xgen-doc2chunk 0.1.0__py3-none-any.whl → 0.1.2__py3-none-any.whl

xgen_doc2chunk/chunking/table_parser.py

@@ -0,0 +1,172 @@
+# chunking_helper/table_parser.py
+"""
+Table Parser - HTML table parsing functions
+
+Main Features:
+- HTML table parsing and structure analysis
+- Cell span information extraction (rowspan, colspan)
+- Table complexity analysis
+"""
+import logging
+import re
+from typing import Dict, List, Optional, Tuple
+
+from xgen_doc2chunk.chunking.constants import ParsedTable, TableRow
+
+logger = logging.getLogger("document-processor")
+
+
+def parse_html_table(table_html: str) -> Optional[ParsedTable]:
+    """
+    Parse an HTML table and extract structured information.
+
+    Args:
+        table_html: HTML table string
+
+    Returns:
+        ParsedTable object or None (if parsing fails)
+    """
+    try:
+        # Extract rows
+        row_pattern = r'<tr[^>]*>(.*?)</tr>'
+        row_matches = re.findall(row_pattern, table_html, re.DOTALL | re.IGNORECASE)
+
+        if not row_matches:
+            logger.debug("No rows found in table")
+            return None
+
+        header_rows: List[TableRow] = []
+        data_rows: List[TableRow] = []
+        max_cols = 0
+
+        for row_content in row_matches:
+            # Extract cells
+            th_cells = re.findall(r'<th[^>]*>(.*?)</th>', row_content, re.DOTALL | re.IGNORECASE)
+            td_cells = re.findall(r'<td[^>]*>(.*?)</td>', row_content, re.DOTALL | re.IGNORECASE)
+
+            is_header = len(th_cells) > 0 and len(td_cells) == 0
+            cell_count = len(th_cells) if is_header else len(td_cells)
+            max_cols = max(max_cols, cell_count)
+
+            # Reconstruct original row HTML
+            row_html = f"<tr>{row_content}</tr>"
+            row_length = len(row_html)
+
+            table_row = TableRow(
+                html=row_html,
+                is_header=is_header,
+                cell_count=cell_count,
+                char_length=row_length
+            )
+
+            if is_header and not data_rows:
+                # Header row before any data rows
+                header_rows.append(table_row)
+            else:
+                data_rows.append(table_row)
+
+        # Build header HTML
+        if header_rows:
+            header_html = "\n".join(row.html for row in header_rows)
+            header_size = sum(row.char_length for row in header_rows) + len(header_rows)  # Including newlines
+        else:
+            header_html = ""
+            header_size = 0
+
+        return ParsedTable(
+            header_rows=header_rows,
+            data_rows=data_rows,
+            total_cols=max_cols,
+            original_html=table_html,
+            header_html=header_html,
+            header_size=header_size
+        )
+
+    except Exception as e:
+        logger.warning(f"Failed to parse HTML table: {e}")
+        return None
+
+
+def extract_cell_spans(row_html: str) -> List[Tuple[int, int]]:
+    """
+    Extract rowspan/colspan information from cells in a row.
+
+    Args:
+        row_html: Row HTML
+
+    Returns:
+        [(rowspan, colspan), ...] list
+    """
+    spans = []
+
+    # Find th and td cells
+    cell_pattern = r'<(th|td)([^>]*)>'
+
+    for match in re.finditer(cell_pattern, row_html, re.IGNORECASE):
+        attrs = match.group(2)
+
+        # Extract rowspan
+        rowspan_match = re.search(r'rowspan=["\']?(\d+)["\']?', attrs, re.IGNORECASE)
+        rowspan = int(rowspan_match.group(1)) if rowspan_match else 1
+
+        # Extract colspan
+        colspan_match = re.search(r'colspan=["\']?(\d+)["\']?', attrs, re.IGNORECASE)
+        colspan = int(colspan_match.group(1)) if colspan_match else 1
+
+        spans.append((rowspan, colspan))
+
+    return spans
+
+
+def extract_cell_spans_with_positions(row_html: str) -> Dict[int, int]:
+    """
+    Extract rowspan information by column position from a row (considering colspan).
+
+    Args:
+        row_html: Row HTML
+
+    Returns:
+        {column_position: rowspan} dictionary (only cells with rowspan > 1)
+    """
+    spans: Dict[int, int] = {}
+    cell_pattern = r'<(th|td)([^>]*)>'
+
+    current_col = 0
+    for match in re.finditer(cell_pattern, row_html, re.IGNORECASE):
+        attrs = match.group(2)
+
+        # Extract rowspan
+        rowspan_match = re.search(r'rowspan=["\']?(\d+)["\']?', attrs, re.IGNORECASE)
+        rowspan = int(rowspan_match.group(1)) if rowspan_match else 1
+
+        # Extract colspan
+        colspan_match = re.search(r'colspan=["\']?(\d+)["\']?', attrs, re.IGNORECASE)
+        colspan = int(colspan_match.group(1)) if colspan_match else 1
+
+        if rowspan > 1:
+            spans[current_col] = rowspan
+
+        current_col += colspan
+
+    return spans
+
+
+def has_complex_spans(table_html: str) -> bool:
+    """
+    Check if a table has complex rowspan.
+    (colspan does not affect row splitting, only rowspan is problematic)
+
+    Args:
+        table_html: Table HTML
+
+    Returns:
+        True if there are cells with rowspan > 1
+    """
+    rowspan_pattern = r'rowspan=["\']?(\d+)["\']?'
+    matches = re.findall(rowspan_pattern, table_html, re.IGNORECASE)
+
+    for val in matches:
+        if int(val) > 1:
+            return True
+
+    return False

xgen_doc2chunk/chunking/text_chunker.py

@@ -0,0 +1,443 @@
+# chunking_helper/text_chunker.py
+"""
+Text Chunker - Text chunking functionality
+
+Main Features:
+- Plain text chunking
+- Table-free text chunking
+- Row-preserving chunking (for tables)
+- Code text chunking
+- Markdown table support with NO overlap
+"""
+import logging
+import re
+from typing import Any, List, Optional, Tuple
+
+from langchain_text_splitters import RecursiveCharacterTextSplitter
+
+from xgen_doc2chunk.chunking.constants import (
+    LANGCHAIN_CODE_LANGUAGE_MAP, HTML_TABLE_PATTERN, MARKDOWN_TABLE_PATTERN
+)
+
+logger = logging.getLogger("document-processor")
+
+
+def chunk_plain_text(text: str, chunk_size: int, chunk_overlap: int) -> List[str]:
+    """
+    Chunk plain text using RecursiveCharacterTextSplitter.
+    """
+    if not text or not text.strip():
+        return []
+
+    splitter = RecursiveCharacterTextSplitter(
+        chunk_size=chunk_size,
+        chunk_overlap=chunk_overlap,
+        length_function=len,
+        separators=["\n\n", "\n", " ", ""]
+    )
+
+    return splitter.split_text(text)
+
+
+def chunk_text_without_tables(
+    text: str,
+    chunk_size: int,
+    chunk_overlap: int,
+    metadata: Optional[str],
+    prepend_metadata_func,
+    page_tag_processor: Optional[Any] = None
+) -> List[str]:
+    """
+    Chunk text that does not contain tables.
+
+    Args:
+        text: Text to chunk
+        chunk_size: Maximum chunk size
+        chunk_overlap: Overlap size between chunks
+        metadata: Metadata to prepend to chunks
+        prepend_metadata_func: Function to prepend metadata
+        page_tag_processor: PageTagProcessor instance (for custom tag patterns)
+
+    Returns:
+        List of chunks
+    """
+    if not text or not text.strip():
+        return []
+
+    # Handle HTML code blocks (```html ... ```) separately
+    html_code_pattern = r'```html\s*(.*?)\s*```'
+
+    html_chunks = []
+    matches = list(re.finditer(html_code_pattern, text, re.DOTALL))
+
+    if matches:
+        current_pos = 0
+        for m in matches:
+            s, e = m.span()
+            before = text[current_pos:s].strip()
+            if before:
+                html_chunks.append(('text', before))
+            html_chunks.append(('html', text[s:e]))
+            current_pos = e
+        after = text[current_pos:].strip()
+        if after:
+            html_chunks.append(('text', after))
+    else:
+        html_chunks = [('text', text)]
+
+    final_chunks: List[str] = []
+
+    for kind, content in html_chunks:
+        if kind == 'html':
+            # Keep HTML code blocks as-is
+            final_chunks.append(content)
+            continue
+
+        # Plain text uses RecursiveCharacterTextSplitter for chunking
+        text_chunks = chunk_plain_text(content, chunk_size, chunk_overlap)
+        final_chunks.extend(text_chunks)
+
+    cleaned_chunks = clean_chunks(final_chunks, page_tag_processor)
+    cleaned_chunks = prepend_metadata_func(cleaned_chunks, metadata)
+
+    return cleaned_chunks
+
+
+def _is_markdown_table(text: str) -> bool:
+    """
+    Check if text is a Markdown table.
+    """
+    lines = text.strip().split('\n')
+    if len(lines) < 2:
+        return False
+    has_pipe_rows = any(line.strip().startswith('|') for line in lines)
+    has_separator = any('---' in line and '|' in line for line in lines)
+    return has_pipe_rows and has_separator
+
+
+def chunk_with_row_protection(
+    text: str,
+    chunk_size: int,
+    chunk_overlap: int,
+    split_with_protected_regions_func,
+    chunk_large_table_func
+) -> List[str]:
+    """
+    Chunk with row-level protection when table protection is disabled.
+
+    HTML tables are processed with chunk_large_table_func to maintain structure.
+    Markdown tables are processed with chunk_large_markdown_table for proper row-level splitting.
+    Both table types have NO overlap applied.
+
+    Args:
+        text: Text to chunk
+        chunk_size: Maximum chunk size
+        chunk_overlap: Overlap size between chunks (NOT applied to tables)
+        split_with_protected_regions_func: Protected region splitting function
+        chunk_large_table_func: Large table chunking function (for HTML)
+
+    Returns:
+        List of chunks
+    """
+    if not text or not text.strip():
+        return []
+
+    # === Extract both HTML and Markdown tables for separate processing ===
+    segments: List[Tuple[str, str]] = []  # [(type, content), ...]
+    
+    # Find all HTML tables
+    html_matches = list(re.finditer(HTML_TABLE_PATTERN, text, re.DOTALL | re.IGNORECASE))
+    
+    # Find all Markdown tables
+    markdown_matches = list(re.finditer(MARKDOWN_TABLE_PATTERN, text, re.MULTILINE))
+    
+    # Combine and sort by start position
+    all_matches = []
+    for match in html_matches:
+        all_matches.append((match.start(), match.end(), 'html_table', match.group(0)))
+    for match in markdown_matches:
+        start = match.start()
+        if match.group(0).startswith('\n'):
+            start += 1
+        all_matches.append((start, match.end(), 'markdown_table', match.group(0).strip()))
+    
+    # Sort by start position
+    all_matches.sort(key=lambda x: x[0])
+    
+    # Remove overlapping matches (first non-overlapping match by position wins)
+    filtered_matches = []
+    last_end = 0
+    for start, end, ttype, content in all_matches:
+        if start >= last_end:
+            filtered_matches.append((start, end, ttype, content))
+            last_end = end
+    
+    # Build segments
+    last_end = 0
+    for start, end, ttype, content in filtered_matches:
+        # Text before table
+        if start > last_end:
+            before_text = text[last_end:start].strip()
+            if before_text:
+                segments.append(('text', before_text))
+        
+        # Table
+        segments.append((ttype, content))
+        last_end = end
+    
+    # Text after last table
+    if last_end < len(text):
+        after_text = text[last_end:].strip()
+        if after_text:
+            segments.append(('text', after_text))
+
+    # If no tables, use simple row protection
+    if not any(seg_type in ('html_table', 'markdown_table') for seg_type, _ in segments):
+        return chunk_with_row_protection_simple(
+            text, chunk_size, chunk_overlap, split_with_protected_regions_func
+        )
+
+    # === Process each segment ===
+    all_chunks: List[str] = []
+
+    for seg_type, content in segments:
+        if seg_type == 'html_table':
+            # HTML table -> split efficiently by rows with NO overlap
+            table_chunks = chunk_large_table_func(content, chunk_size, 0, "")
+            all_chunks.extend(table_chunks)
+        elif seg_type == 'markdown_table':
+            # Markdown table -> split efficiently by rows with NO overlap
+            from .table_chunker import chunk_large_markdown_table
+            table_chunks = chunk_large_markdown_table(content, chunk_size, 0, "")
+            all_chunks.extend(table_chunks)
+        else:
+            # Plain text -> chunk with Markdown row protection
+            text_chunks = chunk_with_row_protection_simple(
+                content, chunk_size, chunk_overlap, split_with_protected_regions_func
+            )
+            all_chunks.extend(text_chunks)
+
+    return all_chunks
+
+
+def chunk_with_row_protection_simple(
+    text: str,
+    chunk_size: int,
+    chunk_overlap: int,
+    split_with_protected_regions_func
+) -> List[str]:
+    """
+    Chunk while protecting Markdown table rows from being split mid-row.
+    Assumes HTML tables have already been separated.
+
+    NOTE: If a complete Markdown table is found, it will be chunked with NO overlap
+    using chunk_large_markdown_table. Only individual rows (not part of a complete table)
+    are protected as regions.
+
+    Args:
+        text: Text to chunk
+        chunk_size: Maximum chunk size
+        chunk_overlap: Overlap size between chunks (NOT applied to Markdown tables)
+        split_with_protected_regions_func: Protected region splitting function
+
+    Returns:
+        List of chunks
+    """
+    if not text or not text.strip():
+        return []
+
+    # Check if text contains a complete Markdown table
+    if _is_markdown_table(text):
+        # Process as a complete Markdown table with NO overlap
+        from .table_chunker import chunk_large_markdown_table
+        return chunk_large_markdown_table(text, chunk_size, 0, "")
+
+    # Protect individual Markdown table rows (for mixed content)
+    row_patterns = [
+        r'\|[^\n]+\|',  # Markdown table row (headers, data, separators)
+    ]
+
+    # Find all row positions
+    row_positions: List[Tuple[int, int]] = []
+    for pattern in row_patterns:
+        for match in re.finditer(pattern, text, re.DOTALL | re.IGNORECASE):
+            row_positions.append((match.start(), match.end()))
+
+    # Sort by position
+    row_positions.sort(key=lambda x: x[0])
+
+    # Merge overlapping regions
+    merged_rows: List[Tuple[int, int]] = []
+    for start, end in row_positions:
+        if merged_rows and start < merged_rows[-1][1]:
+            # Overlap -> merge
+            prev_start, prev_end = merged_rows[-1]
+            merged_rows[-1] = (prev_start, max(prev_end, end))
+        else:
+            merged_rows.append((start, end))
+
+    if not merged_rows:
+        # No rows to protect -> use plain chunking
+        return chunk_plain_text(text, chunk_size, chunk_overlap)
+
+    # Chunk while protecting rows
+    return split_with_protected_regions_func(text, merged_rows, chunk_size, chunk_overlap)
+
+
+def clean_chunks(
+    chunks: List[str],
+    page_tag_processor: Optional[Any] = None
+) -> List[str]:
+    """
+    Clean chunks: remove empty chunks and chunks with only page markers.
+
+    Args:
+        chunks: List of chunks
+        page_tag_processor: PageTagProcessor instance (for custom tag patterns)
+
+    Returns:
+        Cleaned list of chunks
+    """
+    cleaned_chunks = []
+
+    # Build patterns from PageTagProcessor or use defaults
+    if page_tag_processor is not None:
+        config = page_tag_processor.config
+        # Page pattern with optional OCR suffix
+        page_prefix = re.escape(config.tag_prefix)
+        page_suffix = re.escape(config.tag_suffix)
+        slide_prefix = re.escape(config.slide_prefix)
+        slide_suffix = re.escape(config.slide_suffix)
+
+        page_marker_patterns = [
+            f"{page_prefix}\\d+(\\s*\\(OCR[+Ref]*\\))?{page_suffix}",
+            f"{slide_prefix}\\d+(\\s*\\(OCR\\))?{slide_suffix}",
+        ]
+    else:
+        # Default patterns
+        page_marker_patterns = [
+            r"\[Page Number:\s*\d+(\s*\(OCR[+Ref]*\))?\]",
+            r"\[Slide Number:\s*\d+(\s*\(OCR\))?\]",
+        ]
+
+    for chunk in chunks:
+        if not chunk.strip():
+            continue
+
+        # Check if chunk contains only page marker
+        is_page_marker_only = False
+        for pattern in page_marker_patterns:
+            if re.fullmatch(pattern, chunk.strip()):
+                is_page_marker_only = True
+                break
+
+        if not is_page_marker_only:
+            cleaned_chunks.append(chunk)
+
+    return cleaned_chunks
+
+
+def chunk_code_text(
+    text: str,
+    file_type: str,
+    chunk_size: int = 1500,
+    chunk_overlap: int = 300
+) -> List[str]:
+    """
+    Chunk code text using language-specific splitter.
+
+    Args:
+        text: Code text
+        file_type: File extension (e.g., 'py', 'js')
+        chunk_size: Maximum chunk size
+        chunk_overlap: Overlap size between chunks
+
+    Returns:
+        List of chunks
+    """
+    if not text or not text.strip():
+        return [""]
+
+    lang = LANGCHAIN_CODE_LANGUAGE_MAP.get(file_type.lower())
+
+    if lang:
+        splitter = RecursiveCharacterTextSplitter.from_language(
+            language=lang, chunk_size=chunk_size, chunk_overlap=chunk_overlap
+        )
+    else:
+        splitter = RecursiveCharacterTextSplitter(
+            chunk_size=chunk_size, chunk_overlap=chunk_overlap,
+            length_function=len, separators=["\n\n", "\n", " ", ""]
+        )
+
+    chunks = splitter.split_text(text)
+    logger.info(f"Code text split into {len(chunks)} chunks (size: {chunk_size}, overlap: {chunk_overlap})")
+
+    return chunks
+
+
+def reconstruct_text_from_chunks(chunks: List[str], chunk_overlap: int) -> str:
+    """
+    Reconstruct original text from chunks.
+    Removes overlap portions to avoid duplication.
+
+    Args:
+        chunks: List of chunks
+        chunk_overlap: Overlap size between chunks
+
+    Returns:
+        Reconstructed text
+    """
+    if not chunks:
+        return ""
+    if len(chunks) == 1:
+        return chunks[0]
+
+    out = chunks[0]
+    for i in range(1, len(chunks)):
+        prev = chunks[i - 1]
+        cur = chunks[i]
+        ov = find_overlap_length(prev, cur, chunk_overlap)
+        out += cur[ov:] if ov > 0 else cur
+
+    return out
+
+
+def find_overlap_length(c1: str, c2: str, max_overlap: int) -> int:
+    """
+    Find the actual overlap length between two chunks.
+
+    Args:
+        c1: Previous chunk
+        c2: Current chunk
+        max_overlap: Maximum overlap size
+
+    Returns:
+        Actual overlap length
+    """
+    max_check = min(len(c1), len(c2), max_overlap)
+    for ov in range(max_check, 0, -1):
+        if c1[-ov:] == c2[:ov]:
+            return ov
+    return 0
+
+
+def estimate_chunks_count(text: str, chunk_size: int = 1000, chunk_overlap: int = 200) -> int:
+    """
+    Estimate the number of chunks when text is chunked.
+
+    Args:
+        text: Text
+        chunk_size: Maximum chunk size
+        chunk_overlap: Overlap size between chunks
+
+    Returns:
+        Estimated chunk count
+    """
+    if not text:
+        return 0
+    if len(text) <= chunk_size:
+        return 1
+
+    eff = chunk_size - chunk_overlap
+    return max(1, (len(text) - chunk_overlap) // eff + 1)

xgen_doc2chunk/core/__init__.py

@@ -0,0 +1,64 @@
+# xgen_doc2chunk/core/__init__.py
+"""
+Core - Document Processing Core Module
+
+This package provides core functionality for processing various document formats.
+
+Module Structure:
+- document_processor: Main DocumentProcessor class
+- processor/: Individual document type handlers
+    - pdf_handler: PDF document processing
+    - docx_handler: DOCX document processing
+    - doc_handler: DOC document processing
+    - ppt_handler: PPT/PPTX document processing
+    - excel_handler: Excel document processing
+    - hwp_handler: HWP document processing
+    - hwpx_handler: HWPX document processing
+    - csv_handler: CSV file processing
+    - text_handler: Text file processing
+- functions/: Utility functions
+    - utils: Text cleaning, code cleaning, and common utilities
+    - img_processor: Image processing and saving (ImageProcessor class)
+    - ppt2pdf: PPT to PDF conversion
+
+Usage:
+    from xgen_doc2chunk import DocumentProcessor
+    from xgen_doc2chunk.core.processor import PDFHandler, DocxHandler
+    from xgen_doc2chunk.core.functions import clean_text, ImageProcessor
+"""
+
+# === Main Class ===
+from xgen_doc2chunk.core.document_processor import DocumentProcessor
+
+# === Utility Functions ===
+from xgen_doc2chunk.core.functions.utils import (
+    clean_text,
+    clean_code_text,
+    sanitize_text_for_json,
+)
+
+# === Image Processing ===
+from xgen_doc2chunk.core.functions.img_processor import (
+    ImageProcessor,
+    save_image_to_file,
+)
+
+# === Explicit Subpackage Imports ===
+from xgen_doc2chunk.core import processor
+from xgen_doc2chunk.core import functions
+
+__all__ = [
+    # Main Class
+    "DocumentProcessor",
+    # Utility Functions
+    "clean_text",
+    "clean_code_text",
+    "sanitize_text_for_json",
+    # Image Processing
+    "ImageProcessor",
+    "save_image_to_file",
+    # Subpackages
+    "processor",
+    "functions",
+]
+