Keep the video spec response encodable with a half-deleted emoji

backend/app/routers/video.py

@@ -238,6 +238,20 @@ def video_spec(payload: VideoRenderRequest, request: Request, session: Session =
         note_id=note.id,
     )
     return DataResponse(data=VideoSpecRead(
-        spec=imported.spec.model_dump(mode="json", by_alias=True, exclude_none=True),
-        warnings=imported.warnings,
+        spec=_utf8_safe(imported.spec.model_dump(mode="json", by_alias=True, exclude_none=True)),
+        warnings=_utf8_safe(imported.warnings),
     ))
+
+
+def _utf8_safe(value):
+    """Replace lone UTF-16 surrogates — what a half-deleted emoji leaves in a
+    note — which a Python string can hold but a UTF-8 response cannot, so one
+    stray character would otherwise fail the whole response. Rendering keeps
+    the text as it is; only this document's copy is touched."""
+    if isinstance(value, str):
+        return value.encode("utf-16", "surrogatepass").decode("utf-16", "replace")
+    if isinstance(value, list):
+        return [_utf8_safe(v) for v in value]
+    if isinstance(value, dict):
+        return {k: _utf8_safe(v) for k, v in value.items()}
+    return value

backend/tests/test_video_spec.py

@@ -484,3 +484,17 @@ def test_the_spec_route_refuses_another_users_note_and_bad_options(notes):
         _spec_route(notes, options={"aspect": "4:3"})
     assert invalid.value.status_code == 400
 
+
+def test_the_spec_route_survives_half_of_a_deleted_emoji(notes):
+    # JSON.stringify writes a lone surrogate as \ud83d; decoded, it is a valid
+    # Python string but not valid UTF-8, so the response could not be encoded.
+    note = notes.get(Note, "mine")
+    note.content = json.dumps([
+        {"id": "p", "type": "paragraph", "content": [{"type": "text", "text": "Hi \ud83d there \U0001f98e"}]},
+    ])
+    notes.add(note)
+    notes.commit()
+    data = _spec_route(notes, options={"title_card": False})
+    encoded = json.dumps(data.model_dump(), ensure_ascii=False).encode("utf-8")
+    assert "Hi \ufffd there \U0001f98e." in encoded.decode("utf-8")
+