DOCX comment bookmarks cause `Unknown node type: bookmarkStart` and drop comments

[ohcedar]

Summary

DOCX comments containing w:bookmarkStart / w:bookmarkEnd nodes fail during comment HTML conversion with:

RangeError: Unknown node type: bookmarkStart
Failed to convert comment

The affected comment is then omitted from the loaded comments list.

Minimal shape

A comment containing an invisible bookmark pair around visible mention text is sufficient:

<w:comment w:id="1">
  <w:p>
    <w:r><w:t>Before mention </w:t></w:r>
    <w:bookmarkStart w:id="42" w:name="_mention"/>
    <w:r><w:t>@Person</w:t></w:r>
    <w:bookmarkEnd w:id="42"/>
  </w:p>
</w:comment>

Root cause

The full DOCX editor schema includes bookmarkStart and bookmarkEnd, so the comment XML is converted to valid DOCX-schema JSON.

During processLoadedDocxComments, getHtmlFromComment normalizes that JSON and creates a headless text editor using getRichTextExtensions(). That smaller schema does not include bookmark nodes, but normalizeCommentForEditor currently passes them through. schema.nodeFromJSON throws, the catch returns no HTML, and this guard skips the non-tracked comment:

if (!htmlContent && !comment.trackedChange) {
  return;
}

Expected behavior

Invisible bookmark boundary nodes should not prevent the visible comment content from importing.

Suggested narrow fix

Ignore only bookmark boundary nodes while producing the comment's display HTML:

if (node.type === 'bookmarkStart' || node.type === 'bookmarkEnd') {
  return null;
}

This leaves the original docxCommentJSON unchanged, so the imported DOCX representation still retains the bookmark metadata. In a local reproduction with two affected comments, both comments imported successfully, visible text and mentions remained intact, the original bookmark JSON was preserved, and the console errors disappeared.

Versions checked

• @harbour-enterprises/superdoc@1.44.1
• @harbour-enterprises/superdoc@1.45.0
• Current main implementation of packages/superdoc/src/stores/comments-store.js

[linear-code]

IT-1290

[gpardhivvarma]

im picking this up

[gpardhivvarma]

Opened #3829 to fix this.

The fix normalizes comment content defensively: normalizeCommentForEditor now drops/unwraps any node type the reduced rich-text schema (getRichTextExtensions()) doesn't register — not just bookmarkStart/bookmarkEnd — so the whole class of Unknown node type crashes is prevented. Invisible boundary nodes are dropped, nodes wrapping content are unwrapped so the visible text/mention survives, and the original docxCommentJSON is left untouched so exported bookmark metadata is preserved.

Includes a pure unit test plus an end-to-end regression test that exercises the real headless Editor (verified to fail before the fix and pass after).