_id: Optional[str] = Header(None, alias="X-File-ID")
):
    """
    Process files and store them in vector DB with user session isolation.
    INCLUDES FIX: Strips metadata headers before DB storage to prevent AST Parser crashes.
    """
    if not files:
        raise HTTPException(status_code=400, detail="No files provided")
    
    start_time = time.time()
    logger.info(f"📤 Processing {len(files)} files for user {x_user_id[:8]}...")
    
    # 1. Extract text from files (Async processing)
    tasks = [process_file_async(file) for file in files]
    results = await asyncio.gather(*tasks, return_exceptions=True)
    
    combined_result = ""
    files_processed = 0
    storage_errors = []
    
    # Regex to strip the "Wrapper" headers (e.g., --- TEXT FILE: app.py ---)
    # Matches: Header -> Metadata Block -> Double Newline -> CONTENT -> Double Newline -> Footer
    wrapper_pattern = r"(?s)(?:---|===)\s+.*?(?:FILE|DOCUMENT).*?[-=]+\n.*?\n\n(.*?)\n\n(?:---|===) END"
    
    # 2. Process each file and store in vector DB
    for i, result in enumerate(results):
        if isinstance(result, Exception):
            error_msg = f"Error processing {files[i].filename}: {str(result)}"
            logger.error(error_msg)
            combined_result += f"\n[ERROR: {error_msg}]\n"
            continue
        
        # Add to combined result (Keep headers for the User UI!)
        combined_result += result
        files_processed += 1
        
        # 3. Prepare Clean Content for Vector DB
        filename = files[i].filename
        clean_text_for_db = result
        
        # Attempt to unwrap the content so the AST parser works
        match = re.search(wrapper_pattern, result)
        if match:
            # Found the "meat" of the file, use that
            clean_text_for_db = match.group(1)
        else:
            # Fallback: If regex misses (e.g. short file), use original but trim whitespace
            clean_text_for_db = result.strip()

        try:
            # Get vector DB instance
            from vector import vdb
            
            # 4. SYNC storage in vector DB using CLEAN TEXT
            # We pass the pure code (clean_text_for_db) but the real filename
            # This allows V3 to parse classes/functions correctly while linking them to the source file.
            storage_success = vdb.store_session_document(
                text=clean_text_for_db, 
                filename=filename,
                user_id=x_user_id,
                chat_id=x_chat_id,
                file_id=x_file_id
            )
            
            if not storage_success:
                error_msg = f"Vector storage failed for {filename}"
                logger.error(error_msg)
                storage_errors.append(error_msg)
                combined_result += f"\n[WARNING: Vector storage failed for {filename}]\n"
            else:
                logger.info(f"✅ Vector storage successful for {filename}")
                
        except Exception as e:
            error_msg = f"Vector DB error for {filename}: {str(e)}"
            logger.error(error_msg)
            storage_errors.append(error_msg)
            combined_result += f"\n[WARNING: {error_msg}]\n"
    
    elapsed = time.time() - start_time
    
    # 5. Return response
    response_data = {
        "status": "success",
        "extracted_text": combined_result,
        "files_processed": files_processed,
        "total_files": len(files),
        "processing_time": round(elapsed, 2),
        "vector_status": "stored_synchronously",
        "session_id": x_user_id,
        "storage_errors": storage_errors if storage_errors else []
    }
    
    logger.info(f"✅ Interaction completed in {elapsed:.2f}s for user {x_user_id[:8]}")
    
    return response_data

@app.delete("/api/deletefile")
async def delete_specific_file_endpoint(
    file_id: str,  # Expects ?file_id=... in the URL
    x_user_id: str = Header(..., alias="X-User-ID"),
    x_chat_id: str = Header(..., alias="X-Chat-ID")
):
    """
    Surgical Deletion Endpoint:
    Removes ONLY the vector chunks associated with a specific file_id.
    """
    from vector import vdb
    
    # Run in thread to prevent blocking the main event loop
    success = await asyncio.to_thread(vdb.delete_file, x_user_id, x_chat_id, file_id)
    
    if success:
        logger.info(f"🗑️ Deleted file {file_id} for user {x_user_id[:8]}")
        return {"status": "deleted", "file_id": file_id}
    else:
        # 404 indicates the file wasn't found (maybe already deleted or never existed)
        return JSONResponse(
            status_code=404, 
            content={"status": "not_found", "message": "File ID not found in current session"}
        )
# Add debug endpoints for monitoring
@app.get("/api/vector/debug")
async def debug_vector_status(x_user_id: str = Header(..., alias="X-User-ID")):
    """Debug endpoint to check vector DB status"""
    from vector import vdb
    
    stats = vdb.get_user_stats(x_user_id)
    
    return {
        "user_id": x_user_id,
        "stats": stats,
        "index_status": {
            "total_vectors": vdb.index.ntotal,
            "total_metadata": len(vdb.metadata),
            "index_type": vdb.index.__class__.__name__
        }
    }

@app.post("/api/vector/cleanup")
async def cleanup_vector_db(
    max_age_hours: int = 24,
    x_user_id: str = Header(..., alias="X-User-ID")
):
    """Clean up old session data"""
    from vector import vdb
    
    try:
        cleaned = vdb.cleanup_old_sessions(max_age_hours)
        return {
            "status": "success",
            "cleaned_vectors": cleaned,
            "max_age_hours": max_age_hours,
            "user_id": x_user_id
        }
    except Exception as e:
        logger.error(f"Cleanup failed: {e}")
        raise HTTPException(status_code=500, detail=str(e))

@app.delete("/api/session")
async def delete_specific_session(
    x_user_id: str = Header(..., alias="X-User-ID"),
    x_chat_id: str = Header(..., alias="X-Chat-ID")
):
    """Triggered when user clicks 'Delete Chat' in UI"""
    from vector import vdb
    
    # Run in thread to not block other users while rebuilding index
    success = await asyncio.to_thread(vdb.delete_session, x_user_id, x_chat_id)
    
    if success:
        return {"status": "deleted", "chat_id": x_chat_id}
    else:
        return {"status": "not_found", "message": "Session was already empty"}

@app.post("/api/search")
async def search_vector_db(
    payload: SearchRequest,
    x_user_id: str = Header(..., alias="X-User-ID"),
    x_chat_id: str = Header(..., alias="X-Chat-ID")
):
    """
    Search within user's session data with proper JSON serialization.
    """
    from vector import vdb
    
    logger.info(f"🔍 Search request from user {x_user_id[:8]}: '{payload.query[:50]}...'")
    
    try:
        results = vdb.retrieve_session_context(
            query=payload.query,
            user_id=x_user_id,
            chat_id=x_chat_id,
            filter_type=payload.target,
            top_k=50,
            final_k=2
        )
        
        logger.info(f"✅ Search completed: {len(results)} results for user {x_user_id[:8]}")
        
        # MANUALLY serialize to handle numpy types
        def serialize(obj):
            if isinstance(obj, (np.integer, np.floating)):
                return float(obj)
            elif isinstance(obj, np.ndarray):
                return obj.tolist()
            elif isinstance(obj, dict):
                return {k: serialize(v) for k, v in obj.items()}
            elif isinstance(obj, list):
                return [serialize(item) for item in obj]
            return obj
        
        serialized_results = serialize(results)
        
        # Use JSONResponse with custom encoder
        return JSONResponse(
            content={"results": serialized_results},
            media_type="application/json"
        )
        
    except Exception as e:
        logger.error(f"Search failed: {e}")
        raise HTTPException(status_code=500, detail=f"Search failed: {str(e)}")

@app.post("/api/sync")
async def sync_chat_history(
    background_tasks: BackgroundTasks,
    messages: List[Dict] = Body(...),
    x_user_id: str = Header(..., alias="X-User-ID"), # <--- 1. Catch the ID
    x_chat_id: str = Header(..., alias="X-Chat-ID")
):
    """
    Syncs chat history for the specific user session.
    """
    if not messages:
        return {"status": "ignored", "reason": "empty"}

    # Trigger Secure Storage
    background_tasks.add_task(
        vdb.store_chat_context,      # <--- Renamed Function
        messages=messages, 
        user_id=x_user_id,            # <--- Pass the ID
        chat_id=x_chat_id,
    )
    
    return {"status": "syncing_started"}

@app.post("/api/single")
async def ingest_single_file(file: UploadFile = File(...)):
    """Process a single file endpoint."""
    start_time = time.time()
    
    result = await process_file_async(file)
    elapsed = time.time() - start_time
    
    logger.info(f"Single file processed in {elapsed:.2f}s")
    
    return {
        "status": "success",
        "extracted_text": result,
        "filename": file.filename,
        "processing_time": elapsed,
        "file_size": file.size
    }

@app.get("/health")
async def health_check():
    """Comprehensive health check endpoint."""
    return {
        "status": "active",
        "version": "1.0.0",
        "engine": "High-Performance Production Extractor",
        "config": {
            "max_file_size_mb": Config.MAX_FILE_SIZE_MB,
            "max_zip_depth": Config.MAX_ZIP_DEPTH,
            "max_files_in_zip": Config.MAX_FILES_IN_ZIP,
            "worker_threads": Config.WORKER_THREADS,
            "enable_ocr": Config.ENABLE_OCR
        },
        "metrics": {
            "files_processed": metrics["files_processed"],
            "total_bytes_processed": metrics["total_bytes"],
            "error_count": len(metrics["errors"])
        },
        "supported_types": [
            "Documents: .pdf, .docx, .pptx, .xlsx, .xls",
            "Code: 20+ programming languages",
            "Archives: .zip, .tar, .tar.gz, .tar.bz2",
            "Data: .json, .xml, .csv, .tsv",
            "Text: .txt, .md, .log, .ini, .yaml",
            "Images: .png, .jpg, .jpeg, .tiff (OCR)"
        ]
    }

@app.get("/metrics")
async def get_metrics():
    """Get detailed performance metrics."""
    avg_bytes = metrics["total_bytes"] / max(1, metrics["files_processed"]) if metrics["files_processed"] > 0 else 0
    
    return {
        "status": "ok",
        "metrics": {
            **metrics,
            "average_bytes_per_file": round(avg_bytes, 2),
            "uptime_seconds": metrics["processing_time"],
            "latest_errors": metrics["errors"][-10:] if len(metrics["errors"]) > 10 else metrics["errors"]
        }
    }





# ==================== STRUCTURED IMPORT ENDPOINTS ====================

def _compute_median_font_size(blocks: list) -> float:
    """Compute the median font size from all text spans — this is our 'body text' baseline."""
    sizes = []
    for block in blocks:
        if block.get("type") != 0:  # type 0 = text block
            continue
        for line in block.get("lines", []):
            for span in line.get("spans", []):
                text = span.get("text", "").strip()
                if text:
                    sizes.append(span.get("size", 12))
    if not sizes:
        return 12.0
    sizes.sort()
    mid = len(sizes) // 2
    return sizes[mid] if len(sizes) % 2 == 1 else (sizes[mid - 1] + sizes[mid]) / 2


def _classify_heading(font_size: float, median: float, flags: int) -> str:
    """Classify a text block as heading or paragraph based on font size ratio to median."""
    if median == 0:
        return "p"
    ratio = font_size / median
    is_bold = bool(flags & (1 << 4))
    if ratio >= 1.6:
        return "h1"
    elif ratio >= 1.35:
        return "h2"
    elif ratio >= 1.15 or (ratio >= 1.08 and is_bold):
        return "h3"
    return "p"


def _detect_list_prefix(text: str):
    """Detect list item prefixes. Returns (type, cleaned_text) or None."""
    stripped = text.strip()
    # Bullet list: •, ●, ○, ■, –, -, *
    bullet_match = re.match(r'^[\u2022\u25cf\u25cb\u25a0\u2013\-\*]\s+(.+)', stripped)
    if bullet_match:
        return ("ul", bullet_match.group(1))
    # Numbered list: 1., 2., (1), (a), i., etc.
    num_match = re.match(r'^(?:\d+[\.\)]\s+|[a-z][\.\)]\s+|[ivxlcdm]+[\.\)]\s+)(.+)', stripped, re.IGNORECASE)
    if num_match:
        return ("ol", num_match.group(1))
    return None


def _format_span_html(text: str, flags: int) -> str:
    """Wrap text in <strong>/<em> based on PyMuPDF span flags."""
    if not text:
        return ""
    escaped = text.replace("&", "&amp;").replace("<", "&lt;").replace(">", "&gt;")
    is_bold = bool(flags & (1 << 4))
    is_italic = bool(flags & (1 << 1))
    result = escaped
    if is_bold:
        result = f"<strong>{result}</strong>"
    if is_italic:
        result = f"<em>{result}</em>"
    return result


# ── Page number detection ────────────────────────────────────────────────
_PAGE_NUM_RE = re.compile(
    r'^\s*'
    r'(?:'
    r'\d{1,4}'                           # standalone number: 1, 23, 456
    r'|[Pp]age\s+\d{1,4}'               # Page 3, page 12
    r'|\d{1,4}\s+of\s+\d{1,4}'         # 3 of 10
    r'|[Pp]age\s+\d{1,4}\s+of\s+\d+'  # Page 3 of 10
    r'|[-–—]\s*\d{1,4}\s*[-–—]'        # - 3 -, – 12 –
    r')'
    r'\s*$'
)

def _is_page_number(block: dict, page_height: float) -> bool:
    """Detect if a text block is a page number (header/footer region + matching pattern)."""
    if block.get("type") != 0:
        return False
    bbox = block.get("bbox", (0, 0, 0, 0))
    # Block must be in top 8% or bottom 8% of the page
    margin = page_height * 0.08
    in_header = bbox[1] < margin             # y0 near top
    in_footer = bbox[3] > page_height - margin  # y1 near bottom
    if not (in_header or in_footer):
        return False
    # Extract all text from the block
    text = ""
    for line in block.get("lines", []):
        for span in line.get("spans", []):
            text += span.get("text", "")
    text = text.strip()
    if not text:
        return False
    return bool(_PAGE_NUM_RE.match(text))


def _extract_table_html(table, page=None, text_flags=0) -> str:
    """Extract table to HTML with direct cell-level text extraction for accuracy.
    
    Instead of relying on table.extract() (which uses default flags internally),
    we extract text from each cell rect ourselves using page.get_text() with
    our custom flags. This ensures Hindi ligatures, whitespace, and special
    characters are preserved exactly as they appear in the PDF.
    """
    try:
        # ── Primary path: direct extraction from page ──
        if page is not None and hasattr(table, 'rows'):
            rows_data = []
            for row_obj in table.rows:
                row_cells = []
                for cell_rect in row_obj.cells:
                    if cell_rect is None:
                        row_cells.append("")  # Merged cell placeholder
                    else:
                        rect = fitz.Rect(cell_rect)
                        text = page.get_text("text", clip=rect, flags=text_flags, sort=True).strip()
                        row_cells.append(text)
                rows_data.append(row_cells)
        else:
            # Fallback: use table.extract() if page not available
            raw = table.extract()
            if not raw:
                return ""
            rows_data = [[(c or "") for c in row] for row in raw]
    except Exception as e:
        logger.warning(f"Table extraction failed: {e}")
        return ""
    
    if not rows_data:
        return ""
    
    # Drop rows where every cell is empty
    rows_data = [r for r in rows_data if any(c.strip() for c in r)]
    if not rows_data:
        return ""
    
    html = '<table><tbody>\n'
    for i, row in enumerate(rows_data):
        tag = "th" if i == 0 else "td"
        html += "  <tr>"
        for cell_text in row:
            escaped = cell_text.strip()
            escaped = escaped.replace("&", "&amp;").replace("<", "&lt;").replace(">", "&gt;")
            escaped = escaped.replace("\n", "<br>")
            html += f"<{tag}>{escaped}</{tag}>"
        html += "</tr>\n"
    html += "</tbody></table>\n"
    return html


def extract_pdf_to_html(content: bytes) -> dict:
    """
    Convert a searchable PDF to structured HTML using PyMuPDF dict-mode extraction.
    
    Pipeline:
    1. Extract all text blocks with font metadata via page.get_text("dict")
    2. Extract tables via page.find_tables()
    3. Compute median font size as body-text baseline
    4. Classify blocks as headings (h1-h3) or paragraphs based on font size ratio
    5. Detect bold/italic from span flags
    6. Detect list patterns from line prefixes
    7. Assemble clean HTML ready for TipTap editor
    """
    start_time = time.time()
    
    with fitz.open(stream=content, filetype="pdf") as doc:
        if doc.is_encrypted:
            try:
                doc.authenticate("")
            except:
                return {"html": "<p>This PDF is encrypted and cannot be imported.</p>", "title": "Encrypted PDF", "pages": 0}
        
        # Extract title from metadata
        metadata = doc.metadata or {}
        title = metadata.get("title", "").strip() or "Imported PDF"
        total_pages = len(doc)
        
        # First pass: collect all blocks from all pages to compute global median font size
        all_page_data = []
        all_blocks_flat = []
        
        # Text extraction flags — preserve whitespace AND ligatures (critical for Hindi/Devanagari)
        # This same flags value is passed to _extract_table_html for direct cell extraction
        text_flags = fitz.TEXT_PRESERVE_WHITESPACE | fitz.TEXT_PRESERVE_LIGATURES
        
        for page in doc:
            try:
                page_dict = page.get_text("dict", flags=text_flags)
                blocks = page_dict.get("blocks", [])
            except Exception as e:
                logger.warning(f"Skipping corrupt page: {e}")
                blocks = []
            
            # Extract tables for this page (if PyMuPDF version supports it)
            page_tables = []
            try:
                tables = page.find_tables()
                if tables and tables.tables:
                    page_tables = tables.tables
            except (AttributeError, Exception):
                pass  # Older PyMuPDF version without find_tables()
            
            # Get table bounding boxes to exclude table text from block processing
            table_rects = []
            for t in page_tables:
                try:
                    table_rects.append(fitz.Rect(t.bbox))
                except:
                    pass
            
            all_page_data.append({
                "blocks": blocks,
                "tables": page_tables,
                "table_rects": table_rects,
                "page_height": page.rect.height,
            })
            all_blocks_flat.extend(blocks)
        
        median_size = _compute_median_font_size(all_blocks_flat)
        
        # Second pass: convert blocks to HTML
        html_parts = []
        
        for page_idx, page_data in enumerate(all_page_data):
            blocks = page_data["blocks"]
            tables = page_data["tables"]
            table_rects = page_data["table_rects"]
            page_height = page_data["page_height"]
            page_obj = doc[page_idx]  # re-access page for direct cell text extraction
            
            # Track which tables we've already inserted
            tables_inserted = set()
            
            for block in blocks:
                if block.get("type") != 0:  # Skip image blocks
                    continue
                
                # Skip page numbers (headers/footers like "Page 3", "- 5 -", etc.)
                if _is_page_number(block, page_height):
                    continue
                
                block_bbox = fitz.Rect(block.get("bbox", (0, 0, 0, 0)))
                
                # Check if this block overlaps with any table region
                is_in_table = False
                for t_idx, t_rect in enumerate(table_rects):
                    if block_bbox.intersects(t_rect):
                        is_in_table = True
                        if t_idx not in tables_inserted:
                            tables_inserted.add(t_idx)
                            html_parts.append(_extract_table_html(tables[t_idx], page_obj, text_flags))
                        break
                
                if is_in_table:
                    continue
                
                # Process all lines in this block together
                lines = block.get("lines", [])
                if not lines:
                    continue
                
                # Get dominant font size and flags for the block (from first substantial span)
                dominant_size = median_size
                dominant_flags = 0
                for line in lines:
                    for span in line.get("spans", []):
                        if span.get("text", "").strip():
                            dominant_size = span.get("size", median_size)
                            dominant_flags = span.get("flags", 0)
                            break
                    else:
                        continue
                    break
                
                # Determine the HTML tag
                tag = _classify_heading(dominant_size, median_size, dominant_flags)
                
                # Build the inner HTML from all spans
                block_html_parts = []
                for line in lines:
                    line_parts = []
                    for span in line.get("spans", []):
                        text = span.get("text", "")
                        if not text:
                            continue
                        flags = span.get("flags", 0)
                        # For headings, don't double-wrap in bold if heading is already implied
                        if tag.startswith("h") and bool(flags & (1 << 4)):
                            formatted = text.replace("&", "&amp;").replace("<", "&lt;").replace(">", "&gt;")
                            if bool(flags & (1 << 1)):  # Still apply italic
                                formatted = f"<em>{formatted}</em>"
                        else:
                            formatted = _format_span_html(text, flags)
                        line_parts.append(formatted)
                    if line_parts:
                        block_html_parts.append("".join(line_parts))
                
                if not block_html_parts:
                    continue
                
                full_text = " ".join(block_html_parts)
                clean_text = re.sub(r'<[^>]+>', '', full_text).strip()
                
                if not clean_text:
                    continue
                
                # Check for list items
                if tag == "p":
                    # Check each line for list patterns
                    list_items = []
                    list_type = None
                    is_list = True
                    
                    for line_html in block_html_parts:
                        plain = re.sub(r'<[^>]+>', '', line_html).strip()
                        result = _detect_list_prefix(plain)
                        if result:
                            lt, cleaned = result
                            if list_type is None:
                                list_type = lt
                            elif lt != list_type:
                                is_list = False
                                break
                            # Replace the plain text prefix in the HTML
                            list_items.append(f"<li>{cleaned}</li>")
                        else:
                            is_list = False
                            break
                    
                    if is_list and list_items and list_type:
                        list_tag = list_type
                        html_parts.append(f"<{list_tag}>{''.join(list_items)}</{list_tag}>")
                        continue
                
                html_parts.append(f"<{tag}>{full_text}</{tag}>")
            
            # Insert any remaining tables that weren't matched to text blocks
            for t_idx, table in enumerate(tables):
                if t_idx not in tables_inserted:
                    html_parts.append(_extract_table_html(table, page_obj, text_flags))
            
            # Page separator (not after the last page)
            if page_idx < len(all_page_data) - 1:
                html_parts.append("<hr>")
        
        elapsed = time.time() - start_time
        final_html = "\n".join(html_parts)
        if not final_html.strip():
            final_html = "<p>No readable text found. The PDF may be scanned or image-only.</p>"
        
        logger.info(f"PDF→HTML conversion: {total_pages} pages in {elapsed:.2f}s, {len(final_html)} chars")
        
        return {
            "html": final_html,
            "title": title,
            "pages": total_pages,
            "processing_time": round(elapsed, 2),
        }


def extract_docx_to_html(content: bytes) -> dict:
    """
    Convert a DOCX file to structured HTML using python-docx.
    Preserves headings, bold, italic, underline, lists, and tables.
    """
    start_time = time.time()
    
    doc = docx.Document(io.BytesIO(content))
    title = doc.core_properties.title or "Imported Document"
    html_parts = []
    
    for para in doc.paragraphs:
        if not para.text.strip():
            continue
        
        # Determine tag from paragraph style
        style_name = (para.style.name or "").lower()
        if "heading 1" in style_name:
            tag = "h1"
        elif "heading 2" in style_name:
            tag = "h2"
        elif "heading 3" in style_name:
            tag = "h3"
        elif "heading 4" in style_name:
            tag = "h4"
        elif "list" in style_name and "bullet" in style_name:
            # Collect as list item — simplified
            run_html = _docx_runs_to_html(para.runs)
            html_parts.append(f"<ul><li>{run_html}</li></ul>")
            continue
        elif "list" in style_name:
            run_html = _docx_runs_to_html(para.runs)
            html_parts.append(f"<ol><li>{run_html}</li></ol>")
            continue
        else:
            tag = "p"
        
        run_html = _docx_runs_to_html(para.runs)
        if run_html.strip():
            html_parts.append(f"<{tag}>{run_html}</{tag}>")
    
    # Extract tables
    for table in doc.tables:
        html_parts.append("<table><tbody>")
        for i, row in enumerate(table.rows):
            cell_tag = "th" if i == 0 else "td"
            html_parts.append("  <tr>")
            for cell in row.cells:
                cell_text = cell.text.strip().replace("&", "&amp;").replace("<", "&lt;").replace(">", "&gt;")
                html_parts.append(f"    <{cell_tag}>{cell_text}</{cell_tag}>")
            html_parts.append("  </tr>")
        html_parts.append("</tbody></table>")
    
    elapsed = time.time() - start_time
    return {
        "html": "\n".join(html_parts),
        "title": title,
        "processing_time": round(elapsed, 2),
    }


def _docx_runs_to_html(runs) -> str:
    """Convert DOCX paragraph runs to HTML with inline formatting."""
    parts = []
    for run in runs:
        text = run.text
        if not text:
            continue
        escaped = text.replace("&", "&amp;").replace("<", "&lt;").replace(">", "&gt;")
        if run.bold:
            escaped = f"<strong>{escaped}</strong>"
        if run.italic:
            escaped = f"<em>{escaped}</em>"
        if run.underline:
            escaped = f"<u>{escaped}</u>"
        parts.append(escaped)
    return "".join(parts)


def extract_pptx_to_html(content: bytes) -> dict:
    """
    Convert a PPTX file to structured HTML.
    Each slide becomes a section with its text and tables.
    """
    start_time = time.time()
    
    prs = pptx.Presentation(io.BytesIO(content))
    html_parts = []
    
    for i, slide in enumerate(prs.slides):
        slide_parts = []
        
        for shape in slide.shapes:
            if hasattr(shape, "text_frame"):
                for para in shape.text_frame.paragraphs:
                    # Build HTML from runs to preserve bold/italic
                    run_parts = []
                    for run in para.runs:
                        t = run.text
                        if not t:
                            continue
                        t = t.replace("&", "&amp;").replace("<", "&lt;").replace(">", "&gt;")
                        if run.font.bold:
                            t = f"<strong>{t}</strong>"
                        if run.font.italic:
                            t = f"<em>{t}</em>"
                        run_parts.append(t)
                    text = "".join(run_parts)
                    if not text.strip():
                        continue
                    level = para.level
                    if level == 0 and not slide_parts:
                        slide_parts.append(f"<h2>{text}</h2>")
                    elif level == 0:
                        slide_parts.append(f"<p>{text}</p>")
                    else:
                        slide_parts.append(f"<ul><li>{text}</li></ul>")
            
            if shape.has_table:
                table_html = "<table><tbody>"
                for r_idx, row in enumerate(shape.table.rows):
                    cell_tag = "th" if r_idx == 0 else "td"
                    table_html += "<tr>"
                    for cell in row.cells:
                        cell_text = cell.text.strip().replace("&", "&amp;").replace("<", "&lt;").replace(">", "&gt;")
                        table_html += f"<{cell_tag}>{cell_text}</{cell_tag}>"
                    table_html += "</tr>"
                table_html += "</tbody></table>"
                slide_parts.append(table_html)
        
        if slide_parts:
            html_parts.append(f"<!-- Slide {i+1} -->")
            html_parts.extend(slide_parts)
            if i < len(prs.slides) - 1:
                html_parts.append("<hr>")
    
    elapsed = time.time() - start_time
    return {
        "html": "\n".join(html_parts),
        "title": "Imported Presentation",
        "slides": len(prs.slides),
        "processing_time": round(elapsed, 2),
    }


# ── Import Endpoints ─────────────────────────────────────────────────────

@app.post("/api/pdf-to-html")
async def pdf_to_html_endpoint(file: UploadFile = File(...)):
    """
    Convert a searchable PDF to structured HTML with formatting preservation.
    Returns { html, title, pages, processing_time }.
    """
    if not file.filename.lower().endswith('.pdf'):
        raise HTTPException(status_code=400, detail="Only PDF files are accepted")
    
    content = await file.read()
    if len(content) > Config.MAX_FILE_SIZE_MB * 1024 * 1024:
        raise HTTPException(status_code=413, detail=f"File exceeds {Config.MAX_FILE_SIZE_MB}MB limit")
    
    loop = asyncio.get_event_loop()
    try:
        result = await loop.run_in_executor(executor, extract_pdf_to_html, content)
        return result
    except Exception as e:
        logger.error(f"PDF-to-HTML conversion failed: {e}")
        raise HTTPException(status_code=500, detail=f"Conversion failed: {str(e)}")


@app.post("/api/docx-to-html")
async def docx_to_html_endpoint(file: UploadFile = File(...)):
    """
    Convert a DOCX file to structured HTML with formatting preservation.
    Returns { html, title, processing_time }.
    """
    if not file.filename.lower().endswith('.docx'):
        raise HTTPException(status_code=400, detail="Only DOCX files are accepted")
    
    content = await file.read()
    if len(content) > Config.MAX_FILE_SIZE_MB * 1024 * 1024:
        raise HTTPException(status_code=413, detail=f"File exceeds {Config.MAX_FILE_SIZE_MB}MB limit")
    
    loop = asyncio.get_event_loop()
    try:
        result = await loop.run_in_executor(executor, extract_docx_to_html, content)
        return result
    except Exception as e:
        logger.error(f"DOCX-to-HTML conversion failed: {e}")
        raise HTTPException(status_code=500, detail=f"Conversion failed: {str(e)}")


@app.post("/api/pptx-to-html")
async def pptx_to_html_endpoint(file: UploadFile = File(...)):
    """
    Convert a PPTX file to structured HTML.
    Returns { html, title, slides, processing_time }.
    """
    if not file.filename.lower().endswith('.pptx'):
        raise HTTPException(status_code=400, detail="Only PPTX files are accepted")
    
    content = await file.read()
    if len(content) > Config.MAX_FILE_SIZE_MB * 1024 * 1024:
        raise HTTPException(status_code=413, detail=f"File exceeds {Config.MAX_FILE_SIZE_MB}MB limit")
    
    loop = asyncio.get_event_loop()
    try:
        result = await loop.run_in_executor(executor, extract_pptx_to_html, content)
        return result
    except Exception as e:
        logger.error(f"PPTX-to-HTML conversion failed: {e}")
        raise HTTPException(status_code=500, detail=f"Conversion failed: {str(e)}")


# ==================== MAIN ====================

if __name__ == "__main__":
    import sys
    
    port = int(os.getenv("PORT", 7860))
    workers = int(os.getenv("WORKERS", 1))
    host = os.getenv("HOST", "0.0.0.0")
    
    logger.info(f"Starting NeuralStream Production Extractor on {host}:{port}")
    logger.info(f"Worker processes: {workers}")
    logger.info(f"File size limit: {Config.MAX_FILE_SIZE_MB}MB")
    logger.info(f"ZIP processing depth: {Config.MAX_ZIP_DEPTH}")
    logger.info(f"OCR Enabled: {Config.ENABLE_OCR}")
    logger.info(f"OCR Language: {Config.OCR_LANGUAGE}")
    logger.info(f"Supported file types: 50+ formats")
    
    if '--dev' in sys.argv:
        uvicorn.run("app:app", host="127.0.0.1", port=port, reload=True)
    else:
        uvicorn.run(
            "app:app",
            host=host,
            port=port,
            workers=workers,
            log_level="info",
            access_log=True,
            loop="asyncio"
        )