openchamber/packages/ui/src/sync/document-attachments.ts

import { unzip, unzipSync, type UnzipFileInfo, type Unzipped } from "fflate"

const MAX_ARCHIVE_BYTES = 20 * 1024 * 1024
const MAX_UNCOMPRESSED_BYTES = 100 * 1024 * 1024
const MAX_ENTRY_BYTES = 25 * 1024 * 1024
const MAX_XML_ENTRY_BYTES = 8 * 1024 * 1024
const MAX_ARCHIVE_ENTRIES = 5_000
const MAX_EMBEDDED_IMAGES = 50
const MAX_EMBEDDED_IMAGE_BYTES = 20 * 1024 * 1024
const MAX_EMBEDDED_IMAGES_BYTES = 40 * 1024 * 1024
const MAX_EXTRACTED_TEXT_CHARS = 2_000_000
const MAX_ODF_SPACES_PER_ELEMENT = 100
const TEXT_TRUNCATION_NOTICE = "\n\n[Document text truncated by OpenChamber]\n"

const OFFICE_EXTENSIONS = new Set(["docx", "pptx", "xlsx", "odt", "odp", "ods"])
const IMAGE_MIMES = new Map([
  ["png", "image/png"],
  ["jpg", "image/jpeg"],
  ["jpeg", "image/jpeg"],
  ["gif", "image/gif"],
  ["webp", "image/webp"],
])

type Relationship = { target: string; type: string }
type Relationships = Map<string, Relationship>

type ExtractedDocumentAttachments = {
  textFile: File
  images: File[]
}

const extensionOf = (name: string): string => {
  const index = name.lastIndexOf(".")
  return index === -1 ? "" : name.slice(index + 1).toLowerCase()
}

const basenameWithoutExtension = (name: string): string => {
  const basename = name.replace(/\\/g, "/").split("/").pop() || "document"
  const index = basename.lastIndexOf(".")
  return (index > 0 ? basename.slice(0, index) : basename).replace(/[^a-zA-Z0-9._-]+/g, "-") || "document"
}

const normalizeArchivePath = (path: string): string | undefined => {
  const segments: string[] = []
  for (const segment of path.replace(/\\/g, "/").split("/")) {
    if (!segment || segment === ".") continue
    if (segment === "..") {
      if (segments.length === 0) return
      segments.pop()
      continue
    }
    segments.push(segment)
  }
  return segments.join("/")
}

const resolveArchivePath = (sourcePath: string, target: string): string | undefined => {
  if (target.startsWith("/")) return normalizeArchivePath(target.slice(1))
  const sourceDirectory = sourcePath.includes("/") ? sourcePath.slice(0, sourcePath.lastIndexOf("/") + 1) : ""
  return normalizeArchivePath(`${sourceDirectory}${target}`)
}

const relationshipsPath = (sourcePath: string): string => {
  const index = sourcePath.lastIndexOf("/")
  const directory = index === -1 ? "" : sourcePath.slice(0, index + 1)
  const filename = sourcePath.slice(index + 1)
  return `${directory}_rels/${filename}.rels`
}

const decodeXmlCodePoint = (code: string, radix: number): string => {
  const value = Number.parseInt(code, radix)
  if (value < 0 || value > 0x10FFFF || (value >= 0xD800 && value <= 0xDFFF)) return "�"
  return String.fromCodePoint(value)
}

const decodeXml = (value: string): string => value
  .replace(/&#x([0-9a-f]+);/gi, (_, code: string) => decodeXmlCodePoint(code, 16))
  .replace(/&#([0-9]+);/g, (_, code: string) => decodeXmlCodePoint(code, 10))
  .replace(/&lt;/g, "<")
  .replace(/&gt;/g, ">")
  .replace(/&quot;/g, '"')
  .replace(/&apos;/g, "'")
  .replace(/&amp;/g, "&")

const attribute = (tag: string, name: string): string | undefined => {
  const escaped = name.replace(/[.*+?^${}()|[\]\\]/g, "\\$&")
  const match = tag.match(new RegExp(`(?:^|\\s)${escaped}\\s*=\\s*(?:"([^"]*)"|'([^']*)')`, "i"))
  return decodeXml(match?.[1] ?? match?.[2] ?? "") || undefined
}

const tagBlocks = (xml: string, tag: string): string[] => {
  const escaped = tag.replace(/[.*+?^${}()|[\]\\]/g, "\\$&")
  return Array.from(xml.matchAll(new RegExp(`<${escaped}\\b[^>]*>[\\s\\S]*?<\\/${escaped}>`, "gi")), (match) => match[0])
}

const textDecoder = new TextDecoder()
const xml = (archive: Unzipped, path: string): string => {
  const bytes = archive[path]
  return bytes ? textDecoder.decode(bytes) : ""
}

const parseRelationships = (archive: Unzipped, sourcePath: string): Relationships => {
  const result: Relationships = new Map()
  const source = xml(archive, relationshipsPath(sourcePath))
  for (const match of source.matchAll(/<Relationship\b[^>]*\/?\s*>/gi)) {
    const id = attribute(match[0], "Id")
    const target = attribute(match[0], "Target")
    if (!id || !target) continue
    result.set(id, { target, type: attribute(match[0], "Type") ?? "" })
  }
  return result
}

const isControlCharacter = (character: string): boolean => {
  const code = character.charCodeAt(0)
  return code <= 0x1F || code === 0x7F
}

const hasControlCharacters = (value: string): boolean => Array.from(value).some(isControlCharacter)

const embeddedImageLabel = (path: string): string => {
  const basename = path.replace(/\\/g, "/").split("/").pop() || "embedded image"
  return Array.from(basename.slice(0, 200), (character) => {
    if (character === "[" || character === "]") return "_"
    return isControlCharacter(character) ? "_" : character
  }).join("")
}

const hasBytes = (bytes: Uint8Array, expected: number[]): boolean => expected.every((value, index) => bytes[index] === value)

const hasAscii = (bytes: Uint8Array, offset: number, expected: string): boolean => {
  if (bytes.byteLength < offset + expected.length) return false
  for (let index = 0; index < expected.length; index += 1) {
    if (bytes[offset + index] !== expected.charCodeAt(index)) return false
  }
  return true
}

const hasValidImageSignature = (bytes: Uint8Array, extension: string): boolean => {
  switch (extension) {
    case "png":
      return hasBytes(bytes, [0x89, 0x50, 0x4E, 0x47, 0x0D, 0x0A, 0x1A, 0x0A])
    case "jpg":
    case "jpeg":
      return hasBytes(bytes, [0xFF, 0xD8, 0xFF])
    case "gif":
      return hasAscii(bytes, 0, "GIF87a") || hasAscii(bytes, 0, "GIF89a")
    case "webp":
      return hasAscii(bytes, 0, "RIFF") && hasAscii(bytes, 8, "WEBP")
    default:
      return false
  }
}

class EmbeddedImages {
  private readonly files: File[] = []
  private readonly filenames = new Map<string, string>()
  private count = 0
  private imageBytes = 0
  private readonly reservedFilenames: Set<string>

  constructor(
    private readonly archive: Unzipped,
    private readonly documentName: string,
    reservedFilenames: Iterable<string>,
  ) {
    this.reservedFilenames = new Set(Array.from(reservedFilenames, (filename) => filename.toLowerCase()))
  }

  citation(path: string | undefined): string {
    if (!path) return "[Embedded image reference could not be resolved]"
    const normalized = normalizeArchivePath(path)
    if (!normalized) return "[Unsafe embedded image path omitted]"
    const existing = this.filenames.get(normalized)
    if (existing) return `[${existing}]`

    const bytes = this.archive[normalized]
    const extension = extensionOf(normalized)
    const mime = IMAGE_MIMES.get(extension)
    const label = embeddedImageLabel(normalized)
    if (!bytes || !mime) return `[Unsupported embedded image omitted: ${label}]`
    if (!hasValidImageSignature(bytes, extension)) return `[Invalid embedded image omitted: ${label}]`
    if (
      this.files.length >= MAX_EMBEDDED_IMAGES
      || bytes.byteLength > MAX_EMBEDDED_IMAGE_BYTES
      || this.imageBytes + bytes.byteLength > MAX_EMBEDDED_IMAGES_BYTES
    ) {
      return `[Embedded image omitted by attachment limits: ${label}]`
    }

    const outputExtension = extension === "jpeg" ? "jpg" : extension
    let filename: string
    do {
      this.count += 1
      filename = `${basenameWithoutExtension(this.documentName)}-image-${this.count}.${outputExtension}`
    } while (this.reservedFilenames.has(filename.toLowerCase()))
    this.reservedFilenames.add(filename.toLowerCase())
    this.filenames.set(normalized, filename)
    this.files.push(new File([bytes], filename, { type: mime }))
    this.imageBytes += bytes.byteLength
    return `[${filename}]`
  }

  all(): File[] {
    return this.files
  }
}

const relationshipTarget = (sourcePath: string, relationships: Relationships, id: string | undefined): string | undefined => {
  if (!id) return
  const relationship = relationships.get(id)
  return relationship ? resolveArchivePath(sourcePath, relationship.target) : undefined
}

const inlineText = (
  block: string,
  sourcePath: string,
  relationships: Relationships,
  images: EmbeddedImages,
): string => {
  const pieces: string[] = []
  const tokenPattern = /<(?:w:t|a:t|text:span)\b[^>]*>([\s\S]*?)<\/(?:w:t|a:t|text:span)>|<(?:w:tab|text:tab)\b[^>]*\/?>|<(?:w:br|a:br|text:line-break)\b[^>]*\/?>|<(?:a:blip|v:imagedata)\b[^>]*>|<draw:image\b[^>]*>/gi
  for (const match of block.matchAll(tokenPattern)) {
    if (match[1] !== undefined) {
      pieces.push(decodeXml(match[1]).replace(/<[^>]+>/g, ""))
      continue
    }
    if (/tab/i.test(match[0])) {
      pieces.push("\t")
      continue
    }
    if (/br|line-break/i.test(match[0])) {
      pieces.push("\n")
      continue
    }
    const relationshipId = attribute(match[0], "r:embed") ?? attribute(match[0], "r:id")
    const directPath = attribute(match[0], "xlink:href")
    const target = directPath
      ? resolveArchivePath(sourcePath, directPath)
      : relationshipTarget(sourcePath, relationships, relationshipId)
    pieces.push(`\n${images.citation(target)}\n`)
  }
  return pieces.join("").replace(/[ \t]+\n/g, "\n").trim()
}

const paragraphs = (
  source: string,
  paragraphTag: string,
  sourcePath: string,
  relationships: Relationships,
  images: EmbeddedImages,
): string[] => tagBlocks(source, paragraphTag)
  .map((block) => inlineText(block, sourcePath, relationships, images))
  .filter(Boolean)

const extractDocx = (archive: Unzipped, images: EmbeddedImages): string | undefined => {
  const documentPath = "word/document.xml"
  const documentXml = xml(archive, documentPath)
  if (!documentXml) return
  const sections = ["# Document", ...paragraphs(documentXml, "w:p", documentPath, parseRelationships(archive, documentPath), images)]

  const extras = Object.keys(archive)
    .filter((path) => /^word\/(?:header|footer)\d+\.xml$/i.test(path))
    .sort()
  for (const path of extras) {
    const content = paragraphs(xml(archive, path), "w:p", path, parseRelationships(archive, path), images)
    if (content.length > 0) sections.push(`## ${path.includes("header") ? "Header" : "Footer"}`, ...content)
  }
  return `${sections.join("\n\n")}\n`
}

const numberedPaths = (archive: Unzipped, pattern: RegExp): string[] => Object.keys(archive)
  .filter((path) => pattern.test(path))
  .sort((left, right) => {
    const leftNumber = Number(left.match(/(\d+)(?=\.xml$)/)?.[1] ?? 0)
    const rightNumber = Number(right.match(/(\d+)(?=\.xml$)/)?.[1] ?? 0)
    return leftNumber - rightNumber
  })

const extractPptx = (archive: Unzipped, images: EmbeddedImages): string | undefined => {
  const slidePaths = numberedPaths(archive, /^ppt\/slides\/slide\d+\.xml$/i)
  if (slidePaths.length === 0) return
  const sections: string[] = ["# Presentation"]

  slidePaths.forEach((slidePath, index) => {
    const relationships = parseRelationships(archive, slidePath)
    const content = Array.from(
      xml(archive, slidePath).matchAll(/<a:p\b[^>]*>[\s\S]*?<\/a:p>|<p:pic\b[^>]*>[\s\S]*?<\/p:pic>/gi),
      (match) => inlineText(match[0], slidePath, relationships, images),
    ).filter(Boolean)
    sections.push(`## Slide ${index + 1}`, ...(content.length > 0 ? content : ["[Empty slide]"]))

    const notesRelationship = Array.from(relationships.values()).find((relationship) => relationship.type.endsWith("/notesSlide"))
    const notesPath = notesRelationship ? resolveArchivePath(slidePath, notesRelationship.target) : undefined
    if (!notesPath) return
    const notes = paragraphs(xml(archive, notesPath), "a:p", notesPath, parseRelationships(archive, notesPath), images)
    if (notes.length > 0) sections.push(`### Slide ${index + 1} notes`, ...notes)
  })
  return `${sections.join("\n\n")}\n`
}

const columnName = (index: number): string => {
  let value = index + 1
  let result = ""
  while (value > 0) {
    value -= 1
    result = String.fromCharCode(65 + (value % 26)) + result
    value = Math.floor(value / 26)
  }
  return result
}

const cellValue = (cell: string, sharedStrings: string[]): string => {
  const type = attribute(cell.match(/^<c\b[^>]*>/i)?.[0] ?? "", "t")
  if (type === "inlineStr") {
    return Array.from(cell.matchAll(/<t\b[^>]*>([\s\S]*?)<\/t>/gi), (match) => decodeXml(match[1]).replace(/<[^>]+>/g, "")).join("")
  }
  const value = cell.match(/<v\b[^>]*>([\s\S]*?)<\/v>/i)?.[1] ?? ""
  if (type === "s") return sharedStrings[Number(value)] ?? ""
  if (type === "b") return value === "1" ? "TRUE" : "FALSE"
  return decodeXml(value)
}

const drawingCitations = (
  archive: Unzipped,
  worksheetPath: string,
  images: EmbeddedImages,
): string[] => {
  const worksheetXml = xml(archive, worksheetPath)
  const worksheetRelationships = parseRelationships(archive, worksheetPath)
  const output: string[] = []
  for (const drawing of worksheetXml.matchAll(/<drawing\b[^>]*r:id=(?:"([^"]+)"|'([^']+)')[^>]*\/?\s*>/gi)) {
    const drawingPath = relationshipTarget(worksheetPath, worksheetRelationships, drawing[1] ?? drawing[2])
    if (!drawingPath) continue
    const drawingXml = xml(archive, drawingPath)
    const drawingRelationships = parseRelationships(archive, drawingPath)
    for (const anchor of drawingXml.matchAll(/<xdr:(?:oneCellAnchor|twoCellAnchor)\b[^>]*>([\s\S]*?)<\/xdr:(?:oneCellAnchor|twoCellAnchor)>/gi)) {
      const content = anchor[1]
      const column = Number(content.match(/<xdr:col>(\d+)<\/xdr:col>/i)?.[1] ?? 0)
      const row = Number(content.match(/<xdr:row>(\d+)<\/xdr:row>/i)?.[1] ?? 0)
      const imageId = content.match(/<a:blip\b[^>]*r:embed=(?:"([^"]+)"|'([^']+)')[^>]*>/i)
      const target = relationshipTarget(drawingPath, drawingRelationships, imageId?.[1] ?? imageId?.[2])
      output.push(`Image at ${columnName(column)}${row + 1}: ${images.citation(target)}`)
    }
  }
  return output
}

const extractXlsx = (archive: Unzipped, images: EmbeddedImages): string | undefined => {
  const workbookPath = "xl/workbook.xml"
  const workbookXml = xml(archive, workbookPath)
  if (!workbookXml) return
  const workbookRelationships = parseRelationships(archive, workbookPath)
  const sharedStrings = tagBlocks(xml(archive, "xl/sharedStrings.xml"), "si")
    .map((item) => Array.from(item.matchAll(/<t\b[^>]*>([\s\S]*?)<\/t>/gi), (match) => decodeXml(match[1]).replace(/<[^>]+>/g, "")).join(""))
  const sections: string[] = ["# Workbook"]

  for (const sheet of workbookXml.matchAll(/<sheet\b[^>]*\/?\s*>/gi)) {
    const name = attribute(sheet[0], "name") ?? "Sheet"
    const relationshipId = attribute(sheet[0], "r:id")
    const worksheetPath = relationshipTarget(workbookPath, workbookRelationships, relationshipId)
    if (!worksheetPath) continue
    sections.push(`## Sheet: ${name}`)

    const rows: string[] = []
    for (const row of tagBlocks(xml(archive, worksheetPath), "row")) {
      const cells = Array.from(row.matchAll(/<c\b[^>]*>[\s\S]*?<\/c>/gi), (match) => {
        const tag = match[0].match(/^<c\b[^>]*>/i)?.[0] ?? ""
        const reference = attribute(tag, "r") ?? "?"
        return `${reference}: ${cellValue(match[0], sharedStrings)}`
      }).filter((value) => !value.endsWith(": "))
      if (cells.length > 0) rows.push(cells.join(" | "))
    }
    sections.push(...(rows.length > 0 ? rows : ["[Empty sheet]"]), ...drawingCitations(archive, worksheetPath, images))
  }
  return `${sections.join("\n\n")}\n`
}

const expandOdfSpaces = (tag: string): string => {
  const rawCount = attribute(tag, "text:c")
  if (!rawCount) return " "

  const count = Number(rawCount)
  if (Number.isSafeInteger(count) && count > 0 && count <= MAX_ODF_SPACES_PER_ELEMENT) return " ".repeat(count)

  const omitted = Number.isSafeInteger(count) && count > MAX_ODF_SPACES_PER_ELEMENT
    ? `${count - MAX_ODF_SPACES_PER_ELEMENT} additional spaces omitted`
    : "Additional spaces omitted"
  return `${" ".repeat(MAX_ODF_SPACES_PER_ELEMENT)}[${omitted}]`
}

const odfInlineText = (source: string, sourcePath: string, images: EmbeddedImages): string => source
    .replace(/<draw:image\b[^>]*>/gi, (tag) => {
      const target = resolveArchivePath(sourcePath, attribute(tag, "xlink:href") ?? "")
      return `\n${images.citation(target)}\n`
    })
    .replace(/<text:tab\b[^>]*\/?\s*>/gi, "\t")
    .replace(/<text:line-break\b[^>]*\/?\s*>/gi, "\n")
    .replace(/<text:s\b[^>]*\/?\s*>/gi, expandOdfSpaces)
    .replace(/<[^>]+>/g, "")

const odfContent = (source: string, sourcePath: string, images: EmbeddedImages): string[] => {
  const output: string[] = []
  const contentPattern = /<text:(p|h)\b[^>]*>[\s\S]*?<\/text:\1>|<draw:image\b[^>]*>/gi
  for (const match of source.matchAll(contentPattern)) {
    const content = /^<draw:image\b/i.test(match[0])
      ? images.citation(resolveArchivePath(sourcePath, attribute(match[0], "xlink:href") ?? ""))
      : decodeXml(odfInlineText(match[0], sourcePath, images)).replace(/[ \t]+\n/g, "\n").trim()
    if (content) output.push(content)
  }
  return output
}

const extractOdt = (archive: Unzipped, images: EmbeddedImages): string | undefined => {
  const contentPath = "content.xml"
  const content = xml(archive, contentPath)
  if (!content) return
  return `${["# Document", ...odfContent(content, contentPath, images)].join("\n\n")}\n`
}

const extractOdp = (archive: Unzipped, images: EmbeddedImages): string | undefined => {
  const contentPath = "content.xml"
  const content = xml(archive, contentPath)
  if (!content) return
  const sections = ["# Presentation"]
  const pages = tagBlocks(content, "draw:page")
  pages.forEach((page, index) => {
    const openTag = page.match(/^<draw:page\b[^>]*>/i)?.[0] ?? ""
    const name = attribute(openTag, "draw:name") ?? String(index + 1)
    sections.push(`## Slide: ${name}`, ...odfContent(page, contentPath, images))
  })
  return `${sections.join("\n\n")}\n`
}

const extractOds = (archive: Unzipped, images: EmbeddedImages): string | undefined => {
  const contentPath = "content.xml"
  const content = xml(archive, contentPath)
  if (!content) return
  const sections = ["# Workbook"]
  for (const table of tagBlocks(content, "table:table")) {
    const openTag = table.match(/^<table:table\b[^>]*>/i)?.[0] ?? ""
    sections.push(`## Sheet: ${attribute(openTag, "table:name") ?? "Sheet"}`)
    for (const shapes of tagBlocks(table, "table:shapes")) {
      sections.push(...odfContent(shapes, contentPath, images))
    }
    for (const row of tagBlocks(table, "table:table-row")) {
      const cells = tagBlocks(row, "table:table-cell")
        .map((cell) => odfContent(cell, contentPath, images).join(" "))
      if (cells.some(Boolean)) sections.push(cells.join(" | "))
    }
  }
  return `${sections.join("\n\n")}\n`
}

const createArchiveEntryValidator = () => {
  let entries = 0
  let uncompressedBytes = 0

  return (info: UnzipFileInfo): void => {
    entries += 1
    uncompressedBytes += info.originalSize
    if (entries > MAX_ARCHIVE_ENTRIES) throw new Error("Document contains too many files")
    if (info.originalSize > MAX_ENTRY_BYTES) throw new Error("Document contains an oversized file")
    if (/\.(?:xml|rels)$/i.test(info.name) && info.originalSize > MAX_XML_ENTRY_BYTES) {
      throw new Error("Document contains XML that is too large to process safely")
    }
    if (uncompressedBytes > MAX_UNCOMPRESSED_BYTES) throw new Error("Document expands beyond the 100 MB safety limit")
    const normalized = normalizeArchivePath(info.name)
    const canonicalName = info.name.endsWith("/") ? info.name.slice(0, -1) : info.name
    if (
      !normalized
      || normalized !== canonicalName
      || info.name.includes("\\")
      || /^[a-z]:/i.test(info.name)
      || hasControlCharacters(info.name)
    ) {
      throw new Error("Document contains an unsafe file path")
    }
  }
}

const shouldExtractArchiveEntry = (info: UnzipFileInfo): boolean => {
  const extension = extensionOf(info.name)
  return extension === "xml" || extension === "rels" || IMAGE_MIMES.has(extension)
}

const validateArchiveMetadata = async (data: Uint8Array): Promise<void> => {
  const validateArchiveEntry = createArchiveEntryValidator()
  if ("Bun" in globalThis) {
    unzipSync(data, { filter: (info) => {
      validateArchiveEntry(info)
      return false
    } })
    return
  }

  await new Promise<void>((resolve, reject) => {
    unzip(data, { filter: (info) => {
      validateArchiveEntry(info)
      return false
    } }, (error) => {
      if (error) reject(error)
      else resolve()
    })
  })
}

const validateExtractedArchive = (archive: Unzipped): void => {
  let uncompressedBytes = 0
  for (const [path, bytes] of Object.entries(archive)) {
    uncompressedBytes += bytes.byteLength
    if (bytes.byteLength > MAX_ENTRY_BYTES) throw new Error("Document contains an oversized file")
    if (/\.(?:xml|rels)$/i.test(path) && bytes.byteLength > MAX_XML_ENTRY_BYTES) {
      throw new Error("Document contains XML that is too large to process safely")
    }
    if (uncompressedBytes > MAX_UNCOMPRESSED_BYTES) throw new Error("Document expands beyond the 100 MB safety limit")
  }
}

const unzipDocument = async (file: File): Promise<Unzipped> => {
  if (file.size > MAX_ARCHIVE_BYTES) throw new Error("Document exceeds the 20 MB attachment limit")
  const data = new Uint8Array(await file.arrayBuffer())
  await validateArchiveMetadata(data)

  // Bun's browser-style Blob workers do not reliably execute fflate's async decoder.
  // Production browser runtimes use the worker-backed path below.
  const archive = "Bun" in globalThis
    ? unzipSync(data, { filter: shouldExtractArchiveEntry })
    : await new Promise<Unzipped>((resolve, reject) => {
      unzip(data, { filter: shouldExtractArchiveEntry }, (error, result) => {
        if (error) reject(error)
        else resolve(result)
      })
    })
  validateExtractedArchive(archive)
  return archive
}

const extractDocumentText = (extension: string, archive: Unzipped, images: EmbeddedImages): string | undefined => {
  switch (extension) {
    case "docx":
      return extractDocx(archive, images)
    case "pptx":
      return extractPptx(archive, images)
    case "xlsx":
      return extractXlsx(archive, images)
    case "odt":
      return extractOdt(archive, images)
    case "odp":
      return extractOdp(archive, images)
    case "ods":
      return extractOds(archive, images)
    default:
      return undefined
  }
}

const boundExtractedText = (text: string, imageFilenames: Set<string>): string => {
  if (text.length <= MAX_EXTRACTED_TEXT_CHARS) return text

  let end = MAX_EXTRACTED_TEXT_CHARS - TEXT_TRUNCATION_NOTICE.length
  const lastOpenBracket = text.lastIndexOf("[", end - 1)
  const lastCloseBracket = text.lastIndexOf("]", end - 1)
  if (lastOpenBracket > lastCloseBracket) {
    const nextCloseBracket = text.indexOf("]", lastOpenBracket)
    const candidate = nextCloseBracket === -1 ? "" : text.slice(lastOpenBracket + 1, nextCloseBracket)
    if (imageFilenames.has(candidate)) end = lastOpenBracket
  }
  return `${text.slice(0, end)}${TEXT_TRUNCATION_NOTICE}`
}

const citedImageFilenames = (text: string): Set<string> => {
  const filenames = new Set<string>()
  for (const match of text.matchAll(/\[([^\]\r\n]+)\]/g)) filenames.add(match[1])
  return filenames
}

export const extractDocumentAttachments = async (
  file: File,
  reservedFilenames: Iterable<string> = [],
): Promise<ExtractedDocumentAttachments | undefined> => {
  const extension = extensionOf(file.name)
  if (!OFFICE_EXTENSIONS.has(extension)) return
  const archive = await unzipDocument(file)
  const images = new EmbeddedImages(archive, file.name, reservedFilenames)
  const text = extractDocumentText(extension, archive, images)
  if (!text) return
  const extractedImages = images.all()
  const imageFilenames = new Set(extractedImages.map((image) => image.name))
  const boundedText = boundExtractedText(text, imageFilenames)
  const citations = citedImageFilenames(boundedText)
  return {
    textFile: new File([boundedText], file.name, { type: "text/plain" }),
    images: extractedImages.filter((image) => citations.has(image.name)),
  }
}