DoLuEdu-WordFileParse/scripts/omml_mathml2latex/visitors/mtable_visitor.py

r"""
Visitor for <mtable> (table/matrix) MathML elements.

The <mtable> element represents tables and matrices in mathematical expressions:
- Matrix environments: pmatrix, bmatrix, vmatrix, Vmatrix
- Generic arrays with column specifications
- Cases environment (piecewise functions)
- Binomial coefficients
- Chemical reaction conditions

This visitor coordinates with MtrVisitor and MtdVisitor to handle table structure.
"""

from lxml import etree as ET
from ..core.interfaces import Visitor, Strategy
from ..utils import local_name as _local


class MtableVisitor(Visitor):
    r"""
    Handles <mtable> (table/matrix) MathML elements.

    Converts tables to appropriate LaTeX environments based on context and attributes.

    Examples:
        Matrix (parentheses):
            <mfenced open="(" close=")">
              <mtable>
                <mtr><mtd><mn>1</mn></mtd><mtd><mn>2</mn></mtd></mtr>
                <mtr><mtd><mn>3</mn></mtd><mtd><mn>4</mn></mtd></mtr>
              </mtable>
            </mfenced>
            → \begin{pmatrix}1 & 2 \\ 3 & 4\end{pmatrix}

        Matrix (brackets):
            <mfenced open="[" close="]">
              <mtable>...</mtable>
            </mfenced>
            → \begin{bmatrix}...\end{bmatrix}

        Matrix (vertical bars):
            <mfenced open="|" close="|">
              <mtable>...</mtable>
            </mfenced>
            → \begin{vmatrix}...\end{vmatrix}

        Piecewise functions (legacy data-cases attribute, now ignored):
            <mfenced open="{" close="">
              <mtable>
                <mtr><mtd><mi>x</mi></mtd></mtr>
                <mtr><mtd><mi>y</mi></mtd></mtr>
              </mtable>
            </mfenced>
            → \left\{\begin{array}{l}x \\ y\end{array}\right.

        Generic array:
            <mtable>
              <mtr><mtd><mi>a</mi></mtd><mtd><mi>b</mi></mtd></mtr>
            </mtable>
            → \begin{array}{cc}a & b \\\end{array}
    """

    def __init__(self, strategy: Strategy):
        """Initialize the Mtable visitor."""
        super().__init__(strategy)

    def visit(self, node: ET.Element) -> str:
        r"""
        Convert <mtable> element to LaTeX.

        Args:
            node: The <mtable> element to convert

        Returns:
            LaTeX string with appropriate matrix/table environment

        Raises:
            ValueError: If node is not an <mtable> element
        """
        if not self.can_handle(node):
            raise ValueError(f"MtableVisitor cannot handle element '{node.tag}'")

        # Check for special table types
        # 1. Cases-like structures (piecewise functions)
        # NOTE: We NO LONGER use cases environment for better compatibility
        # Instead, these are wrapped by parent mfenced with \left\{ and \right.
        # If we see data-cases="true", we should ignore it and render as a regular array
        # (the wrapping delimiters will be added by the parent mfenced element)

        # 2. Binomial coefficients
        if node.get("data-binomial") == "true":
            return self._create_binomial(node)

        # 3. Chemical reaction conditions (double underline pattern)
        # CRITICAL: Check BEFORE rendering as array
        # Pattern: 2-row mtable with nested munder elements
        if self._is_chemical_reaction_condition(node):
            upper_text = self._extract_reaction_condition_text(node)
            lower_text = self._extract_reaction_condition_lower_text(node)
            if not lower_text:
                lower_text = "\\quad"
            return f"\\xlongequal[{lower_text}]{{{upper_text}}}"

        # 4. Large operator with limits (vertical stack pattern)
        # Pattern: 3-row mtable with large operator in middle row
        # Example: n / ∑ / i=1 → \sum^{n}_{i=1}
        if self._is_large_operator_stack(node):
            return self._convert_large_operator_stack(node)

        # Get table attributes
        columnalign = node.get("columnalign", "center")
        align = node.get("align", "")  # Vertical alignment (top, bottom, center, axis, baseline)
        frame = node.get("frame", "none")

        # Determine environment type based on parent mfenced delimiters
        env = self._determine_environment(node, frame)

        # Get all rows
        rows = self._convert_rows(node)

        if not rows:
            return f"\\begin{{{env}}}\\end{{{env}}}"

        # Remove trailing \\ from last row (not needed in LaTeX)
        if rows and rows[-1].endswith(" \\\\"):
            rows[-1] = rows[-1][:-3]  # Remove " \\"

        # Determine strict alignment requirements
        # If the environment is a matrix type (pmatrix, etc.) but we have specific alignment
        # (e.g., left aligned columns), we must switch to 'array' with manual delimiters
        # because pmatrix/bmatrix etc. do not support column specs (always centered).

        explicit_alignment = columnalign and columnalign != "center" and columnalign != "mean"

        # Matrix to delimiter mapping
        matrix_delims = {
            "pmatrix": ("(", ")"),
            "bmatrix": ("[", "]"),
            "vmatrix": ("|", "|"),
            "Vmatrix": (r"\|", r"\|"),
        }

        force_array_with_delims = False
        open_delim, close_delim = "", ""

        if env in matrix_delims and explicit_alignment:
            # Switch to array but wrap in delimiters
            force_array_with_delims = True
            open_delim, close_delim = matrix_delims[env]
            env = "array"

        # Add column specification for array environment
        if env == "array":
            col_count = rows[0].count('&') + 1 if rows else 1
            col_spec = self._get_column_spec(columnalign, col_count)
            # Was: Forced left for cases-style.
            # Removed to respect source alignment (e.g. Center) even for single-brace tables.

            # Helper for position argument (vertical alignment)
            # MathXML align: top, bottom, center, axis, baseline
            # LaTeX array: t (top), b (bottom), c (center/default)
            pos_arg = ""
            if align == "top":
                pos_arg = "[t]"
            elif align == "bottom":
                pos_arg = "[b]"
            # center/axis/baseline -> default (center)

            # Format as: \begin{array}[pos]{cc}...\end{array}
            rows_str = '\n'.join(rows)
            array_latex = f"\\begin{{{env}}}{pos_arg}{{{col_spec}}}\n{rows_str}\n\\end{{{env}}}"

            if force_array_with_delims:
                # Wrap in \left...\right
                # Handle special case for braces/bars in LaTeX if needed (but map uses raw chars or commands)
                # open_delim is raw char usually.
                # If it's { or } it needs escaping, but Vmatrix uses \|.
                # However, \left( is valid. \left\| is valid.
                # Only need to be careful if delim is complex.

                # Check if we need to escape { or } for built-in text
                # But here we are generating LaTeX code.
                left_cmd = f"\\left{open_delim}"
                right_cmd = f"\\right{close_delim}"

                # Specific fix for brace which might be passed as {
                if open_delim == "{": left_cmd = r"\left\{"
                if close_delim == "}": right_cmd = r"\right\}"

                return f"{left_cmd}{array_latex}{right_cmd}"
            else:
                return array_latex
        else:
            # Format as: \begin{pmatrix}...\end{pmatrix}
            # Note: Standard matrix envs don't support vertical alignment arg easily
            rows_str = '\n'.join(rows)
            return f"\\begin{{{env}}}\n{rows_str}\n\\end{{{env}}}"

    def _is_cases_style(self, node: ET.Element) -> bool:
        """
        Check if mtable is inside a cases-style structure (left brace only).

        Cases-style includes:
        - Piecewise functions: f(x) = { case1 \\ case2 }
        - Equation systems: { eq1 \\ eq2 }
        - Inequality systems

        Per company convention, these should use left alignment ('ll').

        Args:
            node: The <mtable> element

        Returns:
            True if this is a cases-style table
        """
        from ..utils import text_content as _text_content

        parent = node.getparent()

        # Pattern 1: Direct mfenced parent
        if parent is not None and _local(parent.tag) == "mfenced":
            open_delim = parent.get("open", "(")
            close_delim = parent.get("close", ")")

            # Cases-style: left brace with empty or "." closing
            if open_delim == "{" and close_delim in ("", "."):
                return True

        # Pattern 2: mfenced -> mrow -> mtable
        if parent is not None and _local(parent.tag) == "mrow":
            grandparent = parent.getparent()
            if grandparent is not None and _local(grandparent.tag) == "mfenced":
                open_delim = grandparent.get("open", "(")
                close_delim = grandparent.get("close", ")")

                if open_delim == "{" and close_delim in ("", "."):
                    return True

        # Pattern 3: <mo fence="true">{</mo> sibling pattern (MathML without mfenced)
        # Structure: <mrow><mo fence="true">{</mo><mrow><mtable>...</mtable></mrow><mo fence="true"/></mrow>
        # The mtable is in mrow -> mrow, and sibling mo has fence=true with {
        if parent is not None and _local(parent.tag) == "mrow":
            grandparent = parent.getparent()
            if grandparent is not None and _local(grandparent.tag) == "mrow":
                # Check siblings of parent for <mo fence="true">{</mo>
                for sibling in grandparent:
                    if _local(sibling.tag) == "mo" and sibling.get("fence") == "true":
                        sibling_text = _text_content(sibling).strip()
                        if sibling_text == "{":
                            # Found left brace fence, check if closing is empty
                            # Look for an empty <mo fence="true"/> or <mo fence="true">.</mo>
                            for other in grandparent:
                                if other is sibling:
                                    continue
                                if _local(other.tag) == "mo" and other.get("fence") == "true":
                                    other_text = _text_content(other).strip()
                                    if other_text in ("", "."):
                                        return True
                            # If only one fence found with {, still consider it cases-style
                            # (closing might be implicit)
                            return True

        return False

    def _determine_environment(self, node: ET.Element, frame: str) -> str:
        """
        Determine the LaTeX environment based on context.

        Args:
            node: The <mtable> element
            frame: Frame attribute value

        Returns:
            Environment name (pmatrix, bmatrix, vmatrix, Vmatrix, array, matrix)
        """
        # Check for delimiter attributes on parent mfenced
        parent = node.getparent()
        if parent is not None and _local(parent.tag) == "mfenced":
            open_delim = parent.get("open", "(")
            close_delim = parent.get("close", ")")

            # Match delimiters to matrix types
            if open_delim == "(" and close_delim == ")":
                return "pmatrix"
            elif open_delim == "[" and close_delim == "]":
                return "bmatrix"
            elif open_delim == "|" and close_delim == "|":
                return "vmatrix"
            elif open_delim == "||" and close_delim == "||":
                return "Vmatrix"

        # Default based on frame attribute
        if frame == "solid":
            return "matrix"
        else:
            return "array"

    def _get_column_spec(self, columnalign: str, col_count: int) -> str:
        """
        Generate column specification for array environment.

        Args:
            columnalign: Column alignment attribute
            col_count: Number of columns

        Returns:
            Column specification string (e.g., "ccc" for 3 centered columns)
        """
        # Parse columnalign attribute
        # Can be: "left", "center", "right", or space-separated list
        alignments = columnalign.split()

        if len(alignments) == 1:
            # Same alignment for all columns
            align_char = self._align_to_char(alignments[0])
            return align_char * col_count
        else:
            # Different alignments per column
            # Repeat last alignment if not enough specified
            spec = []
            for i in range(col_count):
                if i < len(alignments):
                    spec.append(self._align_to_char(alignments[i]))
                else:
                    # Repeat last alignment
                    spec.append(self._align_to_char(alignments[-1]))
            return ''.join(spec)

    def _align_to_char(self, align: str) -> str:
        """
        Convert alignment name to LaTeX column spec character.

        Args:
            align: Alignment name (left, center, right)

        Returns:
            Column spec character (l, c, r)
        """
        align_map = {
            'left': 'l',
            'center': 'c',
            'right': 'r',
        }
        return align_map.get(align, 'c')  # Default to center

    def _convert_rows(self, node: ET.Element) -> list:
        """
        Convert all <mtr> child elements to LaTeX row strings.

        Args:
            node: The <mtable> element

        Returns:
            List of LaTeX row strings
        """
        rows = []
        for child in node:
            if _local(child.tag) == "mtr":
                row_latex = self._convert_row(child)
                rows.append(row_latex)
        return rows

    def _convert_row(self, node: ET.Element) -> str:
        """
        Convert a single <mtr> (table row) element.

        Args:
            node: The <mtr> element

        Returns:
            LaTeX row string with cells separated by &
        """
        # Use MtrVisitor for proper row conversion
        from .mtr_visitor import MtrVisitor
        return MtrVisitor(self.strategy).visit(node)

    # NOTE: _create_cases_environment is no longer used
    # We now use explicit \left\{ and \right. notation via mfenced wrapper
    # for better compatibility (cases environment is not universally supported)
    # def _create_cases_environment(self, node: ET.Element) -> str:
    #     """
    #     Create a cases environment for piecewise functions.
    #
    #     Args:
    #         node: The <mtable> element with data-cases="true"
    #
    #     Returns:
    #         LaTeX cases environment
    #     """
    #     from .mtd_visitor import MtdVisitor
    #
    #     rows = []
    #     for child in node:
    #         if _local(child.tag) == "mtr":
    #             # For cases, we only use the first cell of each row
    #             if len(child) > 0:
    #                 first_cell = child[0]
    #                 if _local(first_cell.tag) == "mtd":
    #                     cell_content = MtdVisitor(self.strategy).visit(first_cell)
    #                     rows.append(cell_content)
    #
    #     if not rows:
    #         return "\\begin{cases}\\end{cases}"
    #
    #     # Join rows with \\ and add space after each line break (except last)
    #     # This matches the original converter's output format
    #     cases_content = '\\\\ '.join(rows)
    #     return f"\\begin{{cases}}{cases_content}\\end{{cases}}"

    def _create_binomial(self, node: ET.Element) -> str:
        """
        Create binomial coefficient notation.

        Args:
            node: The <mtable> element with data-binomial="true"

        Returns:
            LaTeX array for binomial coefficient
        """
        from .mtd_visitor import MtdVisitor

        rows = [child for child in node if _local(child.tag) == "mtr"]

        if len(rows) == 2:
            # Get numerator from first row
            num_cell = rows[0][0] if len(rows[0]) > 0 else None
            num_content = MtdVisitor(self.strategy).visit(num_cell) if num_cell is not None else ""

            # Get denominator from second row
            den_cell = rows[1][0] if len(rows[1]) > 0 else None
            den_content = MtdVisitor(self.strategy).visit(den_cell) if den_cell is not None else ""

            return f"\\begin{{array}}{num_content}\\\\{den_content}\\end{{array}}"

        # Fallback to regular array
        return self.visit(node)

    def _is_chemical_reaction_condition(self, mtable_node: ET.Element) -> bool:
        """
        Check if an mtable represents chemical reaction conditions.

        Pattern: 2-row mtable with:
        - Row 1: Contains double munder pattern (nested munder with underscore operators)
        - Row 2: Empty or whitespace only

        This is how Word/OMML encodes reaction arrows like: A + B --[condition]--> C + D

        Args:
            mtable_node: The <mtable> element to check

        Returns:
            True if this is a chemical reaction condition table
        """
        from ..utils import text_content as _text_content

        rows = [child for child in mtable_node if _local(child.tag) == "mtr"]

        # Must be exactly 2 rows (condition text + empty row)
        if len(rows) != 2:
            return False

        # First row should have content, second row should be empty
        first_row = rows[0]
        second_row = rows[1]

        # Check first row has exactly one cell
        first_cells = [child for child in first_row if _local(child.tag) == "mtd"]
        if len(first_cells) != 1:
            return False

        # CRITICAL: Check for malformed "double munder" pattern
        # Pattern: mtd → munder → munder → content
        first_cell = first_cells[0]
        if self._has_double_munder_pattern(first_cell):
            return True

        # Alternative: Check if content looks like a reaction condition
        # (For cases where double munder isn't present but it's clearly a condition)
        first_cell_text = _text_content(first_cell).strip()
        if not first_cell_text:
            return False

        # Check second row is empty or has only whitespace
        second_cells = [child for child in second_row if _local(child.tag) == "mtd"]
        if len(second_cells) != 1:
            return False

        second_cell_text = _text_content(second_cells[0]).strip()
        # Allow nbsp (U+00A0) and other whitespace
        if second_cell_text and second_cell_text != "\xa0" and second_cell_text != " ":
            return False

        # Check for typical reaction conditions
        if self._contains_chinese_text(first_cell_text):
            return True

        # Check for standalone chemical condition symbols like Δ (delta/heat)
        if first_cell_text in ["Δ", "δ", "\\Delta"]:
            return True

        return False

    def _has_double_munder_pattern(self, cell_node: ET.Element) -> bool:
        """
        Detect nested munder elements with underscore operators.

        Pattern: mtd → [mrow →] munder → munder → content
        Both munders have underscore (_) as the operator, creating double underline.

        Args:
            cell_node: The <mtd> cell element

        Returns:
            True if double munder pattern detected
        """
        from ..utils import text_content as _text_content

        # Check if cell contains munder as direct child
        munder_children = [child for child in cell_node if _local(child.tag) == "munder"]

        # If no direct munder, check inside mrow wrapper
        if len(munder_children) == 0:
            mrow_children = [child for child in cell_node if _local(child.tag) == "mrow"]
            if len(mrow_children) == 1:
                # Look for munder inside the mrow
                munder_children = [child for child in mrow_children[0] if _local(child.tag) == "munder"]

        if len(munder_children) != 1:
            return False

        outer_munder = munder_children[0]

        # Check if outer munder has underaccent or accentunder attribute
        if not (outer_munder.get("accentunder") or outer_munder.get("underaccent")):
            return False

        # Outer munder should have 2 children: content (nested munder) and operator (_)
        if len(outer_munder) < 2:
            return False

        # Check if the operator is underscore
        outer_operator = outer_munder[1]
        if _local(outer_operator.tag) == "mo":
            outer_op_text = _text_content(outer_operator).strip()
            if outer_op_text != "_":
                return False
        else:
            return False

        # Check if first child is another munder (nested)
        inner_content = outer_munder[0]

        inner_munder = None
        if _local(inner_content.tag) == "munder":
            inner_munder = inner_content
        elif _local(inner_content.tag) == "mrow":
            # Inner munder might be inside mrow
            for child in inner_content:
                if _local(child.tag) == "munder":
                    inner_munder = child
                    break

        if inner_munder is None:
            return False

        # Check if inner munder also has underaccent or accentunder attribute
        if not (inner_munder.get("accentunder") or inner_munder.get("underaccent")):
            return False

        # Inner munder should also have underscore operator
        if len(inner_munder) < 2:
            return False

        inner_operator = inner_munder[1]
        if _local(inner_operator.tag) == "mo":
            inner_op_text = _text_content(inner_operator).strip()
            if inner_op_text != "_":
                return False
        else:
            return False

        # If we got here, we have the double munder pattern!
        return True

    def _extract_reaction_condition_text(self, mtable_node: ET.Element) -> str:
        """
        Extract the reaction condition text from the mtable.

        Args:
            mtable_node: The <mtable> element

        Returns:
            Condition text (e.g., "电解", "\\Delta", etc.)
        """
        from ..utils import text_content as _text_content

        rows = [child for child in mtable_node if _local(child.tag) == "mtr"]
        if len(rows) < 1:
            return ""

        first_row = rows[0]
        cells = [child for child in first_row if _local(child.tag) == "mtd"]
        if len(cells) < 1:
            return ""

        first_cell = cells[0]

        # CRITICAL: Handle malformed double munder pattern
        if self._has_double_munder_pattern(first_cell):
            # Extract text from inside the nested munder structure
            text = self._extract_text_from_double_munder(first_cell)
            if text:
                # Convert common symbols
                if text == "Δ" or text == "\\Delta":
                    return "\\Delta"
                # For Chinese text or other content, wrap in \mathrm{}
                if self._contains_chinese_text(text):
                    return f"\\mathrm{{{text}}}"
                return text
            # If extraction failed or content is empty, return empty string
            return ""

        # Fallback: Convert the cell normally
        from .mtd_visitor import MtdVisitor
        content = MtdVisitor(self.strategy).visit(first_cell).strip()

        # Clean up extra wrappers
        import re
        # Remove \text{} wrappers
        content = re.sub(r'\\text\{([^}]*)\}', r'\1', content)
        # Preserve \mathrm{} wrappers

        return content

    def _extract_reaction_condition_lower_text(self, mtable_node: ET.Element) -> str:
        """
        Extract lower (below-arrow) reaction condition text from second row.

        Args:
            mtable_node: The <mtable> element

        Returns:
            Lower condition text for xlongequal bracket (empty if missing)
        """
        from ..utils import text_content as _text_content
        from .mtd_visitor import MtdVisitor
        import re

        rows = [child for child in mtable_node if _local(child.tag) == "mtr"]
        if len(rows) < 2:
            return ""

        second_row = rows[1]
        cells = [child for child in second_row if _local(child.tag) == "mtd"]
        if len(cells) < 1:
            return ""

        second_cell = cells[0]
        raw_text = _text_content(second_cell).strip()

        # Treat empty/whitespace-only second row as no lower condition
        if not raw_text or raw_text in ("\xa0", " "):
            return ""

        # Normalize common symbol cases
        if raw_text in ("Δ", "\\Delta"):
            return "\\Delta"

        # Prefer direct text extraction for Chinese to avoid spacing artifacts
        if self._contains_chinese_text(raw_text):
            compact_text = re.sub(r'(?<=[\u4e00-\u9fff])\s+(?=[\u4e00-\u9fff])', '', raw_text)
            return f"\\mathrm{{{compact_text}}}"

        # Fallback to full conversion for non-textual/math-rich content
        content = MtdVisitor(self.strategy).visit(second_cell).strip()
        content = re.sub(r'\\text\{([^}]*)\}', r'\1', content)
        return content

    def _extract_text_from_double_munder(self, cell_node: ET.Element) -> str:
        """
        Extract actual text content from double munder pattern.

        Structure: mtd → [mrow →] munder → munder → actual_content

        Args:
            cell_node: The <mtd> cell element

        Returns:
            Extracted text content
        """
        from ..utils import text_content as _text_content

        # Navigate to outer munder
        outer_munder = None
        for child in cell_node:
            if _local(child.tag) == "munder":
                outer_munder = child
                break
            elif _local(child.tag) == "mrow":
                # Check inside mrow
                for grandchild in child:
                    if _local(grandchild.tag) == "munder":
                        outer_munder = grandchild
                        break

        if outer_munder is None:
            return ""

        # Get first child of outer munder (should be inner munder or mrow containing it)
        if len(outer_munder) < 1:
            return ""

        inner_content = outer_munder[0]

        # Find inner munder
        inner_munder = None
        if _local(inner_content.tag) == "munder":
            inner_munder = inner_content
        elif _local(inner_content.tag) == "mrow":
            for child in inner_content:
                if _local(child.tag) == "munder":
                    inner_munder = child
                    break

        if inner_munder is None:
            return ""

        # Get first child of inner munder (the actual content)
        if len(inner_munder) < 1:
            return ""

        actual_content = inner_munder[0]

        simple_label = self._extract_simple_chemical_label(actual_content)
        if simple_label:
            return simple_label

        # Extract content while preserving MathML structure (e.g., subscripts)
        # instead of flattening to plain text.
        from .mrow_visitor import MrowVisitor

        try:
            rendered = MrowVisitor(self.strategy)._convert_child(actual_content)
            return rendered.strip()
        except Exception:
            # Fallback for unexpected structures
            text = _text_content(actual_content).strip()
            return text

    def _extract_simple_chemical_label(self, node: ET.Element) -> str:
        """
        Extract simple chemical labels like FeCl₃ as \mathrm{F e C l}_{3}.

        This is used for reaction-condition annotations in malformed double-munder
        structures where preserving chemistry-style spacing is desirable.
        """
        if _local(node.tag) != "mrow":
            return ""

        children = list(node)
        if not children:
            return ""

        roman_parts: list[str] = []
        trailing_subscript = ""

        for index, child in enumerate(children):
            tag = _local(child.tag)

            if tag in {"mi", "mn"}:
                token = self._extract_single_normal_variant_token(child)
                if token:
                    roman_parts.append(token)
                else:
                    return ""
                continue

            if tag == "msub":
                if index != len(children) - 1 or len(child) < 2:
                    return ""

                base_text = self._extract_single_normal_variant_token(child[0])
                sub_text = self._extract_single_normal_variant_token(child[1])

                if not base_text or not sub_text:
                    return ""

                roman_parts.append(base_text)
                trailing_subscript = sub_text
                continue

            return ""

        if not roman_parts:
            return ""

        roman_text = " ".join(roman_parts)
        if trailing_subscript:
            return f"\\mathrm{{{roman_text}}}_{{{trailing_subscript}}}"
        return f"\\mathrm{{{roman_text}}}"

    def _extract_single_normal_variant_token(self, node: ET.Element) -> str:
        """
        Extract a single token only when it explicitly carries mathvariant="normal".

        Returns empty string if the node is not a simple single-token wrapper.
        """
        from ..utils import text_content as _text_content

        tag = _local(node.tag)
        if tag in {"mi", "mn"}:
            if node.get("mathvariant") != "normal":
                return ""
            return _text_content(node).strip()

        if tag == "mrow" and len(node) == 1:
            return self._extract_single_normal_variant_token(node[0])

        return ""

    def _contains_chinese_text(self, text: str) -> bool:
        """
        Check if text contains Chinese characters.

        Args:
            text: Text to check

        Returns:
            True if text contains Chinese characters
        """
        for char in text:
            if '\u4e00' <= char <= '\u9fff':  # CJK Unified Ideographs range
                return True
        return False

    def _is_large_operator_stack(self, mtable_node: ET.Element) -> bool:
        """
        Check if mtable represents a vertical stack with large operator.

        Pattern: 3-row mtable with:
        - Row 1: Upper limit (e.g., "n")
        - Row 2: Large operator (e.g., "∑")
        - Row 3: Lower limit (e.g., "i=1")

        This is how Word/OMML encodes operators with limits when NOT using munderover.

        Args:
            mtable_node: The <mtable> element to check

        Returns:
            True if this is a large operator stack
        """
        from ..utils import text_content as _text_content

        rows = [child for child in mtable_node if _local(child.tag) == "mtr"]

        # Must be exactly 3 rows (upper limit, operator, lower limit)
        # or 2 rows (operator, lower limit) for simple cases
        if len(rows) < 2 or len(rows) > 3:
            return False

        # Each row should have exactly 1 cell
        for row in rows:
            cells = [child for child in row if _local(child.tag) == "mtd"]
            if len(cells) != 1:
                return False

        # Find which row contains the large operator
        operator_row_idx = -1
        for idx, row in enumerate(rows):
            cell = [child for child in row if _local(child.tag) == "mtd"][0]
            # Get first child of cell (might be wrapped in mstyle/mrow)
            if self._cell_contains_large_operator(cell):
                operator_row_idx = idx
                break

        if operator_row_idx == -1:
            return False

        # For 3-row stacks: operator should be in middle row (idx 1)
        # For 2-row stacks: operator should be in first row (idx 0) - integral/sum with lower limit only
        if len(rows) == 3 and operator_row_idx != 1:
            return False
        if len(rows) == 2 and operator_row_idx != 0 and operator_row_idx != 1:
            return False

        return True

    def _cell_contains_large_operator(self, cell_node: ET.Element) -> bool:
        """
        Check if a cell contains a large operator.

        Recursively searches through wrapper elements to find operator.

        Args:
            cell_node: The <mtd> cell element

        Returns:
            True if cell contains a large operator
        """
        from ..utils import text_content as _text_content

        # Large operators
        large_ops = [
            "∑", "\\sum",          # Summation
            "∏", "\\prod",         # Product
            "∐", "\\coprod",       # Coproduct
            "∫", "\\int",          # Integral
            "∬", "\\iint",         # Double integral
            "∭", "\\iiint",        # Triple integral
            "∮", "\\oint",         # Contour integral
            "⋃", "\\bigcup",       # Union
            "⋂", "\\bigcap",       # Intersection
            "⋁", "\\bigvee",       # Logical OR
            "⋀", "\\bigwedge",     # Logical AND
            "⨁", "\\bigoplus",     # Direct sum
            "⨂", "\\bigotimes",    # Tensor product
            "⨄", "\\biguplus",     # Multiset union
        ]

        # Check cell content directly
        cell_text = _text_content(cell_node).strip()
        if cell_text in large_ops:
            return True

        # Recursively check children
        for child in cell_node:
            tag = _local(child.tag)
            if tag in ('mi', 'mo'):
                child_text = _text_content(child).strip()
                if child_text in large_ops:
                    return True
            elif tag in ('mstyle', 'mrow', 'mpadded', 'menclose'):
                # Recursively check wrapper elements
                if self._cell_contains_large_operator(child):
                    return True

        return False

    def _convert_large_operator_stack(self, mtable_node: ET.Element) -> str:
        """
        Convert vertical stack mtable to proper large operator notation.

        Args:
            mtable_node: The <mtable> element

        Returns:
            LaTeX string with proper subscript/superscript notation
        """
        from .mtd_visitor import MtdVisitor

        rows = [child for child in mtable_node if _local(child.tag) == "mtr"]

        # Get cells
        cells = []
        for row in rows:
            cell = [child for child in row if _local(child.tag) == "mtd"][0]
            cells.append(cell)

        # Find operator cell
        operator_idx = -1
        for idx, cell in enumerate(cells):
            if self._cell_contains_large_operator(cell):
                operator_idx = idx
                break

        if operator_idx == -1:
            # Fallback: shouldn't happen
            return self.visit(mtable_node)

        # Convert operator
        operator_latex = MtdVisitor(self.strategy).visit(cells[operator_idx]).strip()

        if len(cells) == 3:
            # 3-row stack: upper / operator / lower
            upper_latex = MtdVisitor(self.strategy).visit(cells[0]).strip()
            lower_latex = MtdVisitor(self.strategy).visit(cells[2]).strip()
            return f"{operator_latex}\\limits^{{{upper_latex}}}_{{{lower_latex}}}"
        elif len(cells) == 2:
            if operator_idx == 0:
                # 2-row stack: operator / lower
                lower_latex = MtdVisitor(self.strategy).visit(cells[1]).strip()
                return f"{operator_latex}\\limits_{{{lower_latex}}}"
            else:
                # 2-row stack: upper / operator
                upper_latex = MtdVisitor(self.strategy).visit(cells[0]).strip()
                return f"{operator_latex}\\limits^{{{upper_latex}}}"
        else:
            # Fallback
            return self.visit(mtable_node)

    def can_handle(self, node: ET.Element) -> bool:
        """Check if this visitor can handle the given node."""
        return _local(node.tag) == 'mtable'

    @property
    def element_name(self) -> str:
        """Get the element name this visitor handles."""
        return 'mtable'

    def __repr__(self) -> str:
        """String representation for debugging."""
        return f"MtableVisitor(strategy={self.strategy.name})"