theseus/bundled-addons/pdf-editor/lib/create.js

// Making a PDF where there wasn't one: photos in, or several files welded
// into one. Both paths hand back plain bytes, which the editor then opens as
// if they had come off disk — so everything downstream (marks, page rail,
// save, convert) works on them without knowing where they came from.

import { PDFDocument } from "../vendor/pdf-lib/pdf-lib.esm.min.js";

// Points, the PDF's own unit. A4 is the world's paper; Letter is North
// America's. "fit" below means neither: the page takes the image's shape.
export const PAGE_SIZES = {
  a4: [595.28, 841.89],
  letter: [612, 792],
};
// The long edge of a fitted page. A phone photo is 4000 px on its longest
// side; read as points that would be a 55-inch page, technically valid and
// useless to anyone who prints it. Mapping the long edge to A4's long edge
// keeps the image's proportions and gives a page a printer understands.
const FIT_LONG_EDGE = 841.89;

const JPEG = [0xff, 0xd8, 0xff];
const PNG = [0x89, 0x50, 0x4e, 0x47];
const starts = (bytes, sig) => sig.every((b, i) => bytes[i] === b);

/**
 * pdf-lib embeds JPEG and PNG, and nothing else. Rather than refuse a WebP or
 * a HEIC-turned-WebP screenshot, hand it to the browser — which can decode far
 * more than pdf-lib can embed — and re-encode as PNG. Lossless, so a screenshot
 * survives intact; a photo re-encoded this way grows, which is why anything
 * already JPEG is passed through untouched.
 */
async function toEmbeddable(file) {
  const bytes = new Uint8Array(await file.arrayBuffer());
  if (starts(bytes, JPEG)) return { bytes, kind: "jpg" };
  if (starts(bytes, PNG)) return { bytes, kind: "png" };
  let bitmap;
  try { bitmap = await createImageBitmap(file); }
  catch { throw new Error(`${file.name} is not an image this browser can read`); }
  const canvas = new OffscreenCanvas(bitmap.width, bitmap.height);
  canvas.getContext("2d").drawImage(bitmap, 0, 0);
  bitmap.close();
  const blob = await canvas.convertToBlob({ type: "image/png" });
  return { bytes: new Uint8Array(await blob.arrayBuffer()), kind: "png" };
}

/**
 * One image per page, in the order given.
 *
 * `page` is "fit" (the page takes each image's shape), "a4" or "letter". For
 * the two paper sizes the sheet turns landscape when the image is wider than
 * it is tall, because a landscape photo on a portrait page is mostly margin.
 * `margin` is in points and ignored by "fit" — the whole point of fitting the
 * page to the image is that there is no border.
 */
export async function imagesToPdf(files, { page = "fit", margin = 18, onProgress } = {}) {
  if (!files.length) throw new Error("no images given");
  const pdf = await PDFDocument.create();
  const failed = [];
  let n = 0;
  for (const file of files) {
    onProgress?.(++n, files.length, file.name);
    let img;
    try {
      const { bytes, kind } = await toEmbeddable(file);
      img = kind === "jpg" ? await pdf.embedJpg(bytes) : await pdf.embedPng(bytes);
    } catch (e) {
      failed.push({ name: file.name, why: e?.message || String(e) });
      continue;
    }
    if (page === "fit") {
      const scale = FIT_LONG_EDGE / Math.max(img.width, img.height);
      const w = img.width * scale, h = img.height * scale;
      pdf.addPage([w, h]).drawImage(img, { x: 0, y: 0, width: w, height: h });
      continue;
    }
    const [pw, ph] = PAGE_SIZES[page] || PAGE_SIZES.a4;
    const landscape = img.width > img.height;
    const sheet = pdf.addPage(landscape ? [ph, pw] : [pw, ph]);
    const boxW = sheet.getWidth() - margin * 2;
    const boxH = sheet.getHeight() - margin * 2;
    // contain, never crop: the smaller ratio wins, and the leftover becomes
    // an even border on the two sides that did not bind.
    const scale = Math.min(boxW / img.width, boxH / img.height);
    const w = img.width * scale, h = img.height * scale;
    sheet.drawImage(img, {
      x: (sheet.getWidth() - w) / 2,
      y: (sheet.getHeight() - h) / 2,
      width: w, height: h,
    });
  }
  if (!pdf.getPageCount()) {
    throw new Error(failed.length ? `none of those files could be read (${failed[0].why})` : "no pages were made");
  }
  return { bytes: await pdf.save(), pages: pdf.getPageCount(), failed };
}

/**
 * Several PDFs, end to end, in the order given.
 *
 * Page content, annotations and links inside each document survive, because
 * copyPages carries the whole page object across. What does NOT survive is
 * anything that lives above the page: form fields, the outline, and
 * attachments. Those belong to the document they came from, and merging four
 * documents that each define a field called "name" has no correct answer.
 * The editor says so plainly rather than silently dropping them.
 */
export async function mergePdfs(files, { onProgress } = {}) {
  if (files.length < 2) throw new Error("pick at least two PDFs to merge");
  const out = await PDFDocument.create();
  const parts = [];
  const failed = [];
  let hadForms = false;
  let n = 0;
  for (const file of files) {
    onProgress?.(++n, files.length, file.name);
    try {
      const bytes = new Uint8Array(await file.arrayBuffer());
      // An encrypted-but-openable PDF is the common case (print/copy flags,
      // no password). ignoreEncryption lets those through; a real
      // password-protected file still throws, and lands in `failed`.
      const src = await PDFDocument.load(bytes, { ignoreEncryption: true });
      try { if (src.getForm().getFields().length) hadForms = true; } catch {}
      const copied = await out.copyPages(src, src.getPageIndices());
      for (const p of copied) out.addPage(p);
      parts.push({ name: file.name, pages: copied.length });
    } catch (e) {
      failed.push({ name: file.name, why: e?.message || String(e) });
    }
  }
  if (!out.getPageCount()) {
    throw new Error(failed.length ? `nothing could be read (${failed[0].why})` : "no pages were copied");
  }
  return { bytes: await out.save(), pages: out.getPageCount(), parts, failed, hadForms };
}