undoc 0.1.19

//! DOCX parser implementation.

use std::collections::HashMap;

use crate::charts;
use crate::container::OoxmlContainer;
use crate::error::{Error, Result};
use crate::model::{
    Block, Cell, CellAlignment, Document, ListInfo, ListType, Metadata, Paragraph, Resource,
    ResourceType, RevisionType, Row, Section, Table, TextAlignment, TextRun, TextStyle,
    VerticalAlignment,
};

use super::numbering::NumberingMap;
use super::styles::StyleMap;

/// Parser for DOCX (Word) documents.
pub struct DocxParser {
    container: OoxmlContainer,
    styles: StyleMap,
    numbering: NumberingMap,
    relationships: crate::container::Relationships,
    /// Footnote id → plain text content
    footnotes: HashMap<String, String>,
    /// Endnote id → plain text content
    endnotes: HashMap<String, String>,
}

impl DocxParser {
    /// Open a DOCX file for parsing.
    pub fn open(path: impl AsRef<std::path::Path>) -> Result<Self> {
        let container = OoxmlContainer::open(path)?;
        Self::from_container(container)
    }

    /// Create a parser from bytes.
    pub fn from_bytes(data: Vec<u8>) -> Result<Self> {
        let container = OoxmlContainer::from_bytes(data)?;
        Self::from_container(container)
    }

    /// Create a parser from a container.
    fn from_container(container: OoxmlContainer) -> Result<Self> {
        // Parse styles
        let styles = if let Ok(xml) = container.read_xml("word/styles.xml") {
            StyleMap::parse(&xml)?
        } else {
            StyleMap::default()
        };

        // Parse numbering
        let numbering = if let Ok(xml) = container.read_xml("word/numbering.xml") {
            NumberingMap::parse(&xml)?
        } else {
            NumberingMap::default()
        };

        // Parse document relationships
        let relationships = container
            .read_relationships("word/document.xml")
            .unwrap_or_default();

        // Parse footnotes
        let footnotes = if let Ok(xml) = container.read_xml("word/footnotes.xml") {
            parse_notes_xml(&xml, b"w:footnote")
        } else {
            HashMap::new()
        };

        // Parse endnotes
        let endnotes = if let Ok(xml) = container.read_xml("word/endnotes.xml") {
            parse_notes_xml(&xml, b"w:endnote")
        } else {
            HashMap::new()
        };

        Ok(Self {
            container,
            styles,
            numbering,
            relationships,
            footnotes,
            endnotes,
        })
    }

    /// Parse the document and return a Document model.
    pub fn parse(&mut self) -> Result<Document> {
        let mut doc = Document::new();

        // Parse metadata
        doc.metadata = self.parse_metadata()?;

        // Parse main document content
        let mut main_section = self.parse_document_xml()?;

        // Parse charts and add as tables for RAG-ready output
        let chart_tables = self.parse_charts()?;
        for table in chart_tables {
            main_section.add_block(Block::Table(table));
        }

        // Append footnote definitions at end of section
        if !self.footnotes.is_empty() {
            let mut ids: Vec<&String> = self.footnotes.keys().collect();
            ids.sort_by(|a, b| {
                a.parse::<u64>()
                    .unwrap_or(u64::MAX)
                    .cmp(&b.parse::<u64>().unwrap_or(u64::MAX))
            });
            for id in ids {
                if let Some(text) = self.footnotes.get(id) {
                    let para = Paragraph::with_text(format!("[^{}]: {}", id, text));
                    main_section.add_block(Block::Paragraph(para));
                }
            }
        }

        // Append endnote definitions at end of section
        if !self.endnotes.is_empty() {
            let mut ids: Vec<&String> = self.endnotes.keys().collect();
            ids.sort_by(|a, b| {
                a.parse::<u64>()
                    .unwrap_or(u64::MAX)
                    .cmp(&b.parse::<u64>().unwrap_or(u64::MAX))
            });
            for id in ids {
                if let Some(text) = self.endnotes.get(id) {
                    let para = Paragraph::with_text(format!("[^e{}]: {}", id, text));
                    main_section.add_block(Block::Paragraph(para));
                }
            }
        }

        doc.add_section(main_section);

        // Extract resources (images)
        self.extract_resources(&mut doc)?;

        Ok(doc)
    }

    /// Parse document metadata from docProps/core.xml.
    fn parse_metadata(&self) -> Result<Metadata> {
        // Use shared metadata parsing from container
        self.container.parse_core_metadata()
    }

    /// Parse charts from word/charts/ and convert to tables for RAG-ready output.
    fn parse_charts(&self) -> Result<Vec<Table>> {
        let mut tables = Vec::new();

        // Find chart relationships in document.xml.rels
        for (rel_type, rels) in &self.relationships.by_type {
            if !rel_type.contains("chart") {
                continue;
            }

            for rel in rels {
                // Resolve chart path relative to document.xml
                // Target is like "charts/chart1.xml"
                let chart_path = if rel.target.starts_with('/') {
                    rel.target[1..].to_string()
                } else {
                    format!("word/{}", rel.target)
                };

                // Read and parse chart XML
                if let Ok(chart_xml) = self.container.read_xml(&chart_path) {
                    match charts::parse_chart_xml(&chart_xml) {
                        Ok(chart_data) => {
                            if !chart_data.is_empty() {
                                let mut table = chart_data.to_table();
                                // Add chart title if available
                                if let Some(ref title) = chart_data.title {
                                    if !title.is_empty() {
                                        if let Some(first_row) = table.rows.first_mut() {
                                            if let Some(first_cell) = first_row.cells.first_mut() {
                                                let original = first_cell.plain_text();
                                                first_cell.content.clear();
                                                first_cell.content.push(Paragraph::with_text(
                                                    format!("{} ({})", original, title),
                                                ));
                                            }
                                        }
                                    }
                                }
                                tables.push(table);
                            }
                        }
                        Err(_) => {
                            // Chart parsing failed, skip
                        }
                    }
                }
            }
        }

        Ok(tables)
    }

    /// Parse the main document.xml content.
    fn parse_document_xml(&mut self) -> Result<Section> {
        let xml = self.container.read_xml("word/document.xml")?;
        let mut section = Section::new(0);

        let mut reader = quick_xml::Reader::from_str(&xml);
        // IMPORTANT: Don't trim text - preserve whitespace from xml:space="preserve" elements
        // This fixes the "DATE OF BIRTH" -> "DATEOFBIRTH" bug (GitHub Issue #2)
        reader.config_mut().trim_text(false);

        let mut buf = Vec::new();
        let mut in_body = false;
        let mut paragraph_xml = String::new();
        let mut table_xml = String::new();
        let mut in_paragraph = false;
        let mut para_depth: u32 = 0; // Track nested w:p depth (for text boxes)
        let mut table_depth: u32 = 0; // Track nested table depth
        let mut in_sect_pr = false; // Track w:sectPr for header/footer references
        let mut default_header_rid: Option<String> = None;
        let mut default_footer_rid: Option<String> = None;

        loop {
            match reader.read_event_into(&mut buf) {
                Ok(quick_xml::events::Event::Start(ref e)) => {
                    let name = e.name();
                    match name.as_ref() {
                        b"w:body" => {
                            in_body = true;
                        }
                        b"w:p" if in_body && table_depth == 0 && !in_paragraph => {
                            in_paragraph = true;
                            paragraph_xml.clear();
                            paragraph_xml.push_str("<w:p");
                            for attr in e.attributes().flatten() {
                                paragraph_xml.push_str(&format!(
                                    " {}=\"{}\"",
                                    String::from_utf8_lossy(attr.key.as_ref()),
                                    String::from_utf8_lossy(&attr.value)
                                ));
                            }
                            paragraph_xml.push('>');
                        }
                        b"w:sectPr" if in_body && !in_paragraph && table_depth == 0 => {
                            in_sect_pr = true;
                        }
                        b"w:tbl" if in_body => {
                            if table_depth == 0 {
                                // Start collecting table XML
                                table_xml.clear();
                            }
                            table_depth += 1;
                            table_xml.push_str("<w:tbl>");
                        }
                        _ => {
                            if in_paragraph {
                                // Track nested w:p depth for text boxes
                                if name.as_ref() == b"w:p" {
                                    para_depth += 1;
                                }
                                paragraph_xml.push('<');
                                paragraph_xml.push_str(&String::from_utf8_lossy(name.as_ref()));
                                for attr in e.attributes().flatten() {
                                    paragraph_xml.push_str(&format!(
                                        " {}=\"{}\"",
                                        String::from_utf8_lossy(attr.key.as_ref()),
                                        String::from_utf8_lossy(&attr.value)
                                    ));
                                }
                                paragraph_xml.push('>');
                            } else if table_depth > 0 {
                                table_xml.push('<');
                                table_xml.push_str(&String::from_utf8_lossy(name.as_ref()));
                                for attr in e.attributes().flatten() {
                                    table_xml.push_str(&format!(
                                        " {}=\"{}\"",
                                        String::from_utf8_lossy(attr.key.as_ref()),
                                        String::from_utf8_lossy(&attr.value)
                                    ));
                                }
                                table_xml.push('>');
                            }
                        }
                    }
                }
                Ok(quick_xml::events::Event::Empty(ref e)) => {
                    // Handle header/footer references inside w:sectPr
                    if in_sect_pr {
                        let name = e.name();
                        match name.as_ref() {
                            b"w:headerReference" | b"w:footerReference" => {
                                let mut ref_type = String::new();
                                let mut r_id = String::new();
                                for attr in e.attributes().flatten() {
                                    match attr.key.as_ref() {
                                        b"w:type" => {
                                            ref_type =
                                                String::from_utf8_lossy(&attr.value).to_string();
                                        }
                                        b"r:id" => {
                                            r_id = String::from_utf8_lossy(&attr.value).to_string();
                                        }
                                        _ => {}
                                    }
                                }
                                if ref_type == "default" && !r_id.is_empty() {
                                    if name.as_ref() == b"w:headerReference" {
                                        default_header_rid = Some(r_id);
                                    } else {
                                        default_footer_rid = Some(r_id);
                                    }
                                }
                            }
                            _ => {}
                        }
                    } else if in_paragraph {
                        let name = e.name();
                        paragraph_xml.push('<');
                        paragraph_xml.push_str(&String::from_utf8_lossy(name.as_ref()));
                        for attr in e.attributes().flatten() {
                            paragraph_xml.push_str(&format!(
                                " {}=\"{}\"",
                                String::from_utf8_lossy(attr.key.as_ref()),
                                String::from_utf8_lossy(&attr.value)
                            ));
                        }
                        paragraph_xml.push_str("/>");
                    } else if table_depth > 0 {
                        let name = e.name();
                        table_xml.push('<');
                        table_xml.push_str(&String::from_utf8_lossy(name.as_ref()));
                        for attr in e.attributes().flatten() {
                            table_xml.push_str(&format!(
                                " {}=\"{}\"",
                                String::from_utf8_lossy(attr.key.as_ref()),
                                String::from_utf8_lossy(&attr.value)
                            ));
                        }
                        table_xml.push_str("/>");
                    }
                }
                Ok(quick_xml::events::Event::Text(ref e)) => {
                    if in_paragraph {
                        let text = e.unescape().unwrap_or_default();
                        paragraph_xml.push_str(&escape_xml(&text));
                    } else if table_depth > 0 {
                        let text = e.unescape().unwrap_or_default();
                        table_xml.push_str(&escape_xml(&text));
                    }
                }
                Ok(quick_xml::events::Event::End(ref e)) => {
                    let name = e.name();
                    match name.as_ref() {
                        b"w:body" => {
                            in_body = false;
                        }
                        b"w:sectPr" if in_sect_pr => {
                            in_sect_pr = false;
                        }
                        b"w:p" if in_paragraph && table_depth == 0 && para_depth == 0 => {
                            paragraph_xml.push_str("</w:p>");
                            // Extract text box paragraphs before parsing the main paragraph
                            let textbox_paras = self.extract_textbox_paragraphs(&paragraph_xml);
                            if let Ok(para) = self.parse_paragraph(&paragraph_xml) {
                                section.add_block(Block::Paragraph(para));
                            }
                            // Add text box paragraphs as separate blocks
                            for tb_para in textbox_paras {
                                section.add_block(Block::Paragraph(tb_para));
                            }
                            in_paragraph = false;
                        }
                        b"w:tbl" if table_depth > 0 => {
                            table_xml.push_str("</w:tbl>");
                            table_depth -= 1;
                            if table_depth == 0 {
                                // Finished collecting outermost table - now parse it
                                if let Ok(table) = self.parse_table(&table_xml) {
                                    section.add_block(Block::Table(table));
                                }
                            }
                        }
                        _ => {
                            if in_paragraph {
                                // Track nested w:p depth for text boxes
                                if name.as_ref() == b"w:p" {
                                    para_depth = para_depth.saturating_sub(1);
                                }
                                paragraph_xml.push_str("</");
                                paragraph_xml.push_str(&String::from_utf8_lossy(name.as_ref()));
                                paragraph_xml.push('>');
                            } else if table_depth > 0 {
                                table_xml.push_str("</");
                                table_xml.push_str(&String::from_utf8_lossy(name.as_ref()));
                                table_xml.push('>');
                            }
                        }
                    }
                }
                Ok(quick_xml::events::Event::Eof) => break,
                Err(e) => {
                    return Err(Error::xml_parse_with_context(
                        e.to_string(),
                        "word/document.xml",
                    ))
                }
                _ => {}
            }
            buf.clear();
        }

        // Resolve and parse header/footer from sectPr references
        if let Some(rid) = default_header_rid {
            if let Some(paragraphs) = self.parse_header_footer_by_rid(&rid) {
                if !paragraphs.is_empty() {
                    section.header = Some(paragraphs);
                }
            }
        }
        if let Some(rid) = default_footer_rid {
            if let Some(paragraphs) = self.parse_header_footer_by_rid(&rid) {
                if !paragraphs.is_empty() {
                    section.footer = Some(paragraphs);
                }
            }
        }

        Ok(section)
    }

    /// Resolve a relationship ID to a header/footer XML path and parse its paragraphs.
    fn parse_header_footer_by_rid(&self, rid: &str) -> Option<Vec<Paragraph>> {
        let rel = self.relationships.get(rid)?;
        let path = OoxmlContainer::resolve_path("word/document.xml", &rel.target);
        let xml = self.container.read_xml(&path).ok()?;
        Some(parse_header_footer_xml(&xml))
    }

    /// Parse a single paragraph element.
    fn parse_paragraph(&mut self, xml: &str) -> Result<Paragraph> {
        use crate::model::InlineImage;

        let mut para = Paragraph::new();
        let mut reader = quick_xml::Reader::from_str(xml);
        // Don't trim text - preserve whitespace from xml:space="preserve" elements
        reader.config_mut().trim_text(false);

        let mut buf = Vec::new();
        let mut in_ppr = false;
        let mut in_rpr = false;
        let mut in_run = false;
        let mut in_text = false; // Track w:t elements (regular text)
        let mut in_instr_text = false; // Track w:instrText elements (field codes to skip)
        let mut in_drawing = false; // Track w:drawing elements for images
        let mut in_ins = false; // Track w:ins elements (tracked changes - insertions)
        let mut in_del = false; // Track w:del elements (tracked changes - deletions)
        let mut txbx_content_depth: u32 = 0; // Track w:txbxContent nesting (suppress text capture)
        let mut mc_fallback_depth: u32 = 0; // Track mc:Fallback nesting (skip entirely)
        let mut current_style = TextStyle::default();
        let mut current_hyperlink: Option<String> = None;
        let mut current_image_alt: Option<String> = None;

        loop {
            match reader.read_event_into(&mut buf) {
                Ok(quick_xml::events::Event::Start(ref e)) => match e.name().as_ref() {
                    // Skip mc:Fallback branches to avoid duplicating text box content
                    b"mc:Fallback" => {
                        mc_fallback_depth += 1;
                    }
                    // Track w:txbxContent to suppress text capture (extracted separately)
                    b"w:txbxContent" if mc_fallback_depth == 0 => {
                        txbx_content_depth += 1;
                    }
                    _ if mc_fallback_depth > 0 => {} // Skip everything inside mc:Fallback
                    _ if txbx_content_depth > 0 => {} // Skip everything inside w:txbxContent
                    b"w:pPr" => in_ppr = true,
                    b"w:rPr" => in_rpr = true,
                    b"w:r" => {
                        in_run = true;
                        current_style = TextStyle::default();
                    }
                    b"w:t" => in_text = true,
                    b"w:instrText" => in_instr_text = true,
                    b"w:drawing" => {
                        in_drawing = true;
                        current_image_alt = None;
                    }
                    // Tracked changes - insertions
                    b"w:ins" => in_ins = true,
                    // Tracked changes - deletions
                    b"w:del" => in_del = true,
                    b"w:hyperlink" => {
                        for attr in e.attributes().flatten() {
                            if attr.key.as_ref() == b"r:id" {
                                let rel_id = String::from_utf8_lossy(&attr.value);
                                if let Some(rel) = self.relationships.get(&rel_id) {
                                    current_hyperlink = Some(rel.target.clone());
                                }
                            }
                        }
                    }
                    _ => {}
                },
                Ok(quick_xml::events::Event::Empty(ref e)) => match e.name().as_ref() {
                    _ if mc_fallback_depth > 0 || txbx_content_depth > 0 => {} // Skip
                    b"w:pStyle" if in_ppr => {
                        for attr in e.attributes().flatten() {
                            if attr.key.as_ref() == b"w:val" {
                                let style_id = String::from_utf8_lossy(&attr.value);
                                para.style_id = Some(style_id.to_string());
                                para.heading = self.styles.get_heading_level(&style_id);
                                // Also get style name from StyleMap
                                if let Some(style) = self.styles.styles.get(style_id.as_ref()) {
                                    if !style.name.is_empty() {
                                        para.style_name = Some(style.name.clone());
                                    }
                                }
                            }
                        }
                    }
                    b"w:jc" if in_ppr => {
                        for attr in e.attributes().flatten() {
                            if attr.key.as_ref() == b"w:val" {
                                let val = String::from_utf8_lossy(&attr.value);
                                para.alignment = match val.as_ref() {
                                    "center" => TextAlignment::Center,
                                    "right" => TextAlignment::Right,
                                    "both" | "distribute" => TextAlignment::Justify,
                                    _ => TextAlignment::Left,
                                };
                            }
                        }
                    }
                    b"w:b" if in_rpr => {
                        let val = get_bool_attr(e, b"w:val");
                        current_style.bold = val.unwrap_or(true);
                    }
                    b"w:i" if in_rpr => {
                        let val = get_bool_attr(e, b"w:val");
                        current_style.italic = val.unwrap_or(true);
                    }
                    b"w:u" if in_rpr => {
                        for attr in e.attributes().flatten() {
                            if attr.key.as_ref() == b"w:val" {
                                let val = String::from_utf8_lossy(&attr.value);
                                current_style.underline = val != "none";
                            }
                        }
                    }
                    b"w:strike" if in_rpr => {
                        let val = get_bool_attr(e, b"w:val");
                        current_style.strikethrough = val.unwrap_or(true);
                    }
                    b"w:vertAlign" if in_rpr => {
                        for attr in e.attributes().flatten() {
                            if attr.key.as_ref() == b"w:val" {
                                let val = String::from_utf8_lossy(&attr.value);
                                match val.as_ref() {
                                    "superscript" => current_style.superscript = true,
                                    "subscript" => current_style.subscript = true,
                                    _ => {}
                                }
                            }
                        }
                    }
                    b"w:sz" if in_rpr => {
                        for attr in e.attributes().flatten() {
                            if attr.key.as_ref() == b"w:val" {
                                let val = String::from_utf8_lossy(&attr.value);
                                current_style.size = val.parse().ok();
                            }
                        }
                    }
                    b"w:color" if in_rpr => {
                        for attr in e.attributes().flatten() {
                            if attr.key.as_ref() == b"w:val" {
                                let val = String::from_utf8_lossy(&attr.value);
                                if val != "auto" {
                                    current_style.color = Some(val.to_string());
                                }
                            }
                        }
                    }
                    b"w:highlight" if in_rpr => {
                        for attr in e.attributes().flatten() {
                            if attr.key.as_ref() == b"w:val" {
                                current_style.highlight =
                                    Some(String::from_utf8_lossy(&attr.value).to_string());
                            }
                        }
                    }
                    b"w:rFonts" if in_rpr => {
                        for attr in e.attributes().flatten() {
                            if attr.key.as_ref() == b"w:ascii" {
                                current_style.font =
                                    Some(String::from_utf8_lossy(&attr.value).to_string());
                                break;
                            }
                        }
                    }
                    // Image handling: wp:docPr contains alt text
                    b"wp:docPr" if in_drawing => {
                        for attr in e.attributes().flatten() {
                            if attr.key.as_ref() == b"descr" {
                                current_image_alt =
                                    Some(String::from_utf8_lossy(&attr.value).to_string());
                            }
                        }
                    }
                    // Image handling: a:blip contains the image reference
                    b"a:blip" if in_drawing => {
                        for attr in e.attributes().flatten() {
                            if attr.key.as_ref() == b"r:embed" {
                                let rel_id = String::from_utf8_lossy(&attr.value).to_string();
                                // Create inline image with the relationship ID
                                let image = InlineImage {
                                    resource_id: rel_id,
                                    alt_text: current_image_alt.clone(),
                                    width: None,
                                    height: None,
                                };
                                para.images.push(image);
                            }
                        }
                    }
                    // Break handling - line break or page break
                    b"w:br" if in_run => {
                        // Check for break type: page, column, or text wrapping (default)
                        let mut is_page_break = false;
                        for attr in e.attributes().flatten() {
                            if attr.key.as_ref() == b"w:type" {
                                let break_type = String::from_utf8_lossy(&attr.value);
                                is_page_break = break_type == "page";
                            }
                        }

                        // Compute current revision type
                        let current_revision = if in_del {
                            RevisionType::Deleted
                        } else if in_ins {
                            RevisionType::Inserted
                        } else {
                            RevisionType::None
                        };

                        if is_page_break {
                            // Page break - mark run with page_break flag
                            if let Some(last_run) = para.runs.last_mut() {
                                last_run.page_break = true;
                            } else {
                                para.runs.push(TextRun {
                                    text: String::new(),
                                    style: current_style.clone(),
                                    hyperlink: None,
                                    line_break: false,
                                    page_break: true,
                                    revision: current_revision,
                                });
                            }
                        } else {
                            // Line break (text wrapping or column break treated as line break)
                            if let Some(last_run) = para.runs.last_mut() {
                                last_run.line_break = true;
                            } else {
                                para.runs.push(TextRun {
                                    text: String::new(),
                                    style: current_style.clone(),
                                    hyperlink: None,
                                    line_break: true,
                                    page_break: false,
                                    revision: current_revision,
                                });
                            }
                        }
                    }
                    // Tab character handling - convert <w:tab/> to tab character
                    b"w:tab" if in_run => {
                        let current_revision = if in_del {
                            RevisionType::Deleted
                        } else if in_ins {
                            RevisionType::Inserted
                        } else {
                            RevisionType::None
                        };
                        para.runs.push(TextRun {
                            text: "\t".to_string(),
                            style: current_style.clone(),
                            hyperlink: current_hyperlink.clone(),
                            line_break: false,
                            page_break: false,
                            revision: current_revision,
                        });
                    }
                    // Carriage return handling - convert <w:cr/> to newline
                    b"w:cr" if in_run => {
                        if let Some(last_run) = para.runs.last_mut() {
                            last_run.line_break = true;
                        } else {
                            let current_revision = if in_del {
                                RevisionType::Deleted
                            } else if in_ins {
                                RevisionType::Inserted
                            } else {
                                RevisionType::None
                            };
                            para.runs.push(TextRun {
                                text: String::new(),
                                style: current_style.clone(),
                                hyperlink: None,
                                line_break: true,
                                page_break: false,
                                revision: current_revision,
                            });
                        }
                    }
                    // Non-breaking hyphen handling
                    b"w:noBreakHyphen" if in_run => {
                        let current_revision = if in_del {
                            RevisionType::Deleted
                        } else if in_ins {
                            RevisionType::Inserted
                        } else {
                            RevisionType::None
                        };
                        para.runs.push(TextRun {
                            text: "\u{2011}".to_string(), // Non-breaking hyphen Unicode
                            style: current_style.clone(),
                            hyperlink: current_hyperlink.clone(),
                            line_break: false,
                            page_break: false,
                            revision: current_revision,
                        });
                    }
                    // Soft hyphen handling (optional hyphen, usually invisible)
                    b"w:softHyphen" if in_run => {
                        let current_revision = if in_del {
                            RevisionType::Deleted
                        } else if in_ins {
                            RevisionType::Inserted
                        } else {
                            RevisionType::None
                        };
                        para.runs.push(TextRun {
                            text: "\u{00AD}".to_string(), // Soft hyphen Unicode
                            style: current_style.clone(),
                            hyperlink: current_hyperlink.clone(),
                            line_break: false,
                            page_break: false,
                            revision: current_revision,
                        });
                    }
                    // Non-breaking space handling
                    b"w:noBreakSpace" if in_run => {
                        let current_revision = if in_del {
                            RevisionType::Deleted
                        } else if in_ins {
                            RevisionType::Inserted
                        } else {
                            RevisionType::None
                        };
                        para.runs.push(TextRun {
                            text: "\u{00A0}".to_string(), // Non-breaking space Unicode
                            style: current_style.clone(),
                            hyperlink: current_hyperlink.clone(),
                            line_break: false,
                            page_break: false,
                            revision: current_revision,
                        });
                    }
                    // Footnote reference handling
                    b"w:footnoteReference" if in_run => {
                        for attr in e.attributes().flatten() {
                            if attr.key.as_ref() == b"w:id" {
                                let id = String::from_utf8_lossy(&attr.value).to_string();
                                // Only insert marker if this footnote has content
                                if self.footnotes.contains_key(&id) {
                                    para.runs.push(TextRun::plain(format!("[^{}]", id)));
                                }
                            }
                        }
                    }
                    // Endnote reference handling
                    b"w:endnoteReference" if in_run => {
                        for attr in e.attributes().flatten() {
                            if attr.key.as_ref() == b"w:id" {
                                let id = String::from_utf8_lossy(&attr.value).to_string();
                                // Only insert marker if this endnote has content
                                if self.endnotes.contains_key(&id) {
                                    para.runs.push(TextRun::plain(format!("[^e{}]", id)));
                                }
                            }
                        }
                    }
                    _ => {}
                },
                Ok(quick_xml::events::Event::Text(ref e)) => {
                    // Only extract text from w:t elements, skip w:instrText (field codes)
                    // Also skip text inside mc:Fallback and w:txbxContent (extracted separately)
                    if in_run
                        && in_text
                        && !in_instr_text
                        && mc_fallback_depth == 0
                        && txbx_content_depth == 0
                    {
                        let text = e.unescape().unwrap_or_default().to_string();
                        if !text.is_empty() {
                            let current_revision = if in_del {
                                RevisionType::Deleted
                            } else if in_ins {
                                RevisionType::Inserted
                            } else {
                                RevisionType::None
                            };
                            let run = TextRun {
                                text,
                                style: current_style.clone(),
                                hyperlink: current_hyperlink.clone(),
                                line_break: false,
                                page_break: false,
                                revision: current_revision,
                            };
                            para.runs.push(run);
                        }
                    }
                }
                Ok(quick_xml::events::Event::End(ref e)) => match e.name().as_ref() {
                    b"mc:Fallback" if mc_fallback_depth > 0 => {
                        mc_fallback_depth -= 1;
                    }
                    b"w:txbxContent" if txbx_content_depth > 0 => {
                        txbx_content_depth -= 1;
                    }
                    _ if mc_fallback_depth > 0 || txbx_content_depth > 0 => {} // Skip
                    b"w:pPr" => in_ppr = false,
                    b"w:rPr" => in_rpr = false,
                    b"w:r" => in_run = false,
                    b"w:t" => in_text = false,
                    b"w:instrText" => in_instr_text = false,
                    b"w:hyperlink" => current_hyperlink = None,
                    b"w:drawing" => {
                        in_drawing = false;
                        current_image_alt = None;
                    }
                    b"w:ins" => in_ins = false,
                    b"w:del" => in_del = false,
                    _ => {}
                },
                Ok(quick_xml::events::Event::Eof) => break,
                Err(e) => return Err(Error::xml_parse_with_context(e.to_string(), "paragraph")),
                _ => {}
            }
            buf.clear();
        }

        // Parse numbering (list info)
        para.list_info = self.parse_list_info(xml);

        Ok(para)
    }

    /// Extract paragraphs from `w:txbxContent` elements (text boxes/shapes).
    ///
    /// Text boxes in DOCX appear inside `w:drawing` or `mc:AlternateContent` elements.
    /// This method finds all `w:txbxContent` sections (skipping those inside `mc:Fallback`
    /// to avoid duplication) and parses the inner `<w:p>` elements as regular paragraphs.
    fn extract_textbox_paragraphs(&mut self, xml: &str) -> Vec<Paragraph> {
        let mut paragraphs = Vec::new();
        let mut reader = quick_xml::Reader::from_str(xml);
        reader.config_mut().trim_text(false);

        let mut buf = Vec::new();
        let mut mc_fallback_depth: u32 = 0;
        let mut txbx_content_depth: u32 = 0;
        let mut in_txbx_para = false;
        let mut txbx_para_xml = String::new();
        let mut txbx_para_depth: u32 = 0; // Track nested elements inside the text box <w:p>

        loop {
            match reader.read_event_into(&mut buf) {
                Ok(quick_xml::events::Event::Start(ref e)) => {
                    let name = e.name();
                    match name.as_ref() {
                        b"mc:Fallback" => {
                            mc_fallback_depth += 1;
                        }
                        b"w:txbxContent" if mc_fallback_depth == 0 => {
                            txbx_content_depth += 1;
                        }
                        b"w:p" if txbx_content_depth > 0 && !in_txbx_para => {
                            in_txbx_para = true;
                            txbx_para_depth = 0;
                            txbx_para_xml.clear();
                            txbx_para_xml.push_str("<w:p");
                            for attr in e.attributes().flatten() {
                                txbx_para_xml.push_str(&format!(
                                    " {}=\"{}\"",
                                    String::from_utf8_lossy(attr.key.as_ref()),
                                    String::from_utf8_lossy(&attr.value)
                                ));
                            }
                            txbx_para_xml.push('>');
                        }
                        _ if in_txbx_para => {
                            txbx_para_depth += 1;
                            txbx_para_xml.push('<');
                            txbx_para_xml.push_str(&String::from_utf8_lossy(name.as_ref()));
                            for attr in e.attributes().flatten() {
                                txbx_para_xml.push_str(&format!(
                                    " {}=\"{}\"",
                                    String::from_utf8_lossy(attr.key.as_ref()),
                                    String::from_utf8_lossy(&attr.value)
                                ));
                            }
                            txbx_para_xml.push('>');
                        }
                        _ => {}
                    }
                }
                Ok(quick_xml::events::Event::Empty(ref e)) => {
                    if in_txbx_para {
                        let name = e.name();
                        txbx_para_xml.push('<');
                        txbx_para_xml.push_str(&String::from_utf8_lossy(name.as_ref()));
                        for attr in e.attributes().flatten() {
                            txbx_para_xml.push_str(&format!(
                                " {}=\"{}\"",
                                String::from_utf8_lossy(attr.key.as_ref()),
                                String::from_utf8_lossy(&attr.value)
                            ));
                        }
                        txbx_para_xml.push_str("/>");
                    }
                }
                Ok(quick_xml::events::Event::Text(ref e)) => {
                    if in_txbx_para {
                        let text = e.unescape().unwrap_or_default();
                        txbx_para_xml.push_str(&escape_xml(&text));
                    }
                }
                Ok(quick_xml::events::Event::End(ref e)) => {
                    let name = e.name();
                    match name.as_ref() {
                        b"mc:Fallback" if mc_fallback_depth > 0 => {
                            mc_fallback_depth -= 1;
                        }
                        b"w:txbxContent" if txbx_content_depth > 0 => {
                            txbx_content_depth -= 1;
                        }
                        b"w:p" if in_txbx_para && txbx_para_depth == 0 => {
                            txbx_para_xml.push_str("</w:p>");
                            if let Ok(para) = self.parse_paragraph(&txbx_para_xml) {
                                if !para.plain_text().is_empty() {
                                    paragraphs.push(para);
                                }
                            }
                            in_txbx_para = false;
                        }
                        _ if in_txbx_para => {
                            txbx_para_depth = txbx_para_depth.saturating_sub(1);
                            txbx_para_xml.push_str("</");
                            txbx_para_xml.push_str(&String::from_utf8_lossy(name.as_ref()));
                            txbx_para_xml.push('>');
                        }
                        _ => {}
                    }
                }
                Ok(quick_xml::events::Event::Eof) => break,
                Err(_) => break,
                _ => {}
            }
            buf.clear();
        }

        paragraphs
    }

    /// Parse list info from paragraph XML.
    fn parse_list_info(&mut self, xml: &str) -> Option<ListInfo> {
        let mut reader = quick_xml::Reader::from_str(xml);
        reader.config_mut().trim_text(true);

        let mut buf = Vec::new();
        let mut num_id: Option<String> = None;
        let mut level: u8 = 0;
        let mut in_num_pr = false;

        loop {
            match reader.read_event_into(&mut buf) {
                Ok(quick_xml::events::Event::Start(ref e)) => {
                    if e.name().as_ref() == b"w:numPr" {
                        in_num_pr = true;
                    }
                }
                Ok(quick_xml::events::Event::Empty(ref e)) => {
                    if in_num_pr {
                        match e.name().as_ref() {
                            b"w:numId" => {
                                for attr in e.attributes().flatten() {
                                    if attr.key.as_ref() == b"w:val" {
                                        num_id =
                                            Some(String::from_utf8_lossy(&attr.value).to_string());
                                    }
                                }
                            }
                            b"w:ilvl" => {
                                for attr in e.attributes().flatten() {
                                    if attr.key.as_ref() == b"w:val" {
                                        let val = String::from_utf8_lossy(&attr.value);
                                        level = val.parse().unwrap_or(0);
                                    }
                                }
                            }
                            _ => {}
                        }
                    }
                }
                Ok(quick_xml::events::Event::End(ref e)) => {
                    if e.name().as_ref() == b"w:numPr" {
                        in_num_pr = false;
                    }
                }
                Ok(quick_xml::events::Event::Eof) => break,
                _ => {}
            }
            buf.clear();
        }

        if let Some(ref nid) = num_id {
            if let Some((list_type, number)) = self.numbering.get_list_info(nid, level) {
                return Some(ListInfo {
                    list_type,
                    level,
                    number: if list_type == ListType::Numbered {
                        Some(number)
                    } else {
                        None
                    },
                });
            }
        }

        None
    }

    /// Parse a table element.
    #[allow(clippy::only_used_in_recursion)] // &self needed for recursive nested table parsing
    fn parse_table(&self, xml: &str) -> Result<Table> {
        use crate::model::InlineImage;

        let mut table = Table::new();
        let mut reader = quick_xml::Reader::from_str(xml);
        // Don't trim text - preserve whitespace from xml:space="preserve" elements
        reader.config_mut().trim_text(false);

        let mut buf = Vec::new();
        let mut in_row = false;
        let mut in_cell = false;
        let mut in_paragraph = false;
        let mut in_run = false;
        let mut in_rpr = false; // Track w:rPr (run properties for formatting)
        let mut in_text = false; // Track w:t elements (regular text)
        let mut in_instr_text = false; // Track w:instrText elements (field codes to skip)
        let mut in_drawing = false; // Track w:drawing elements for images
        let mut current_image_alt: Option<String> = None;
        let mut current_row: Option<Row> = None;
        let mut cell_paragraphs: Vec<Paragraph> = Vec::new();
        let mut cell_nested_tables: Vec<Table> = Vec::new();
        let mut current_paragraph: Option<Paragraph> = None;
        let mut current_style = TextStyle::default();
        let mut is_header_row = false;
        let mut col_span = 1u32;
        let mut row_span = 1u32;
        let mut cell_alignment = CellAlignment::Left;
        let mut in_tc_pr = false; // Track w:tcPr (table cell properties)

        // Track nested table depth (0 = we're at the main table level)
        // 1+ = we're inside a nested table and should collect its XML
        let mut nested_table_depth: u32 = 0;
        let mut nested_table_xml = String::new();

        loop {
            match reader.read_event_into(&mut buf) {
                Ok(quick_xml::events::Event::Start(ref e)) => {
                    let name = e.name();

                    // If we're inside a nested table, just collect XML
                    if nested_table_depth > 0 {
                        nested_table_xml.push('<');
                        nested_table_xml.push_str(&String::from_utf8_lossy(name.as_ref()));
                        for attr in e.attributes().flatten() {
                            nested_table_xml.push_str(&format!(
                                " {}=\"{}\"",
                                String::from_utf8_lossy(attr.key.as_ref()),
                                String::from_utf8_lossy(&attr.value)
                            ));
                        }
                        nested_table_xml.push('>');
                        if name.as_ref() == b"w:tbl" {
                            nested_table_depth += 1;
                        }
                        continue;
                    }

                    match name.as_ref() {
                        b"w:tbl" if in_cell => {
                            // Start collecting nested table
                            nested_table_depth = 1;
                            nested_table_xml.clear();
                            nested_table_xml.push_str("<w:tbl>");
                        }
                        b"w:tr" => {
                            in_row = true;
                            current_row = Some(Row {
                                cells: Vec::new(),
                                is_header: false,
                                height: None,
                            });
                            is_header_row = false;
                        }
                        b"w:tc" => {
                            in_cell = true;
                            cell_paragraphs.clear();
                            cell_nested_tables.clear();
                            col_span = 1;
                            row_span = 1;
                            cell_alignment = CellAlignment::Left;
                        }
                        b"w:tcPr" if in_cell => {
                            in_tc_pr = true;
                        }
                        b"w:p" if in_cell => {
                            in_paragraph = true;
                            current_paragraph = Some(Paragraph::new());
                        }
                        b"w:r" if in_paragraph => {
                            in_run = true;
                            current_style = TextStyle::default();
                        }
                        b"w:rPr" if in_run => in_rpr = true,
                        b"w:t" => in_text = true,
                        b"w:instrText" => in_instr_text = true,
                        b"w:drawing" => {
                            in_drawing = true;
                            current_image_alt = None;
                        }
                        _ => {}
                    }
                }
                Ok(quick_xml::events::Event::Empty(ref e)) => {
                    let name = e.name();

                    // If we're inside a nested table, just collect XML
                    if nested_table_depth > 0 {
                        nested_table_xml.push('<');
                        nested_table_xml.push_str(&String::from_utf8_lossy(name.as_ref()));
                        for attr in e.attributes().flatten() {
                            nested_table_xml.push_str(&format!(
                                " {}=\"{}\"",
                                String::from_utf8_lossy(attr.key.as_ref()),
                                String::from_utf8_lossy(&attr.value)
                            ));
                        }
                        nested_table_xml.push_str("/>");
                        continue;
                    }

                    match name.as_ref() {
                        b"w:tblHeader" if in_row => {
                            is_header_row = true;
                        }
                        b"w:gridSpan" if in_cell => {
                            for attr in e.attributes().flatten() {
                                if attr.key.as_ref() == b"w:val" {
                                    let val = String::from_utf8_lossy(&attr.value);
                                    col_span = val.parse().unwrap_or(1);
                                }
                            }
                        }
                        b"w:vMerge" if in_cell => {
                            let mut has_val = false;
                            for attr in e.attributes().flatten() {
                                if attr.key.as_ref() == b"w:val" {
                                    has_val = true;
                                }
                            }
                            if !has_val {
                                row_span = 0;
                            }
                        }
                        b"w:jc" if in_tc_pr => {
                            for attr in e.attributes().flatten() {
                                if attr.key.as_ref() == b"w:val" {
                                    let val = String::from_utf8_lossy(&attr.value);
                                    cell_alignment = match val.as_ref() {
                                        "center" => CellAlignment::Center,
                                        "right" | "end" => CellAlignment::Right,
                                        _ => CellAlignment::Left,
                                    };
                                }
                            }
                        }
                        // Handle formatting in run properties
                        b"w:b" if in_rpr => {
                            let val = get_bool_attr(e, b"w:val");
                            current_style.bold = val.unwrap_or(true);
                        }
                        b"w:i" if in_rpr => {
                            let val = get_bool_attr(e, b"w:val");
                            current_style.italic = val.unwrap_or(true);
                        }
                        b"w:u" if in_rpr => {
                            for attr in e.attributes().flatten() {
                                if attr.key.as_ref() == b"w:val" {
                                    let val = String::from_utf8_lossy(&attr.value);
                                    current_style.underline = val != "none";
                                }
                            }
                        }
                        b"w:strike" if in_rpr => {
                            let val = get_bool_attr(e, b"w:val");
                            current_style.strikethrough = val.unwrap_or(true);
                        }
                        // Image handling: wp:docPr contains alt text
                        b"wp:docPr" if in_drawing => {
                            for attr in e.attributes().flatten() {
                                if attr.key.as_ref() == b"descr" {
                                    current_image_alt =
                                        Some(String::from_utf8_lossy(&attr.value).to_string());
                                }
                            }
                        }
                        // Image handling: a:blip contains the image reference
                        b"a:blip" if in_drawing => {
                            for attr in e.attributes().flatten() {
                                if attr.key.as_ref() == b"r:embed" {
                                    let rel_id = String::from_utf8_lossy(&attr.value).to_string();
                                    // Create inline image with the relationship ID
                                    let image = InlineImage {
                                        resource_id: rel_id,
                                        alt_text: current_image_alt.clone(),
                                        width: None,
                                        height: None,
                                    };
                                    if let Some(ref mut para) = current_paragraph {
                                        para.images.push(image);
                                    }
                                }
                            }
                        }
                        _ => {}
                    }
                }
                Ok(quick_xml::events::Event::Text(ref e)) => {
                    // If we're inside a nested table, just collect XML
                    if nested_table_depth > 0 {
                        let text = e.unescape().unwrap_or_default();
                        nested_table_xml.push_str(&escape_xml(&text));
                        continue;
                    }

                    // Only extract text from w:t elements, skip w:instrText (field codes)
                    if in_run && in_text && !in_instr_text {
                        let text = e.unescape().unwrap_or_default().to_string();
                        if !text.is_empty() {
                            if let Some(ref mut para) = current_paragraph {
                                let run = TextRun {
                                    text,
                                    style: current_style.clone(),
                                    hyperlink: None,
                                    line_break: false,
                                    page_break: false,
                                    revision: RevisionType::None,
                                };
                                para.runs.push(run);
                            }
                        }
                    }
                }
                Ok(quick_xml::events::Event::End(ref e)) => {
                    let name = e.name();

                    // If we're inside a nested table, collect XML and check for end
                    if nested_table_depth > 0 {
                        if name.as_ref() == b"w:tbl" {
                            nested_table_xml.push_str("</w:tbl>");
                            nested_table_depth -= 1;
                            if nested_table_depth == 0 {
                                // Finished collecting nested table - parse recursively
                                if let Ok(nested_table) = self.parse_table(&nested_table_xml) {
                                    cell_nested_tables.push(nested_table);
                                }
                            }
                        } else {
                            nested_table_xml.push_str("</");
                            nested_table_xml.push_str(&String::from_utf8_lossy(name.as_ref()));
                            nested_table_xml.push('>');
                        }
                        continue;
                    }

                    match name.as_ref() {
                        b"w:tr" => {
                            if let Some(mut row) = current_row.take() {
                                row.is_header = is_header_row;
                                table.add_row(row);
                            }
                            in_row = false;
                        }
                        b"w:tcPr" => {
                            in_tc_pr = false;
                        }
                        b"w:tc" => {
                            if row_span > 0 {
                                // Use collected paragraphs, or empty paragraph if none
                                // Deduplicate repeated paragraph blocks within cell
                                // Word may store the same paragraph block twice but only displays once
                                let content = if cell_paragraphs.is_empty() {
                                    vec![Paragraph::new()]
                                } else {
                                    let paragraphs = std::mem::take(&mut cell_paragraphs);
                                    deduplicate_paragraph_block(paragraphs)
                                };
                                let cell = Cell {
                                    content,
                                    nested_tables: std::mem::take(&mut cell_nested_tables),
                                    col_span,
                                    row_span,
                                    alignment: cell_alignment,
                                    vertical_alignment: VerticalAlignment::default(),
                                    is_header: is_header_row,
                                    background: None,
                                };
                                if let Some(ref mut row) = current_row {
                                    row.cells.push(cell);
                                }
                            }
                            in_cell = false;
                        }
                        b"w:p" if in_cell => {
                            // Save the completed paragraph
                            if let Some(para) = current_paragraph.take() {
                                // Only add non-empty paragraphs
                                if !para.is_empty() {
                                    // Skip duplicate paragraphs (same text content as previous)
                                    // Word may store duplicate paragraphs in same cell but only displays one
                                    let is_duplicate = cell_paragraphs
                                        .last()
                                        .map(|last| last.plain_text() == para.plain_text())
                                        .unwrap_or(false);

                                    if !is_duplicate {
                                        cell_paragraphs.push(para);
                                    }
                                }
                            }
                            in_paragraph = false;
                        }
                        b"w:r" => {
                            in_run = false;
                        }
                        b"w:rPr" => in_rpr = false,
                        b"w:t" => in_text = false,
                        b"w:instrText" => in_instr_text = false,
                        b"w:drawing" => {
                            in_drawing = false;
                            current_image_alt = None;
                        }
                        _ => {}
                    }
                }
                Ok(quick_xml::events::Event::Eof) => break,
                Err(e) => return Err(Error::xml_parse_with_context(e.to_string(), "table")),
                _ => {}
            }
            buf.clear();
        }

        Ok(table)
    }

    /// Extract embedded resources (images, etc.).
    fn extract_resources(&self, doc: &mut Document) -> Result<()> {
        for (id, rel) in &self.relationships.by_id {
            if rel.rel_type.contains("/image") && !rel.external {
                let path = OoxmlContainer::resolve_path("word/document.xml", &rel.target);
                if let Ok(data) = self.container.read_binary(&path) {
                    let size = data.len();
                    let ext = std::path::Path::new(&path)
                        .extension()
                        .and_then(|e| e.to_str())
                        .unwrap_or("");
                    let resource = Resource {
                        resource_type: ResourceType::from_extension(ext),
                        filename: Some(
                            std::path::Path::new(&path)
                                .file_name()
                                .unwrap_or_default()
                                .to_string_lossy()
                                .to_string(),
                        ),
                        mime_type: guess_mime_type(&path),
                        data,
                        size,
                        width: None,
                        height: None,
                        alt_text: None,
                    };
                    doc.resources.insert(id.clone(), resource);
                }
            }
        }

        Ok(())
    }

    /// Get a reference to the container.
    pub fn container(&self) -> &OoxmlContainer {
        &self.container
    }
}

/// Parse footnotes.xml or endnotes.xml into a map of id → plain text.
///
/// `note_tag` should be `b"w:footnote"` or `b"w:endnote"`.
/// Entries with `w:type="separator"` or `w:type="continuationSeparator"` are skipped.
fn parse_notes_xml(xml: &str, note_tag: &[u8]) -> HashMap<String, String> {
    let mut notes = HashMap::new();
    let mut reader = quick_xml::Reader::from_str(xml);
    reader.config_mut().trim_text(false);

    let mut buf = Vec::new();
    let mut current_id: Option<String> = None;
    let mut current_text = String::new();
    let mut in_note = false;
    let mut in_text = false;

    loop {
        match reader.read_event_into(&mut buf) {
            Ok(quick_xml::events::Event::Start(ref e)) => {
                if e.name().as_ref() == note_tag {
                    let mut id = None;
                    let mut note_type = None;
                    for attr in e.attributes().flatten() {
                        match attr.key.as_ref() {
                            b"w:id" => {
                                id = Some(String::from_utf8_lossy(&attr.value).to_string());
                            }
                            b"w:type" => {
                                note_type = Some(String::from_utf8_lossy(&attr.value).to_string());
                            }
                            _ => {}
                        }
                    }
                    // Skip separator and continuationSeparator types
                    if let Some(ref t) = note_type {
                        if t == "separator" || t == "continuationSeparator" {
                            // Don't set in_note, so content is ignored
                            buf.clear();
                            continue;
                        }
                    }
                    if let Some(id_val) = id {
                        in_note = true;
                        current_id = Some(id_val);
                        current_text.clear();
                    }
                } else if in_note && e.name().as_ref() == b"w:t" {
                    in_text = true;
                }
            }
            Ok(quick_xml::events::Event::Text(ref e)) => {
                if in_note && in_text {
                    if let Ok(text) = e.unescape() {
                        current_text.push_str(&text);
                    }
                }
            }
            Ok(quick_xml::events::Event::End(ref e)) => {
                if e.name().as_ref() == note_tag {
                    if in_note {
                        if let Some(id) = current_id.take() {
                            let trimmed = current_text.trim().to_string();
                            if !trimmed.is_empty() {
                                notes.insert(id, trimmed);
                            }
                        }
                        in_note = false;
                    }
                } else if e.name().as_ref() == b"w:t" {
                    in_text = false;
                }
            }
            Ok(quick_xml::events::Event::Eof) => break,
            Err(_) => break,
            _ => {}
        }
        buf.clear();
    }

    notes
}

/// Parse a header or footer XML file (w:hdr or w:ftr) into a list of paragraphs.
///
/// Header/footer XML has the same structure as the document body:
/// `<w:hdr>` or `<w:ftr>` containing `<w:p>` paragraphs with `<w:r>` runs and `<w:t>` text.
/// This function extracts plain text only (no styles or formatting).
fn parse_header_footer_xml(xml: &str) -> Vec<Paragraph> {
    let mut paragraphs = Vec::new();
    let mut reader = quick_xml::Reader::from_str(xml);
    reader.config_mut().trim_text(false);

    let mut buf = Vec::new();
    let mut in_paragraph = false;
    let mut in_text = false;
    let mut current_text = String::new();

    loop {
        match reader.read_event_into(&mut buf) {
            Ok(quick_xml::events::Event::Start(ref e)) => match e.name().as_ref() {
                b"w:p" => {
                    in_paragraph = true;
                    current_text.clear();
                }
                b"w:t" if in_paragraph => {
                    in_text = true;
                }
                _ => {}
            },
            Ok(quick_xml::events::Event::Text(ref e)) => {
                if in_paragraph && in_text {
                    if let Ok(text) = e.unescape() {
                        current_text.push_str(&text);
                    }
                }
            }
            Ok(quick_xml::events::Event::End(ref e)) => match e.name().as_ref() {
                b"w:p" => {
                    if in_paragraph {
                        let trimmed = current_text.trim().to_string();
                        if !trimmed.is_empty() {
                            paragraphs.push(Paragraph::with_text(trimmed));
                        }
                        in_paragraph = false;
                    }
                }
                b"w:t" => {
                    in_text = false;
                }
                _ => {}
            },
            Ok(quick_xml::events::Event::Eof) => break,
            Err(_) => break,
            _ => {}
        }
        buf.clear();
    }

    paragraphs
}

/// Helper to get a boolean attribute value.
fn get_bool_attr(e: &quick_xml::events::BytesStart, key: &[u8]) -> Option<bool> {
    for attr in e.attributes().flatten() {
        if attr.key.as_ref() == key {
            let val = String::from_utf8_lossy(&attr.value);
            return Some(val != "0" && val != "false");
        }
    }
    None
}

/// Escape XML special characters.
fn escape_xml(s: &str) -> String {
    s.replace('&', "&amp;")
        .replace('<', "&lt;")
        .replace('>', "&gt;")
        .replace('"', "&quot;")
        .replace('\'', "&apos;")
}

/// Guess MIME type from file extension.
fn guess_mime_type(path: &str) -> Option<String> {
    let ext = std::path::Path::new(path)
        .extension()
        .and_then(|e| e.to_str())
        .map(|e| e.to_lowercase())?;

    Some(
        match ext.as_str() {
            "png" => "image/png",
            "jpg" | "jpeg" => "image/jpeg",
            "gif" => "image/gif",
            "bmp" => "image/bmp",
            "tiff" | "tif" => "image/tiff",
            "svg" => "image/svg+xml",
            "emf" => "image/x-emf",
            "wmf" => "image/x-wmf",
            _ => return None,
        }
        .to_string(),
    )
}

/// Deduplicate repeated paragraph blocks within a table cell.
/// Word may store the same paragraph block twice but only displays one.
/// This function checks if the first half and second half are identical
/// and returns only the first half if so.
fn deduplicate_paragraph_block(paragraphs: Vec<Paragraph>) -> Vec<Paragraph> {
    let len = paragraphs.len();
    if len < 2 {
        return paragraphs;
    }

    // Check if paragraphs form a duplicated block (first half == second half)
    if len.is_multiple_of(2) {
        let half = len / 2;
        let first_half = &paragraphs[..half];
        let second_half = &paragraphs[half..];

        let is_duplicate = first_half
            .iter()
            .zip(second_half.iter())
            .all(|(a, b)| a.plain_text() == b.plain_text());

        if is_duplicate {
            return paragraphs.into_iter().take(half).collect();
        }
    }

    paragraphs
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_open_docx() {
        let path = "test-files/file-sample_1MB.docx";
        if std::path::Path::new(path).exists() {
            let parser = DocxParser::open(path);
            assert!(parser.is_ok());
        }
    }

    #[test]
    fn test_parse_docx() {
        let path = "test-files/file-sample_1MB.docx";
        if std::path::Path::new(path).exists() {
            let mut parser = DocxParser::open(path).unwrap();
            let doc = parser.parse().unwrap();

            assert!(!doc.sections.is_empty());

            let text = doc.plain_text();
            assert!(!text.is_empty());
            assert!(text.contains("Lorem ipsum"));
        }
    }

    #[test]
    fn test_parse_headings() {
        let path = "test-files/file-sample_1MB.docx";
        if std::path::Path::new(path).exists() {
            let mut parser = DocxParser::open(path).unwrap();
            let doc = parser.parse().unwrap();

            let headings: Vec<_> = doc.sections[0]
                .content
                .iter()
                .filter_map(|block| {
                    if let Block::Paragraph(p) = block {
                        if p.is_heading() {
                            return Some(p);
                        }
                    }
                    None
                })
                .collect();

            assert!(!headings.is_empty());
        }
    }

    #[test]
    fn test_extract_resources() {
        let path = "test-files/file-sample_1MB.docx";
        if std::path::Path::new(path).exists() {
            let mut parser = DocxParser::open(path).unwrap();
            let doc = parser.parse().unwrap();

            if !doc.resources.is_empty() {
                let resource = doc.resources.values().next().unwrap();
                assert!(resource.is_image());
            }
        }
    }

    // =========================================================================
    // Whitespace Preservation Tests (GitHub Issue #2)
    // =========================================================================

    #[test]
    fn test_whitespace_preserved_between_runs() {
        // Test case from GitHub Issue #2: "DATE OF BIRTH" was becoming "DATEOFBIRTH"
        // When xml:space="preserve" is used, spaces must be preserved
        let xml = r#"<w:p>
            <w:r><w:t>DATE</w:t></w:r>
            <w:r><w:t xml:space="preserve"> </w:t></w:r>
            <w:r><w:t>OF</w:t></w:r>
            <w:r><w:t xml:space="preserve"> </w:t></w:r>
            <w:r><w:t>BIRTH</w:t></w:r>
        </w:p>"#;

        // Create a minimal container just for testing paragraph parsing
        let container = crate::container::OoxmlContainer::from_bytes(Vec::new());
        if container.is_err() {
            // Can't create empty container, skip test
            return;
        }
        let container = container.unwrap();
        let mut parser = DocxParser {
            container,
            styles: StyleMap::default(),
            numbering: NumberingMap::default(),
            relationships: crate::container::Relationships::default(),
            footnotes: HashMap::new(),
            endnotes: HashMap::new(),
        };

        let para = parser.parse_paragraph(xml).unwrap();
        let text = para.plain_text();

        // The text should have spaces preserved
        assert!(
            text.contains("DATE") && text.contains("OF") && text.contains("BIRTH"),
            "Expected 'DATE OF BIRTH' with spaces, got: '{}'",
            text
        );
        // Check that spaces are actually there
        assert!(
            text.contains(' '),
            "Expected spaces between words, got: '{}'",
            text
        );
    }

    #[test]
    fn test_whitespace_leading_trailing_preserved() {
        // Test leading/trailing whitespace with xml:space="preserve"
        let xml = r#"<w:p>
            <w:r><w:t xml:space="preserve">  Hello World  </w:t></w:r>
        </w:p>"#;

        let container = crate::container::OoxmlContainer::from_bytes(Vec::new());
        if container.is_err() {
            return;
        }
        let container = container.unwrap();
        let mut parser = DocxParser {
            container,
            styles: StyleMap::default(),
            numbering: NumberingMap::default(),
            relationships: crate::container::Relationships::default(),
            footnotes: HashMap::new(),
            endnotes: HashMap::new(),
        };

        let para = parser.parse_paragraph(xml).unwrap();
        let text = para.plain_text();

        // Leading/trailing spaces should be preserved
        assert!(
            text.starts_with("  ") || text.contains("  Hello"),
            "Expected leading spaces, got: '{}'",
            text
        );
    }

    #[test]
    fn test_tab_character_handling() {
        // Test <w:tab/> element is converted to tab character
        let xml = r#"<w:p>
            <w:r>
                <w:t>Column1</w:t>
            </w:r>
            <w:r>
                <w:tab/>
            </w:r>
            <w:r>
                <w:t>Column2</w:t>
            </w:r>
        </w:p>"#;

        let container = crate::container::OoxmlContainer::from_bytes(Vec::new());
        if container.is_err() {
            return;
        }
        let container = container.unwrap();
        let mut parser = DocxParser {
            container,
            styles: StyleMap::default(),
            numbering: NumberingMap::default(),
            relationships: crate::container::Relationships::default(),
            footnotes: HashMap::new(),
            endnotes: HashMap::new(),
        };

        let para = parser.parse_paragraph(xml).unwrap();
        let text = para.plain_text();

        assert!(
            text.contains('\t'),
            "Expected tab character between columns, got: '{}'",
            text
        );
        assert!(
            text.contains("Column1") && text.contains("Column2"),
            "Expected both column texts, got: '{}'",
            text
        );
    }

    #[test]
    fn test_multiple_spaces_preserved() {
        // Test multiple consecutive spaces are preserved
        let xml = r#"<w:p>
            <w:r><w:t xml:space="preserve">Word1     Word2</w:t></w:r>
        </w:p>"#;

        let container = crate::container::OoxmlContainer::from_bytes(Vec::new());
        if container.is_err() {
            return;
        }
        let container = container.unwrap();
        let mut parser = DocxParser {
            container,
            styles: StyleMap::default(),
            numbering: NumberingMap::default(),
            relationships: crate::container::Relationships::default(),
            footnotes: HashMap::new(),
            endnotes: HashMap::new(),
        };

        let para = parser.parse_paragraph(xml).unwrap();
        let text = para.plain_text();

        // Multiple spaces should be preserved
        assert!(
            text.contains("     "),
            "Expected 5 consecutive spaces, got: '{}'",
            text
        );
    }

    #[test]
    fn test_carriage_return_handling() {
        // Test <w:cr/> element creates line break
        let xml = r#"<w:p>
            <w:r>
                <w:t>Line1</w:t>
                <w:cr/>
                <w:t>Line2</w:t>
            </w:r>
        </w:p>"#;

        let container = crate::container::OoxmlContainer::from_bytes(Vec::new());
        if container.is_err() {
            return;
        }
        let container = container.unwrap();
        let mut parser = DocxParser {
            container,
            styles: StyleMap::default(),
            numbering: NumberingMap::default(),
            relationships: crate::container::Relationships::default(),
            footnotes: HashMap::new(),
            endnotes: HashMap::new(),
        };

        let para = parser.parse_paragraph(xml).unwrap();

        // Check that we have a line break somewhere
        let has_line_break = para.runs.iter().any(|r| r.line_break);
        assert!(has_line_break, "Expected line break from <w:cr/>");
    }

    #[test]
    fn test_non_breaking_hyphen() {
        // Test <w:noBreakHyphen/> is converted to non-breaking hyphen Unicode
        let xml = r#"<w:p>
            <w:r>
                <w:t>non</w:t>
                <w:noBreakHyphen/>
                <w:t>breaking</w:t>
            </w:r>
        </w:p>"#;

        let container = crate::container::OoxmlContainer::from_bytes(Vec::new());
        if container.is_err() {
            return;
        }
        let container = container.unwrap();
        let mut parser = DocxParser {
            container,
            styles: StyleMap::default(),
            numbering: NumberingMap::default(),
            relationships: crate::container::Relationships::default(),
            footnotes: HashMap::new(),
            endnotes: HashMap::new(),
        };

        let para = parser.parse_paragraph(xml).unwrap();
        let text = para.plain_text();

        // Should contain non-breaking hyphen (U+2011) or at least the text parts
        assert!(
            text.contains("non") && text.contains("breaking"),
            "Expected 'non' and 'breaking' text, got: '{}'",
            text
        );
        assert!(
            text.contains('\u{2011}'),
            "Expected non-breaking hyphen U+2011, got: '{}'",
            text
        );
    }

    // =========================================================================
    // Tracked Changes Tests (Revisions)
    // =========================================================================

    #[test]
    fn test_tracked_changes_insertion() {
        // Test <w:ins> element marks text as inserted
        let xml = r#"<w:p>
            <w:r><w:t>Original </w:t></w:r>
            <w:ins>
                <w:r><w:t>inserted </w:t></w:r>
            </w:ins>
            <w:r><w:t>text</w:t></w:r>
        </w:p>"#;

        let container = crate::container::OoxmlContainer::from_bytes(Vec::new());
        if container.is_err() {
            return;
        }
        let container = container.unwrap();
        let mut parser = DocxParser {
            container,
            styles: StyleMap::default(),
            numbering: NumberingMap::default(),
            relationships: crate::container::Relationships::default(),
            footnotes: HashMap::new(),
            endnotes: HashMap::new(),
        };

        let para = parser.parse_paragraph(xml).unwrap();

        // Check that we have an inserted revision
        let has_inserted = para
            .runs
            .iter()
            .any(|r| r.revision == RevisionType::Inserted);
        assert!(has_inserted, "Expected to find inserted revision");

        // The inserted text should be marked
        let inserted_text: String = para
            .runs
            .iter()
            .filter(|r| r.revision == RevisionType::Inserted)
            .map(|r| r.text.as_str())
            .collect();
        assert!(
            inserted_text.contains("inserted"),
            "Expected 'inserted' text in revision, got: '{}'",
            inserted_text
        );
    }

    #[test]
    fn test_tracked_changes_deletion() {
        // Test <w:del> element marks text as deleted
        let xml = r#"<w:p>
            <w:r><w:t>Keep this </w:t></w:r>
            <w:del>
                <w:r><w:t>deleted </w:t></w:r>
            </w:del>
            <w:r><w:t>text</w:t></w:r>
        </w:p>"#;

        let container = crate::container::OoxmlContainer::from_bytes(Vec::new());
        if container.is_err() {
            return;
        }
        let container = container.unwrap();
        let mut parser = DocxParser {
            container,
            styles: StyleMap::default(),
            numbering: NumberingMap::default(),
            relationships: crate::container::Relationships::default(),
            footnotes: HashMap::new(),
            endnotes: HashMap::new(),
        };

        let para = parser.parse_paragraph(xml).unwrap();

        // Check that we have a deleted revision
        let has_deleted = para
            .runs
            .iter()
            .any(|r| r.revision == RevisionType::Deleted);
        assert!(has_deleted, "Expected to find deleted revision");

        // The deleted text should be marked
        let deleted_text: String = para
            .runs
            .iter()
            .filter(|r| r.revision == RevisionType::Deleted)
            .map(|r| r.text.as_str())
            .collect();
        assert!(
            deleted_text.contains("deleted"),
            "Expected 'deleted' text in revision, got: '{}'",
            deleted_text
        );
    }

    // =========================================================================
    // Footnote / Endnote Tests
    // =========================================================================

    #[test]
    fn test_parse_footnotes_xml() {
        let xml = r#"<?xml version="1.0" encoding="UTF-8"?>
        <w:footnotes xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main">
            <w:footnote w:id="0" w:type="separator">
                <w:p><w:r><w:t>___</w:t></w:r></w:p>
            </w:footnote>
            <w:footnote w:id="1" w:type="continuationSeparator">
                <w:p><w:r><w:t>---</w:t></w:r></w:p>
            </w:footnote>
            <w:footnote w:id="2">
                <w:p><w:r><w:t>This is footnote two.</w:t></w:r></w:p>
            </w:footnote>
            <w:footnote w:id="3">
                <w:p><w:r><w:t>Another footnote.</w:t></w:r></w:p>
            </w:footnote>
        </w:footnotes>"#;

        let notes = parse_notes_xml(xml, b"w:footnote");

        // Separator and continuationSeparator should be skipped
        assert!(!notes.contains_key("0"), "separator should be skipped");
        assert!(
            !notes.contains_key("1"),
            "continuationSeparator should be skipped"
        );

        // Content footnotes should be parsed
        assert_eq!(notes.get("2").unwrap(), "This is footnote two.");
        assert_eq!(notes.get("3").unwrap(), "Another footnote.");
        assert_eq!(notes.len(), 2);
    }

    #[test]
    fn test_parse_endnotes_xml() {
        let xml = r#"<?xml version="1.0" encoding="UTF-8"?>
        <w:endnotes xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main">
            <w:endnote w:id="0" w:type="separator">
                <w:p><w:r><w:t>___</w:t></w:r></w:p>
            </w:endnote>
            <w:endnote w:id="1">
                <w:p><w:r><w:t>Endnote content here.</w:t></w:r></w:p>
            </w:endnote>
        </w:endnotes>"#;

        let notes = parse_notes_xml(xml, b"w:endnote");

        assert!(!notes.contains_key("0"), "separator should be skipped");
        assert_eq!(notes.get("1").unwrap(), "Endnote content here.");
        assert_eq!(notes.len(), 1);
    }

    #[test]
    fn test_footnote_multi_run_text() {
        // Footnote with multiple runs should concatenate text
        let xml = r#"<?xml version="1.0" encoding="UTF-8"?>
        <w:footnotes xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main">
            <w:footnote w:id="1">
                <w:p>
                    <w:r><w:t>First </w:t></w:r>
                    <w:r><w:t>second </w:t></w:r>
                    <w:r><w:t>third.</w:t></w:r>
                </w:p>
            </w:footnote>
        </w:footnotes>"#;

        let notes = parse_notes_xml(xml, b"w:footnote");
        assert_eq!(notes.get("1").unwrap(), "First second third.");
    }

    #[test]
    fn test_footnote_reference_in_paragraph() {
        // Test that w:footnoteReference inserts [^N] marker
        let xml = r#"<w:p>
            <w:r><w:t>Some text</w:t></w:r>
            <w:r><w:footnoteReference w:id="2"/></w:r>
            <w:r><w:t> more text</w:t></w:r>
        </w:p>"#;

        let container = crate::container::OoxmlContainer::from_bytes(Vec::new());
        if container.is_err() {
            return;
        }
        let container = container.unwrap();
        let mut footnotes = HashMap::new();
        footnotes.insert("2".to_string(), "Footnote content".to_string());

        let mut parser = DocxParser {
            container,
            styles: StyleMap::default(),
            numbering: NumberingMap::default(),
            relationships: crate::container::Relationships::default(),
            footnotes,
            endnotes: HashMap::new(),
        };

        let para = parser.parse_paragraph(xml).unwrap();
        let text = para.plain_text();

        assert!(
            text.contains("[^2]"),
            "Expected footnote reference [^2], got: '{}'",
            text
        );
        assert!(
            text.contains("Some text"),
            "Expected original text, got: '{}'",
            text
        );
    }

    #[test]
    fn test_endnote_reference_in_paragraph() {
        // Test that w:endnoteReference inserts [^eN] marker
        let xml = r#"<w:p>
            <w:r><w:t>Text with endnote</w:t></w:r>
            <w:r><w:endnoteReference w:id="1"/></w:r>
        </w:p>"#;

        let container = crate::container::OoxmlContainer::from_bytes(Vec::new());
        if container.is_err() {
            return;
        }
        let container = container.unwrap();
        let mut endnotes = HashMap::new();
        endnotes.insert("1".to_string(), "Endnote content".to_string());

        let mut parser = DocxParser {
            container,
            styles: StyleMap::default(),
            numbering: NumberingMap::default(),
            relationships: crate::container::Relationships::default(),
            footnotes: HashMap::new(),
            endnotes,
        };

        let para = parser.parse_paragraph(xml).unwrap();
        let text = para.plain_text();

        assert!(
            text.contains("[^e1]"),
            "Expected endnote reference [^e1], got: '{}'",
            text
        );
    }

    #[test]
    fn test_footnote_reference_skipped_when_no_content() {
        // If the footnote id doesn't exist in the map, no marker should be inserted
        let xml = r#"<w:p>
            <w:r><w:t>Text</w:t></w:r>
            <w:r><w:footnoteReference w:id="99"/></w:r>
        </w:p>"#;

        let container = crate::container::OoxmlContainer::from_bytes(Vec::new());
        if container.is_err() {
            return;
        }
        let container = container.unwrap();
        let mut parser = DocxParser {
            container,
            styles: StyleMap::default(),
            numbering: NumberingMap::default(),
            relationships: crate::container::Relationships::default(),
            footnotes: HashMap::new(), // No footnotes
            endnotes: HashMap::new(),
        };

        let para = parser.parse_paragraph(xml).unwrap();
        let text = para.plain_text();

        assert!(
            !text.contains("[^"),
            "Should not insert marker for unknown footnote, got: '{}'",
            text
        );
    }

    #[test]
    fn test_empty_footnote_skipped() {
        // Footnotes with only whitespace should not be included
        let xml = r#"<?xml version="1.0" encoding="UTF-8"?>
        <w:footnotes xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main">
            <w:footnote w:id="1">
                <w:p><w:r><w:t>   </w:t></w:r></w:p>
            </w:footnote>
            <w:footnote w:id="2">
                <w:p><w:r><w:t>Real content.</w:t></w:r></w:p>
            </w:footnote>
        </w:footnotes>"#;

        let notes = parse_notes_xml(xml, b"w:footnote");
        assert!(
            !notes.contains_key("1"),
            "Whitespace-only note should be skipped"
        );
        assert_eq!(notes.get("2").unwrap(), "Real content.");
    }

    #[test]
    fn test_parse_header_footer_xml_basic() {
        let xml = r#"<?xml version="1.0" encoding="UTF-8"?>
        <w:hdr xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main">
            <w:p>
                <w:r><w:t>Company Name</w:t></w:r>
            </w:p>
            <w:p>
                <w:r><w:t>Confidential</w:t></w:r>
            </w:p>
        </w:hdr>"#;

        let paragraphs = parse_header_footer_xml(xml);
        assert_eq!(paragraphs.len(), 2);
        assert_eq!(paragraphs[0].plain_text(), "Company Name");
        assert_eq!(paragraphs[1].plain_text(), "Confidential");
    }

    #[test]
    fn test_parse_header_footer_xml_empty_paragraphs_skipped() {
        let xml = r#"<?xml version="1.0" encoding="UTF-8"?>
        <w:ftr xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main">
            <w:p></w:p>
            <w:p>
                <w:r><w:t>Page 1</w:t></w:r>
            </w:p>
            <w:p>
                <w:r><w:t>   </w:t></w:r>
            </w:p>
        </w:ftr>"#;

        let paragraphs = parse_header_footer_xml(xml);
        assert_eq!(paragraphs.len(), 1);
        assert_eq!(paragraphs[0].plain_text(), "Page 1");
    }

    #[test]
    fn test_parse_header_footer_xml_multiple_runs() {
        let xml = r#"<?xml version="1.0" encoding="UTF-8"?>
        <w:hdr xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main">
            <w:p>
                <w:r><w:t>Draft - </w:t></w:r>
                <w:r><w:t>Do Not Distribute</w:t></w:r>
            </w:p>
        </w:hdr>"#;

        let paragraphs = parse_header_footer_xml(xml);
        assert_eq!(paragraphs.len(), 1);
        assert_eq!(paragraphs[0].plain_text(), "Draft - Do Not Distribute");
    }

    #[test]
    fn test_parse_header_footer_xml_empty() {
        let xml = r#"<?xml version="1.0" encoding="UTF-8"?>
        <w:hdr xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main">
        </w:hdr>"#;

        let paragraphs = parse_header_footer_xml(xml);
        assert!(paragraphs.is_empty());
    }

    // =========================================================================
    // Text Box Content Extraction Tests (w:txbxContent)
    // =========================================================================

    /// Helper to create a minimal DOCX in memory with given document.xml content.
    fn create_minimal_docx(document_xml: &str) -> Vec<u8> {
        use std::io::{Cursor, Write};
        let buf = Cursor::new(Vec::new());
        let mut zip = zip::ZipWriter::new(buf);
        let options = zip::write::SimpleFileOptions::default()
            .compression_method(zip::CompressionMethod::Stored);

        // [Content_Types].xml
        zip.start_file("[Content_Types].xml", options).unwrap();
        zip.write_all(br#"<?xml version="1.0" encoding="UTF-8"?>
<Types xmlns="http://schemas.openxmlformats.org/package/2006/content-types">
  <Default Extension="rels" ContentType="application/vnd.openxmlformats-package.relationships+xml"/>
  <Default Extension="xml" ContentType="application/xml"/>
  <Override PartName="/word/document.xml" ContentType="application/vnd.openxmlformats-officedocument.wordprocessingml.document.main+xml"/>
</Types>"#).unwrap();

        // _rels/.rels
        zip.start_file("_rels/.rels", options).unwrap();
        zip.write_all(br#"<?xml version="1.0" encoding="UTF-8"?>
<Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships">
  <Relationship Id="rId1" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/officeDocument" Target="word/document.xml"/>
</Relationships>"#).unwrap();

        // word/_rels/document.xml.rels
        zip.start_file("word/_rels/document.xml.rels", options)
            .unwrap();
        zip.write_all(
            br#"<?xml version="1.0" encoding="UTF-8"?>
<Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships">
</Relationships>"#,
        )
        .unwrap();

        // word/document.xml
        zip.start_file("word/document.xml", options).unwrap();
        zip.write_all(document_xml.as_bytes()).unwrap();

        zip.finish().unwrap().into_inner()
    }

    #[test]
    fn test_textbox_content_extracted() {
        // Text box via w:drawing > wps:txbx > w:txbxContent
        let doc_xml = r#"<?xml version="1.0" encoding="UTF-8"?>
<w:document xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main"
            xmlns:wps="http://schemas.microsoft.com/office/word/2010/wordprocessingShape">
  <w:body>
    <w:p>
      <w:r>
        <w:t>Normal paragraph</w:t>
      </w:r>
    </w:p>
    <w:p>
      <w:r>
        <w:drawing>
          <wps:wsp>
            <wps:txbx>
              <w:txbxContent>
                <w:p>
                  <w:r><w:t>Text box content here</w:t></w:r>
                </w:p>
              </w:txbxContent>
            </wps:txbx>
          </wps:wsp>
        </w:drawing>
      </w:r>
    </w:p>
  </w:body>
</w:document>"#;

        let data = create_minimal_docx(doc_xml);
        let mut parser = DocxParser::from_bytes(data).unwrap();
        let doc = parser.parse().unwrap();
        let text = doc.plain_text();

        assert!(
            text.contains("Normal paragraph"),
            "Should contain normal paragraph text"
        );
        assert!(
            text.contains("Text box content here"),
            "Should contain text box content, got: {}",
            text
        );
    }

    #[test]
    fn test_textbox_mc_alternate_content_no_duplication() {
        // mc:AlternateContent with text box in both Choice and Fallback
        // Should only extract once (from Choice branch)
        let doc_xml = r#"<?xml version="1.0" encoding="UTF-8"?>
<w:document xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main"
            xmlns:mc="http://schemas.openxmlformats.org/markup-compatibility/2006"
            xmlns:wps="http://schemas.microsoft.com/office/word/2010/wordprocessingShape"
            xmlns:v="urn:schemas-microsoft-com:vml">
  <w:body>
    <w:p>
      <w:r>
        <mc:AlternateContent>
          <mc:Choice>
            <w:drawing>
              <wps:wsp>
                <wps:txbx>
                  <w:txbxContent>
                    <w:p>
                      <w:r><w:t>Unique text box</w:t></w:r>
                    </w:p>
                  </w:txbxContent>
                </wps:txbx>
              </wps:wsp>
            </w:drawing>
          </mc:Choice>
          <mc:Fallback>
            <w:pict>
              <v:shape>
                <v:textbox>
                  <w:txbxContent>
                    <w:p>
                      <w:r><w:t>Unique text box</w:t></w:r>
                    </w:p>
                  </w:txbxContent>
                </v:textbox>
              </v:shape>
            </w:pict>
          </mc:Fallback>
        </mc:AlternateContent>
      </w:r>
    </w:p>
  </w:body>
</w:document>"#;

        let data = create_minimal_docx(doc_xml);
        let mut parser = DocxParser::from_bytes(data).unwrap();
        let doc = parser.parse().unwrap();
        let text = doc.plain_text();

        // Count occurrences - should appear exactly once
        let count = text.matches("Unique text box").count();
        assert_eq!(
            count, 1,
            "Text box content should appear exactly once, not duplicated. Full text: {}",
            text
        );
    }

    #[test]
    fn test_textbox_multiple_paragraphs() {
        // Text box with multiple paragraphs
        let doc_xml = r#"<?xml version="1.0" encoding="UTF-8"?>
<w:document xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main"
            xmlns:wps="http://schemas.microsoft.com/office/word/2010/wordprocessingShape">
  <w:body>
    <w:p>
      <w:r>
        <w:drawing>
          <wps:wsp>
            <wps:txbx>
              <w:txbxContent>
                <w:p>
                  <w:r><w:t>First text box paragraph</w:t></w:r>
                </w:p>
                <w:p>
                  <w:r><w:t>Second text box paragraph</w:t></w:r>
                </w:p>
              </w:txbxContent>
            </wps:txbx>
          </wps:wsp>
        </w:drawing>
      </w:r>
    </w:p>
  </w:body>
</w:document>"#;

        let data = create_minimal_docx(doc_xml);
        let mut parser = DocxParser::from_bytes(data).unwrap();
        let doc = parser.parse().unwrap();
        let text = doc.plain_text();

        assert!(
            text.contains("First text box paragraph"),
            "Should contain first text box paragraph"
        );
        assert!(
            text.contains("Second text box paragraph"),
            "Should contain second text box paragraph"
        );
    }
}