kreuzberg 3.7.0__py3-none-any.whl → 3.8.0__py3-none-any.whl

kreuzberg/_extractors/_spread_sheet.py

@@ -13,7 +13,7 @@ from python_calamine import CalamineWorkbook
 
 from kreuzberg._extractors._base import Extractor
 from kreuzberg._mime_types import MARKDOWN_MIME_TYPE, SPREADSHEET_MIME_TYPES
-from kreuzberg._types import ExtractionResult
+from kreuzberg._types import ExtractionResult, Metadata
 from kreuzberg._utils._string import normalize_spaces
 from kreuzberg._utils._sync import run_sync, run_taskgroup
 from kreuzberg._utils._tmp import create_temp_file
@@ -45,9 +45,14 @@ class SpreadSheetExtractor(Extractor):
             try:
                 results: list[str] = await run_taskgroup(*tasks)
 
-                return ExtractionResult(
-                    content="\n\n".join(results), mime_type=MARKDOWN_MIME_TYPE, metadata={}, chunks=[]
+                result = ExtractionResult(
+                    content="\n\n".join(results),
+                    mime_type=MARKDOWN_MIME_TYPE,
+                    metadata=self._extract_spreadsheet_metadata(workbook),
+                    chunks=[],
                 )
+
+                return self._apply_quality_processing(result)
             except ExceptionGroup as eg:
                 raise ParsingError(
                     "Failed to extract file data",
@@ -87,7 +92,14 @@ class SpreadSheetExtractor(Extractor):
                 sheet_text = self._convert_sheet_to_text_sync(workbook, sheet_name)
                 results.append(sheet_text)
 
-            return ExtractionResult(content="\n\n".join(results), mime_type=MARKDOWN_MIME_TYPE, metadata={}, chunks=[])
+            result = ExtractionResult(
+                content="\n\n".join(results),
+                mime_type=MARKDOWN_MIME_TYPE,
+                metadata=self._extract_spreadsheet_metadata(workbook),
+                chunks=[],
+            )
+
+            return self._apply_quality_processing(result)
         except Exception as e:
             raise ParsingError(
                 "Failed to extract file data",
@@ -181,3 +193,166 @@ class SpreadSheetExtractor(Extractor):
             result = "\n".join(markdown_lines)
 
         return f"## {sheet_name}\n\n{normalize_spaces(result)}"
+
+    def _enhance_sheet_with_table_data(self, workbook: CalamineWorkbook, sheet_name: str) -> str:
+        """Enhanced sheet processing with better table structure preservation."""
+        try:
+            # pandas is optional dependency
+            import pandas as pd
+
+            from kreuzberg._utils._table import enhance_table_markdown
+
+            sheet = workbook.get_sheet_by_name(sheet_name)
+            data = sheet.to_python()
+
+            if not data or not any(row for row in data):
+                return f"## {sheet_name}\n\n*Empty sheet*"
+
+            # Convert to DataFrame
+            df = pd.DataFrame(data)
+
+            # Clean up empty rows and columns
+            df = df.dropna(how="all").dropna(axis=1, how="all")
+
+            if df.empty:
+                return f"## {sheet_name}\n\n*No data*"
+
+            # Create a mock TableData for enhanced formatting
+            from PIL import Image
+
+            from kreuzberg._types import TableData
+
+            # Create a 1x1 transparent image as placeholder
+            placeholder_image = Image.new("RGBA", (1, 1), (0, 0, 0, 0))
+            mock_table: TableData = {"df": df, "text": "", "page_number": 0, "cropped_image": placeholder_image}
+
+            enhanced_markdown = enhance_table_markdown(mock_table)
+            return f"## {sheet_name}\n\n{enhanced_markdown}"
+
+        except (ImportError, AttributeError, ValueError):
+            # Fallback to original method if pandas/table enhancement fails
+            return self._convert_sheet_to_text_sync(workbook, sheet_name)
+
+    @staticmethod
+    def _extract_spreadsheet_metadata(workbook: CalamineWorkbook) -> Metadata:
+        """Extract metadata from spreadsheet using python-calamine.
+
+        Args:
+            workbook: CalamineWorkbook instance
+
+        Returns:
+            Metadata dict using existing metadata keys where possible
+        """
+        metadata: Metadata = {}
+
+        # Extract basic document properties
+        SpreadSheetExtractor._extract_document_properties(workbook, metadata)
+
+        # Add structural information
+        SpreadSheetExtractor._add_structure_info(workbook, metadata)
+
+        # Analyze content complexity
+        SpreadSheetExtractor._analyze_content_complexity(workbook, metadata)
+
+        return metadata
+
+    @staticmethod
+    def _extract_document_properties(workbook: CalamineWorkbook, metadata: Metadata) -> None:
+        """Extract basic document properties from workbook."""
+        with contextlib.suppress(AttributeError, Exception):
+            if not (hasattr(workbook, "metadata") and workbook.metadata):
+                return
+
+            props = workbook.metadata
+
+            # Basic properties mapping
+            property_mapping = {
+                "title": "title",
+                "author": "authors",  # Convert to list
+                "subject": "subject",
+                "comments": "comments",
+                "keywords": "keywords",  # Process separately
+                "category": "categories",  # Convert to list
+                "company": "organization",
+                "manager": "modified_by",
+            }
+
+            for prop_name, meta_key in property_mapping.items():
+                if hasattr(props, prop_name) and (value := getattr(props, prop_name)):
+                    if meta_key in ("authors", "categories"):
+                        metadata[meta_key] = [value]  # type: ignore[literal-required]
+                    elif meta_key == "keywords":
+                        keywords = [k.strip() for k in value.replace(";", ",").split(",") if k.strip()]
+                        if keywords:
+                            metadata[meta_key] = keywords  # type: ignore[literal-required]
+                    else:
+                        metadata[meta_key] = value  # type: ignore[literal-required]
+
+            # Handle dates separately
+            SpreadSheetExtractor._extract_date_properties(props, metadata)
+
+    @staticmethod
+    def _extract_date_properties(props: Any, metadata: Metadata) -> None:
+        """Extract and format date properties."""
+        date_mapping = {"created": "created_at", "modified": "modified_at"}
+
+        for prop_name, meta_key in date_mapping.items():
+            if hasattr(props, prop_name) and (date_value := getattr(props, prop_name)):
+                with contextlib.suppress(Exception):
+                    if hasattr(date_value, "isoformat"):
+                        metadata[meta_key] = date_value.isoformat()  # type: ignore[literal-required]
+                    else:
+                        metadata[meta_key] = str(date_value)  # type: ignore[literal-required]
+
+    @staticmethod
+    def _add_structure_info(workbook: CalamineWorkbook, metadata: Metadata) -> None:
+        """Add structural information about the spreadsheet."""
+        if not (hasattr(workbook, "sheet_names") and workbook.sheet_names):
+            return
+
+        sheet_count = len(workbook.sheet_names)
+        structure_info = f"Spreadsheet with {sheet_count} sheet{'s' if sheet_count != 1 else ''}"
+
+        # Don't list too many sheet names (magic number made constant)
+        max_sheet_names_to_list = 5
+        if sheet_count <= max_sheet_names_to_list:
+            structure_info += f": {', '.join(workbook.sheet_names)}"
+
+        metadata["description"] = structure_info
+
+    @staticmethod
+    def _analyze_content_complexity(workbook: CalamineWorkbook, metadata: Metadata) -> None:
+        """Analyze spreadsheet content for complexity indicators."""
+        with contextlib.suppress(Exception):
+            has_formulas = False
+            total_cells = 0
+
+            # Check only first few sheets for performance
+            max_sheets_to_check = 3
+            max_rows_to_check = 50
+
+            for sheet_name in workbook.sheet_names[:max_sheets_to_check]:
+                with contextlib.suppress(Exception):
+                    sheet = workbook.get_sheet_by_name(sheet_name)
+                    data = sheet.to_python()
+
+                    for row in data[:max_rows_to_check]:
+                        if not row:  # Skip empty rows
+                            continue
+
+                        total_cells += sum(1 for cell in row if cell is not None and str(cell).strip())
+
+                        # Check for formulas (simple heuristic)
+                        if any(isinstance(cell, str) and cell.startswith("=") for cell in row):
+                            has_formulas = True
+                            break
+
+            # Build summary
+            summary_parts = []
+            if total_cells > 0:
+                summary_parts.append(f"Contains {total_cells}+ data cells")
+            if has_formulas:
+                summary_parts.append("includes formulas")
+
+            if summary_parts and "summary" not in metadata:
+                metadata["summary"] = f"Spreadsheet that {', '.join(summary_parts)}."

kreuzberg/_extractors/_structured.py

@@ -0,0 +1,148 @@
+from __future__ import annotations
+
+import json
+from typing import TYPE_CHECKING, Any, ClassVar
+
+from anyio import Path as AsyncPath
+
+from kreuzberg._extractors._base import Extractor
+from kreuzberg._mime_types import JSON_MIME_TYPE, PLAIN_TEXT_MIME_TYPE, TOML_MIME_TYPE, YAML_MIME_TYPE
+from kreuzberg._types import ExtractionResult, normalize_metadata
+from kreuzberg._utils._string import normalize_spaces, safe_decode
+from kreuzberg._utils._sync import run_sync
+
+if TYPE_CHECKING:
+    from pathlib import Path
+
+
+class StructuredDataExtractor(Extractor):
+    SUPPORTED_MIME_TYPES: ClassVar[set[str]] = {
+        JSON_MIME_TYPE,
+        "text/json",
+        YAML_MIME_TYPE,
+        "text/yaml",
+        "text/x-yaml",
+        "application/yaml",
+        TOML_MIME_TYPE,
+        "text/toml",
+    }
+
+    async def extract_bytes_async(self, content: bytes) -> ExtractionResult:
+        return await run_sync(self.extract_bytes_sync, content)
+
+    async def extract_path_async(self, path: Path) -> ExtractionResult:
+        content = await AsyncPath(path).read_bytes()
+        return await self.extract_bytes_async(content)
+
+    def extract_bytes_sync(self, content: bytes) -> ExtractionResult:
+        text_content = safe_decode(content)
+
+        try:
+            if self.mime_type in {JSON_MIME_TYPE, "text/json"}:
+                data = json.loads(text_content)
+            elif self.mime_type in {TOML_MIME_TYPE, "text/toml"}:
+                try:
+                    import tomllib  # type: ignore[import-not-found]
+                except ImportError:
+                    try:
+                        import tomli as tomllib  # type: ignore[import-not-found]
+                    except ImportError:
+                        return ExtractionResult(
+                            content=normalize_spaces(text_content),
+                            mime_type=PLAIN_TEXT_MIME_TYPE,
+                            metadata={"warning": "tomllib/tomli not available, returning raw text"},
+                            chunks=[],
+                        )
+                data = tomllib.loads(text_content)
+            else:
+                try:
+                    import yaml
+
+                    data = yaml.safe_load(text_content)
+                except ImportError:
+                    return ExtractionResult(
+                        content=normalize_spaces(text_content),
+                        mime_type=PLAIN_TEXT_MIME_TYPE,
+                        metadata={"warning": "PyYAML not available, returning raw text"},
+                        chunks=[],
+                    )
+
+            text_parts: list[str] = []
+            metadata: dict[str, Any] = {}
+
+            if isinstance(data, dict):
+                text_parts.extend(self._extract_from_dict(data, metadata))
+            elif isinstance(data, list):
+                text_parts.extend(self._extract_from_list(data, metadata))
+            else:
+                text_parts.append(str(data))
+
+            combined_text = "\n".join(text_parts) if text_parts else text_content
+
+            return ExtractionResult(
+                content=normalize_spaces(combined_text),
+                mime_type=PLAIN_TEXT_MIME_TYPE,
+                metadata=normalize_metadata(metadata),
+                chunks=[],
+            )
+
+        except (ValueError, TypeError, KeyError, AttributeError, UnicodeDecodeError) as e:
+            return ExtractionResult(
+                content=normalize_spaces(text_content),
+                mime_type=PLAIN_TEXT_MIME_TYPE,
+                metadata={"parse_error": str(e)},
+                chunks=[],
+            )
+
+    def extract_path_sync(self, path: Path) -> ExtractionResult:
+        content = path.read_bytes()
+        return self.extract_bytes_sync(content)
+
+    def _extract_from_dict(self, data: dict[str, Any], metadata: dict[str, Any], prefix: str = "") -> list[str]:
+        text_parts = []
+
+        for key, value in data.items():
+            full_key = f"{prefix}.{key}" if prefix else key
+
+            if isinstance(value, str) and value.strip():
+                text_parts.append(f"{full_key}: {value}")
+
+                if any(
+                    text_field in key.lower()
+                    for text_field in ["title", "name", "subject", "description", "content", "body", "text", "message"]
+                ):
+                    metadata[full_key] = value
+
+            elif isinstance(value, (int, float, bool)):
+                text_parts.append(f"{full_key}: {value}")
+
+            elif isinstance(value, dict):
+                text_parts.extend(self._extract_from_dict(value, metadata, full_key))
+
+            elif isinstance(value, list):
+                text_parts.extend(self._extract_from_list(value, metadata, full_key))
+
+            elif value is not None:
+                text_parts.append(f"{full_key}: {value!s}")
+
+        return text_parts
+
+    def _extract_from_list(self, data: list[Any], metadata: dict[str, Any], prefix: str = "") -> list[str]:
+        text_parts = []
+
+        for i, item in enumerate(data):
+            item_key = f"{prefix}[{i}]" if prefix else f"item_{i}"
+
+            if isinstance(item, str) and item.strip():
+                text_parts.append(f"{item_key}: {item}")
+
+            elif isinstance(item, dict):
+                text_parts.extend(self._extract_from_dict(item, metadata, item_key))
+
+            elif isinstance(item, list):
+                text_parts.extend(self._extract_from_list(item, metadata, item_key))
+
+            elif item is not None:
+                text_parts.append(f"{item_key}: {item!s}")
+
+        return text_parts

kreuzberg/_gmft.py

@@ -196,7 +196,7 @@ async def extract_tables(  # noqa: PLR0915
 
     try:
         if use_isolated_process:
-            from kreuzberg._multiprocessing.gmft_isolated import extract_tables_isolated_async
+            from kreuzberg._multiprocessing import extract_tables_isolated_async
 
             result = await extract_tables_isolated_async(file_path, config)
 
@@ -314,7 +314,7 @@ def extract_tables_sync(
         return cached_result  # type: ignore[no-any-return]
 
     if use_isolated_process:
-        from kreuzberg._multiprocessing.gmft_isolated import extract_tables_isolated
+        from kreuzberg._multiprocessing import extract_tables_isolated
 
         result = extract_tables_isolated(file_path, config)
 

kreuzberg/_mime_types.py

@@ -17,6 +17,12 @@ PLAIN_TEXT_MIME_TYPE: Final = "text/plain"
 POWER_POINT_MIME_TYPE: Final = "application/vnd.openxmlformats-officedocument.presentationml.presentation"
 DOCX_MIME_TYPE: Final = "application/vnd.openxmlformats-officedocument.wordprocessingml.document"
 
+EML_MIME_TYPE: Final = "message/rfc822"
+MSG_MIME_TYPE: Final = "application/vnd.ms-outlook"
+JSON_MIME_TYPE: Final = "application/json"
+YAML_MIME_TYPE: Final = "application/x-yaml"
+TOML_MIME_TYPE: Final = "application/toml"
+
 EXCEL_MIME_TYPE: Final = "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet"
 EXCEL_BINARY_MIME_TYPE: Final = "application/vnd.ms-excel"
 EXCEL_MACRO_MIME_TYPE: Final = "application/vnd.ms-excel.sheet.macroEnabled.12"
@@ -127,6 +133,12 @@ EXT_TO_MIME_TYPE: Final[Mapping[str, str]] = {
     ".org": "text/x-org",
     ".epub": "application/epub+zip",
     ".rtf": "application/rtf",
+    ".eml": EML_MIME_TYPE,
+    ".msg": MSG_MIME_TYPE,
+    ".json": JSON_MIME_TYPE,
+    ".yaml": YAML_MIME_TYPE,
+    ".yml": YAML_MIME_TYPE,
+    ".toml": TOML_MIME_TYPE,
     ".odt": "application/vnd.oasis.opendocument.text",
     ".docx": DOCX_MIME_TYPE,
     ".bib": "application/x-bibtex",
@@ -139,7 +151,21 @@ SUPPORTED_MIME_TYPES: Final[set[str]] = (
     | IMAGE_MIME_TYPES
     | PANDOC_SUPPORTED_MIME_TYPES
     | SPREADSHEET_MIME_TYPES
-    | {PDF_MIME_TYPE, POWER_POINT_MIME_TYPE, HTML_MIME_TYPE}
+    | {
+        PDF_MIME_TYPE,
+        POWER_POINT_MIME_TYPE,
+        HTML_MIME_TYPE,
+        EML_MIME_TYPE,
+        MSG_MIME_TYPE,
+        JSON_MIME_TYPE,
+        YAML_MIME_TYPE,
+        TOML_MIME_TYPE,
+        "text/json",
+        "text/yaml",
+        "text/x-yaml",
+        "application/yaml",
+        "text/toml",
+    }
 )
 
 

kreuzberg/_multiprocessing/__init__.py

@@ -1,6 +1,5 @@
 """Multiprocessing utilities for kreuzberg."""
 
-from .process_manager import ProcessPoolManager
-from .tesseract_pool import TesseractProcessPool
+from .gmft_isolated import extract_tables_isolated, extract_tables_isolated_async
 
-__all__ = ["ProcessPoolManager", "TesseractProcessPool"]
+__all__ = ["extract_tables_isolated", "extract_tables_isolated_async"]

kreuzberg/_ocr/__init__.py

@@ -4,9 +4,39 @@ from typing import Any
 from kreuzberg._ocr._base import OCRBackend
 from kreuzberg._ocr._easyocr import EasyOCRBackend
 from kreuzberg._ocr._paddleocr import PaddleBackend
+from kreuzberg._ocr._pool import TesseractProcessPool
+from kreuzberg._ocr._sync import (
+    process_batch_images_process_pool,
+    process_batch_images_sync,
+    process_batch_images_threaded,
+    process_image_bytes_easyocr_sync,
+    process_image_bytes_paddleocr_sync,
+    process_image_bytes_tesseract_sync,
+    process_image_easyocr_sync,
+    process_image_paddleocr_sync,
+    process_image_tesseract_sync,
+)
 from kreuzberg._ocr._tesseract import TesseractBackend
 from kreuzberg._types import OcrBackendType
 
+__all__ = [
+    "EasyOCRBackend",
+    "OCRBackend",
+    "PaddleBackend",
+    "TesseractBackend",
+    "TesseractProcessPool",
+    "get_ocr_backend",
+    "process_batch_images_process_pool",
+    "process_batch_images_sync",
+    "process_batch_images_threaded",
+    "process_image_bytes_easyocr_sync",
+    "process_image_bytes_paddleocr_sync",
+    "process_image_bytes_tesseract_sync",
+    "process_image_easyocr_sync",
+    "process_image_paddleocr_sync",
+    "process_image_tesseract_sync",
+]
+
 
 @lru_cache
 def get_ocr_backend(backend: OcrBackendType) -> OCRBackend[Any]:

kreuzberg/{_multiprocessing/tesseract_pool.py → _ocr/_pool.py}

@@ -1,7 +1,8 @@
-"""Tesseract process pool for parallel OCR processing."""
+"""Process pools for parallel OCR processing."""
 
 from __future__ import annotations
 
+from pathlib import Path
 from typing import TYPE_CHECKING, Any
 
 from PIL import Image
@@ -9,14 +10,11 @@ from typing_extensions import Self
 
 from kreuzberg._ocr._tesseract import TesseractConfig
 from kreuzberg._types import ExtractionResult
-
-from .process_manager import ProcessPoolManager
+from kreuzberg._utils._process_pool import ProcessPoolManager
 
 if TYPE_CHECKING:
     import types
 
-from pathlib import Path
-
 
 def _process_image_with_tesseract(
     image_path: str,