app/metadata/extractor.py

"""PDF metadata extraction using PyMuPDF (fitz) with pypdf fallback."""

from __future__ import annotations

import re
from pathlib import Path

from loguru import logger

from app.models.document import Category, DocumentMeta, Language, PageMeta
from app.utils.file import file_sha256

# Regex pattern to identify Khmer Unicode range U+1780 to U+17FF and U+19E0 to U+19FF
KHMER_CHAR_PATTERN = re.compile(r"[\u1780-\u17ff\u19e0-\u19ff]")
ENGLISH_CHAR_PATTERN = re.compile(r"[a-zA-Z]")


def _detect_language(text: str) -> Language:
    """Heuristic language detection based on character occurrences."""
    if not text.strip():
        return Language.unknown

    has_khmer = bool(KHMER_CHAR_PATTERN.search(text))
    has_english = bool(ENGLISH_CHAR_PATTERN.search(text))

    if has_khmer and has_english:
        return Language.mixed
    elif has_khmer:
        return Language.km
    elif has_english:
        return Language.en
    return Language.unknown


class MetadataExtractor:
    """
    Extracts structural, textual, and spatial metadata from PDF documents.
    Primary engine: PyMuPDF (fitz).
    Fallback engine: pypdf.
    """

    def __init__(self, *, fallback_to_pypdf: bool = True) -> None:
        self._fallback_to_pypdf = fallback_to_pypdf

    def extract(self, pdf_path: Path, category_hint: str = "other") -> DocumentMeta:
        """Extract metadata from a PDF file."""
        pdf_path = Path(pdf_path).resolve()
        if not pdf_path.exists():
            raise FileNotFoundError(f"PDF file not found: {pdf_path}")

        file_size_bytes = pdf_path.stat().st_size
        sha256_hash = file_sha256(pdf_path)

        try:
            return self._extract_pymupdf(pdf_path, file_size_bytes, sha256_hash, category_hint)
        except Exception as exc:
            logger.warning("PyMuPDF failed on {}: {}", pdf_path.name, exc)
            if self._fallback_to_pypdf:
                logger.info("Attempting pypdf fallback for {}", pdf_path.name)
                return self._extract_pypdf(pdf_path, file_size_bytes, sha256_hash, category_hint)
            raise

    def _extract_pymupdf(
        self,
        pdf_path: Path,
        file_size_bytes: int,
        sha256_hash: str,
        category_hint: str,
    ) -> DocumentMeta:
        import fitz  # PyMuPDF

        doc = fitz.open(pdf_path)
        pages_count = len(doc)
        pages_meta: list[PageMeta] = []

        total_text = ""
        has_images = False
        has_tables = False
        has_text_layer = False

        for idx, page in enumerate(doc, start=1):
            rect = page.rect
            width, height = rect.width, rect.height
            text = page.get_text("text")
            char_count = len(text)
            total_text += text + " "

            images = page.get_images()
            img_count = len(images)
            if img_count > 0:
                has_images = True

            page_has_text = char_count > 10
            if page_has_text:
                has_text_layer = True

            # Basic table heuristic via PyMuPDF find_tables if available
            try:
                tables = page.find_tables()
                if tables and len(tables.tables) > 0:
                    has_tables = True
            except AttributeError:
                pass

            pages_meta.append(
                PageMeta(
                    page_number=idx,
                    width_pt=float(width),
                    height_pt=float(height),
                    text_char_count=char_count,
                    image_count=img_count,
                    has_text_layer=page_has_text,
                )
            )

        doc.close()

        lang = _detect_language(total_text)
        native_pdf = has_text_layer
        scanned = not has_text_layer and has_images

        cat_val = category_hint if category_hint in Category.__members__ else "other"

        return DocumentMeta(
            filename=pdf_path.name,
            language=lang,
            category=Category(cat_val),
            pages=pages_count,
            file_size_bytes=file_size_bytes,
            native_pdf=native_pdf,
            scanned=scanned,
            has_tables=has_tables,
            has_images=has_images,
            has_header=False,
            has_footer=False,
            sha256=sha256_hash,
            pdf_path=str(pdf_path),
            pages_meta=pages_meta,
        )

    def _extract_pypdf(
        self,
        pdf_path: Path,
        file_size_bytes: int,
        sha256_hash: str,
        category_hint: str,
    ) -> DocumentMeta:
        from pypdf import PdfReader

        reader = PdfReader(pdf_path)
        pages_count = len(reader.pages)
        pages_meta: list[PageMeta] = []

        total_text = ""
        has_images = False
        has_text_layer = False

        for idx, page in enumerate(reader.pages, start=1):
            box = page.mediabox
            width = float(box.width)
            height = float(box.height)
            text = page.extract_text() or ""
            char_count = len(text)
            total_text += text + " "

            img_count = len(page.images)
            if img_count > 0:
                has_images = True

            page_has_text = char_count > 10
            if page_has_text:
                has_text_layer = True

            pages_meta.append(
                PageMeta(
                    page_number=idx,
                    width_pt=width,
                    height_pt=height,
                    text_char_count=char_count,
                    image_count=img_count,
                    has_text_layer=page_has_text,
                )
            )

        lang = _detect_language(total_text)
        native_pdf = has_text_layer
        scanned = not has_text_layer and has_images
        cat_val = category_hint if category_hint in Category.__members__ else "other"

        return DocumentMeta(
            filename=pdf_path.name,
            language=lang,
            category=Category(cat_val),
            pages=pages_count,
            file_size_bytes=file_size_bytes,
            native_pdf=native_pdf,
            scanned=scanned,
            has_tables=False,
            has_images=has_images,
            has_header=False,
            has_footer=False,
            sha256=sha256_hash,
            pdf_path=str(pdf_path),
            pages_meta=pages_meta,
        )