pinakes/crates/pinakes-core/src/metadata/document.rs

use std::path::Path;

use super::{ExtractedMetadata, MetadataExtractor};
use crate::{
  error::{PinakesError, Result},
  media_type::{BuiltinMediaType, MediaType},
};

pub struct DocumentExtractor;

impl MetadataExtractor for DocumentExtractor {
  fn extract(&self, path: &Path) -> Result<ExtractedMetadata> {
    match MediaType::from_path(path) {
      Some(MediaType::Builtin(BuiltinMediaType::Pdf)) => extract_pdf(path),
      Some(MediaType::Builtin(BuiltinMediaType::Epub)) => extract_epub(path),
      Some(MediaType::Builtin(BuiltinMediaType::Djvu)) => extract_djvu(path),
      _ => Ok(ExtractedMetadata::default()),
    }
  }

  fn supported_types(&self) -> Vec<MediaType> {
    vec![
      MediaType::Builtin(BuiltinMediaType::Pdf),
      MediaType::Builtin(BuiltinMediaType::Epub),
      MediaType::Builtin(BuiltinMediaType::Djvu),
    ]
  }
}

fn extract_pdf(path: &Path) -> Result<ExtractedMetadata> {
  let doc = lopdf::Document::load(path)
    .map_err(|e| PinakesError::MetadataExtraction(format!("PDF load: {e}")))?;

  let mut meta = ExtractedMetadata::default();
  let mut book_meta = crate::model::BookMetadata::default();

  // Find the Info dictionary via the trailer
  if let Ok(info_ref) = doc.trailer.get(b"Info") {
    let info_obj = info_ref
      .as_reference()
      .map_or(Some(info_ref), |reference| doc.get_object(reference).ok());

    if let Some(obj) = info_obj
      && let Ok(dict) = obj.as_dict()
    {
      if let Ok(title) = dict.get(b"Title") {
        meta.title = pdf_object_to_string(title);
      }
      if let Ok(author) = dict.get(b"Author") {
        let author_str = pdf_object_to_string(author);
        meta.artist.clone_from(&author_str);

        // Parse multiple authors if separated by semicolon, comma, or "and"
        if let Some(authors_str) = author_str {
          book_meta.authors = authors_str
            .split(&[';', ','][..])
            .flat_map(|part| part.split(" and "))
            .map(|name| name.trim().to_string())
            .filter(|name| !name.is_empty())
            .enumerate()
            .map(|(pos, name)| {
              let mut author = crate::model::AuthorInfo::new(name);
              author.position = i32::try_from(pos).unwrap_or(i32::MAX);
              author
            })
            .collect();
        }
      }
      if let Ok(subject) = dict.get(b"Subject") {
        meta.description = pdf_object_to_string(subject);
      }
      if let Ok(creator) = dict.get(b"Creator") {
        meta.extra.insert(
          "creator".to_string(),
          pdf_object_to_string(creator).unwrap_or_default(),
        );
      }
      if let Ok(producer) = dict.get(b"Producer") {
        meta.extra.insert(
          "producer".to_string(),
          pdf_object_to_string(producer).unwrap_or_default(),
        );
      }
    }
  }

  // Page count
  let pages = doc.get_pages();
  let page_count = pages.len();
  if page_count > 0 {
    book_meta.page_count = Some(i32::try_from(page_count).unwrap_or(i32::MAX));
  }

  // Try to extract ISBN from first few pages
  // Extract text from up to the first 5 pages and search for ISBN patterns
  let mut extracted_text = String::new();
  let max_pages = page_count.min(5);

  for (_page_num, page_id) in pages.iter().take(max_pages) {
    if let Ok(content) = doc.get_page_content(*page_id) {
      // PDF content streams contain raw operators, but may have text strings
      if let Ok(text) = std::str::from_utf8(&content) {
        extracted_text.push_str(text);
        extracted_text.push(' ');
      }
    }
  }

  // Extract ISBN from the text
  if let Some(isbn) = crate::books::extract_isbn_from_text(&extracted_text)
    && let Ok(normalized) = crate::books::normalize_isbn(&isbn)
  {
    book_meta.isbn13 = Some(normalized);
    book_meta.isbn = Some(isbn);
  }

  // Set format
  book_meta.format = Some("pdf".to_string());

  meta.book_metadata = Some(book_meta);
  Ok(meta)
}

fn pdf_object_to_string(obj: &lopdf::Object) -> Option<String> {
  match obj {
    lopdf::Object::String(bytes, _) => {
      Some(String::from_utf8_lossy(bytes).into_owned())
    },
    lopdf::Object::Name(name) => {
      Some(String::from_utf8_lossy(name).into_owned())
    },
    _ => None,
  }
}

fn extract_epub(path: &Path) -> Result<ExtractedMetadata> {
  let mut doc = epub::doc::EpubDoc::new(path).map_err(|e| {
    PinakesError::MetadataExtraction(format!("EPUB parse: {e}"))
  })?;

  let mut meta = ExtractedMetadata {
    title: doc.mdata("title").map(|item| item.value.clone()),
    artist: doc.mdata("creator").map(|item| item.value.clone()),
    description: doc.mdata("description").map(|item| item.value.clone()),
    ..Default::default()
  };

  let mut book_meta = crate::model::BookMetadata::default();

  // Extract basic metadata
  if let Some(lang) = doc.mdata("language") {
    book_meta.language = Some(lang.value.clone());
  }
  if let Some(publisher) = doc.mdata("publisher") {
    book_meta.publisher = Some(publisher.value.clone());
  }
  if let Some(date) = doc.mdata("date") {
    // Try to parse as YYYY-MM-DD or just YYYY
    if let Ok(parsed_date) =
      chrono::NaiveDate::parse_from_str(&date.value, "%Y-%m-%d")
    {
      book_meta.publication_date = Some(parsed_date);
    } else if let Ok(year) = date.value.parse::<i32>() {
      book_meta.publication_date = chrono::NaiveDate::from_ymd_opt(year, 1, 1);
    }
  }

  // Extract authors - iterate through all metadata items
  let mut authors = Vec::new();
  let mut position = 0;
  for item in &doc.metadata {
    if item.property == "creator" || item.property == "dc:creator" {
      let mut author = crate::model::AuthorInfo::new(item.value.clone());
      author.position = position;
      position += 1;

      // Check for file-as in refinements
      if let Some(file_as_ref) = item.refinement("file-as") {
        author.file_as = Some(file_as_ref.value.clone());
      }

      // Check for role in refinements
      if let Some(role_ref) = item.refinement("role") {
        author.role.clone_from(&role_ref.value);
      }

      authors.push(author);
    }
  }
  book_meta.authors = authors;

  // Extract ISBNs from identifiers
  let mut identifiers = std::collections::HashMap::new();
  for item in &doc.metadata {
    if item.property == "identifier" || item.property == "dc:identifier" {
      // Try to get scheme from refinements
      let scheme = item
        .refinement("identifier-type")
        .map(|r| r.value.to_lowercase());

      let id_type = match scheme.as_deref() {
        Some("isbn" | "isbn-10" | "isbn10") => "isbn",
        Some("isbn-13" | "isbn13") => "isbn13",
        Some("asin") => "asin",
        Some("doi") => "doi",
        _ => {
          // Fallback: detect from value pattern.
          // ISBN-10 = 10 chars bare, ISBN-13 = 13 chars bare,
          // hyphenated ISBN-13 = 17 chars (e.g. 978-0-123-45678-9).
          // Parentheses required: && binds tighter than ||.
          if (item.value.len() == 10 || item.value.len() == 13)
            || (item.value.contains('-')
              && (item.value.len() == 13 || item.value.len() == 17))
          {
            "isbn"
          } else {
            "other"
          }
        },
      };

      // Try to normalize ISBN
      if (id_type == "isbn" || id_type == "isbn13")
        && let Ok(normalized) = crate::books::normalize_isbn(&item.value)
      {
        book_meta.isbn13 = Some(normalized.clone());
        book_meta.isbn = Some(item.value.clone());
      }

      identifiers
        .entry(id_type.to_string())
        .or_insert_with(Vec::new)
        .push(item.value.clone());
    }
  }
  book_meta.identifiers = identifiers;

  // Extract Calibre series metadata by parsing the content.opf file
  // Try common OPF locations
  let opf_paths = vec!["OEBPS/content.opf", "content.opf", "OPS/content.opf"];
  let mut opf_data = None;
  for path in opf_paths {
    if let Some(data) = doc.get_resource_str_by_path(path) {
      opf_data = Some(data);
      break;
    }
  }

  if let Some(opf_content) = opf_data {
    // Look for <meta name="calibre:series" content="Series Name"/>
    if let Some(series_start) = opf_content.find("name=\"calibre:series\"")
      && let Some(content_start) =
        opf_content[series_start..].find("content=\"")
    {
      let after_content = &opf_content[series_start + content_start + 9..];
      if let Some(quote_end) = after_content.find('"') {
        book_meta.series_name = Some(after_content[..quote_end].to_string());
      }
    }

    // Look for <meta name="calibre:series_index" content="1.0"/>
    if let Some(index_start) = opf_content.find("name=\"calibre:series_index\"")
      && let Some(content_start) = opf_content[index_start..].find("content=\"")
    {
      let after_content = &opf_content[index_start + content_start + 9..];
      if let Some(quote_end) = after_content.find('"')
        && let Ok(index) = after_content[..quote_end].parse::<f64>()
      {
        book_meta.series_index = Some(index);
      }
    }
  }

  // Set format
  book_meta.format = Some("epub".to_string());

  meta.book_metadata = Some(book_meta);
  Ok(meta)
}

fn extract_djvu(path: &Path) -> Result<ExtractedMetadata> {
  // DjVu files contain metadata in SEXPR (S-expression) format within
  // ANTa/ANTz chunks, or in the DIRM chunk. We parse the raw bytes to
  // extract any metadata fields we can find.

  // Guard against loading very large DjVu files into memory.
  const MAX_DJVU_SIZE: u64 = 50 * 1024 * 1024; // 50 MB
  let file_meta = std::fs::metadata(path)
    .map_err(|e| PinakesError::MetadataExtraction(format!("DjVu stat: {e}")))?;
  if file_meta.len() > MAX_DJVU_SIZE {
    return Ok(ExtractedMetadata::default());
  }

  let data = std::fs::read(path)
    .map_err(|e| PinakesError::MetadataExtraction(format!("DjVu read: {e}")))?;

  let mut meta = ExtractedMetadata::default();

  // DjVu files start with "AT&T" magic followed by FORM:DJVU or FORM:DJVM
  if data.len() < 16 {
    return Ok(meta);
  }

  // Search for metadata annotations in the file. DjVu metadata is stored
  // as S-expressions like (metadata (key "value") ...) within ANTa chunks.
  let content = String::from_utf8_lossy(&data);

  // Look for (metadata ...) blocks
  if let Some(meta_start) = content.find("(metadata") {
    let remainder = &content[meta_start..];
    // Extract key-value pairs like (title "Some Title")
    extract_djvu_field(remainder, "title", &mut meta.title);
    extract_djvu_field(remainder, "author", &mut meta.artist);

    let mut desc = None;
    extract_djvu_field(remainder, "subject", &mut desc);
    if desc.is_none() {
      extract_djvu_field(remainder, "description", &mut desc);
    }
    meta.description = desc;

    let mut year_str = None;
    extract_djvu_field(remainder, "year", &mut year_str);
    if let Some(ref y) = year_str {
      meta.year = y.parse().ok();
    }

    let mut creator = None;
    extract_djvu_field(remainder, "creator", &mut creator);
    if let Some(c) = creator {
      meta.extra.insert("creator".to_string(), c);
    }
  }

  // Also check for booklet-style metadata that some DjVu encoders write
  // outside the metadata SEXPR
  if meta.title.is_none()
    && let Some(title_start) = content.find("(bookmarks")
  {
    let remainder = &content[title_start..];
    // First bookmark title is often the document title
    if let Some(q1) = remainder.find('"') {
      let after_q1 = &remainder[q1 + 1..];
      if let Some(q2) = after_q1.find('"') {
        let val = &after_q1[..q2];
        if !val.is_empty() {
          meta.title = Some(val.to_string());
        }
      }
    }
  }

  Ok(meta)
}

fn extract_djvu_field(sexpr: &str, key: &str, out: &mut Option<String>) {
  // Look for patterns like (key "value") in the S-expression
  let pattern = format!("({key}");
  if let Some(start) = sexpr.find(&pattern) {
    let remainder = &sexpr[start + pattern.len()..];
    // Find the quoted value
    if let Some(q1) = remainder.find('"') {
      let after_q1 = &remainder[q1 + 1..];
      if let Some(q2) = after_q1.find('"') {
        let val = &after_q1[..q2];
        if !val.is_empty() {
          *out = Some(val.to_string());
        }
      }
    }
  }
}