Skip to content

bug: non-breaking spaces disappear when a document is opened and saved #35

Closed

Description

@Riyooo

Summary

parseXml drops a text node whose content is only a no-break space (U+00A0) or a narrow no-break space (U+202F). A <w:t> holding just that character is therefore emptied when a document is opened, and the space is missing from the saved file, even in paragraphs the caller never touched.

French typography puts a no-break or narrow no-break space before :, ;, !, ? and inside « ». When the text language is French (France), Word inserts one as you type, or converts the ordinary space typed before the sign, as I checked in my own Word, and writers also type them by hand (Ctrl+Shift+Space in Word, AltGr+Space or AltGr+Shift+Space on the French AZERTY keyboard). A phrase like Nom : Dupont comes out as Nom: Dupont, as the reproduction below shows, and nothing reports the loss.

The same happens to any such space that sits alone in its run, for example:

  • Durée : 18 mois → Durée: 18 mois
  • « Oui » → «Oui»
  • 10 000 € → 10000€

I found it by checking that a document opened and saved with no edit comes back unchanged. Two French documents saved by Word did not:

  • word/document.xml lost 5 and 11 characters respectively;
  • every lost character was a no-break or narrow no-break space sitting alone in its own run, with no xml:space="preserve";
  • all the other parts came back byte-identical.

The cause is the test that drops white-space-only text between elements, in src/internal/xml/parser.ts. It keeps a text node only when value.trim().length > 0, or under xml:space="preserve", or in mixed content. String.prototype.trim treats U+00A0 and U+202F as white space; XML does not (details below).

Reproduction

import { strFromU8, strToU8, unzipSync, zipSync } from "fflate";
import { openDocx, paragraphs, paragraphText, toUint8Array } from "@office-kit/docx";

const show = (s) => s.replace(/[\u00a0\u202f]/g, (c) => `\\u${c.charCodeAt(0).toString(16).padStart(4, "0")}`);
const xml = (t) =>
  `<?xml version="1.0" encoding="UTF-8" standalone="yes"?><w:document xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main"><w:body><w:p>` +
  `<w:r><w:t>Nom</w:t></w:r><w:r><w:t>${t}</w:t></w:r><w:r><w:t>: Dupont</w:t></w:r></w:p></w:body></w:document>`;

for (const ch of ["\u202f", "\u00a0"]) {
  const bytes = zipSync({
    "[Content_Types].xml": strToU8(`<?xml version="1.0" encoding="UTF-8" standalone="yes"?><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>`),
    "_rels/.rels": strToU8(`<?xml version="1.0" encoding="UTF-8" standalone="yes"?><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>`),
    "word/document.xml": strToU8(xml(ch)),
  });
  const doc = openDocx(bytes);
  console.log("read: ", show(paragraphText(paragraphs(doc)[0])));
  const saved = strFromU8(unzipSync(toUint8Array(doc))["word/document.xml"]);
  console.log("saved:", show(saved.match(/<w:t[^>]*>[^<]*<\/w:t>|<w:t\/>/g).join("")));
}

Expected behavior

The character is kept, as for any other character data. Output with the condition changed as proposed below:

read:  Nom\u202f: Dupont
saved: <w:t>Nom</w:t><w:t>\u202f</w:t><w:t>: Dupont</w:t>
read:  Nom\u00a0: Dupont
saved: <w:t>Nom</w:t><w:t>\u00a0</w:t><w:t>: Dupont</w:t>

Actual behavior

The character is dropped on read, and the saved part holds an empty <w:t>:

read:  Nom: Dupont
saved: <w:t>Nom</w:t><w:t></w:t><w:t>: Dupont</w:t>
read:  Nom: Dupont
saved: <w:t>Nom</w:t><w:t></w:t><w:t>: Dupont</w:t>

Environment

  • Package version: @office-kit/docx 0.2.0 (the same line is on main at a5fe1b8)
  • Runtime: Node 24.17.0 on Ubuntu 26.04 x86_64
  • Other relevant details: none, the reproduction uses only the published package and fflate

Additional context

The condition is value.trim().length > 0 in src/internal/xml/parser.ts (line 318 on main). String.prototype.trim removes ECMAScript white space and line terminators: U+0009–U+000D, U+0020, U+00A0, U+1680, U+2000–U+200A, U+2028, U+2029, U+202F, U+205F, U+3000, U+FEFF. XML white space (S, production [3] of XML 1.0) is only U+0020, U+0009, U+000D and U+000A, the set cursorSkipWhitespace already uses. So U+3000 (ideographic space) is affected as well.

Testing only XML white space fixes it:

if (xmlSpace === "preserve" || /[^ \t\r\n]/.test(value) || hasNonElementSiblings(children)) {

With this change applied to the 0.2.0 build, both documents round-trip byte-identical (all 40 and 32 parts). A regression test along these lines:

it("keeps a text node made only of non-XML white space", () => {
  for (const ch of ["\u00a0", "\u202f", "\u3000"]) {
    const doc = parseXml(`<w:t xmlns:w="urn:test">${ch}</w:t>`);
    expect(doc.root.children).toEqual([{ kind: "text", value: ch }]);
  }
});

Confirmation

  • I have searched existing issues and confirmed this is not a duplicate.
  • I am on the latest released version, or I have explained above why an older version is relevant.
  • I have included a minimal reproduction that I have actually run myself.
  • If I used an LLM to draft this issue, I have read and verified every claim and am willing to defend them in follow-up.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions