fix(microsoft-word): strip XML-forbidden control characters from generated documents

Author: waleedlatif1

apps/sim/lib/microsoft-word/document.server.test.ts

@@ -12,6 +12,9 @@ import {
   replaceTextInDocx,
 } from '@/lib/microsoft-word/document.server'
 
+/** The characters XML 1.0 forbids in a text node, as a fresh (unstateful) matcher. */
+const INVALID_XML_CHARS_PATTERN = /[\u0000-\u0008\u000B\u000C\u000E-\u001F]/
+
 /** Reads one XML part out of a generated package. */
 async function readPart(buffer: Buffer, path = 'word/document.xml'): Promise<string> {
   const zip = await JSZip.loadAsync(buffer)
@@ -89,6 +92,22 @@ describe('buildDocxFromContent', () => {
     expect(text).toContain('2 * 3 = 6')
   })
 
+  it('strips control characters that would make the package unopenable', async () => {
+    // The docx package writes run text into the XML verbatim, so a forbidden
+    // character reaching it produces invalid XML 1.0 rather than a rendering bug.
+    const buffer = await buildDocxFromContent(`before${String.fromCharCode(8)}after`)
+    const xml = await readDocumentXml(buffer)
+
+    expect(INVALID_XML_CHARS_PATTERN.test(xml)).toBe(false)
+    expect(xml).toContain('beforeafter')
+  })
+
+  it('keeps tabs, which are legal XML, while stripping forbidden characters', async () => {
+    const buffer = await buildDocxFromContent(`a\tb${String.fromCharCode(0)}c`)
+
+    expect(await extractDocxText(buffer)).toContain('a\tbc')
+  })
+
   it('still produces an openable document for empty content', async () => {
     const buffer = await buildDocxFromContent('')
     const xml = await readDocumentXml(buffer)

apps/sim/lib/microsoft-word/document.server.ts

@@ -34,6 +34,11 @@ const INLINE_EMPHASIS_PATTERN = /\*\*([^*]+)\*\*|\*([^*]+)\*/g
 /** Characters XML 1.0 forbids in a text node. */
 const INVALID_XML_CHARS = /[\u0000-\u0008\u000B\u000C\u000E-\u001F]/g
 
+/** Removes the characters XML 1.0 forbids in a text node. */
+function stripInvalidXmlChars(value: string): string {
+  return value.replace(INVALID_XML_CHARS, '')
+}
+
 /**
  * Splits a line into bold / italic runs using the Markdown subset Sim supports:
  * `**bold**` and `*italic*`. Anything else is emitted verbatim, so unmatched
@@ -77,7 +82,12 @@ function parseInlineRuns(line: string): ContentRun[] {
 function parseContentBlocks(content: string): ContentBlock[] {
   const blocks: ContentBlock[] = []
 
-  for (const rawLine of content.replace(/\r\n?/g, '\n').split('\n')) {
+  // Stripped once here rather than per run: the `docx` package writes run text
+  // into the XML verbatim, so a stray control character from an upstream block
+  // would produce a package Word refuses to open. Tab, newline, and carriage
+  // return are legal XML and are outside the class, so the split below is
+  // unaffected.
+  for (const rawLine of stripInvalidXmlChars(content).replace(/\r\n?/g, '\n').split('\n')) {
     const line = rawLine.trimEnd()
     if (line.trim().length === 0) continue
 
@@ -132,8 +142,7 @@ export async function buildDocxFromContent(content: string, title?: string): Pro
 
 /** Escapes text for an XML text node and drops characters XML 1.0 forbids. */
 function escapeXmlText(value: string): string {
-  return value
-    .replace(INVALID_XML_CHARS, '')
+  return stripInvalidXmlChars(value)
     .replace(/&/g, '&')
     .replace(/</g, '&lt;')
     .replace(/>/g, '&gt;')