Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

Studio: let a finished reply be edited when its settle write was refused (#14131)

Daniel Hanea71dc56cbb8
diff --git a/studio/backend/storage/studio_db.py b/studio/backend/storage/studio_db.py
index a5ca2bab702..6eb38a932fc 100644
--- a/studio/backend/storage/studio_db.py
+++ b/studio/backend/storage/studio_db.py
@@ -3834,11 +3834,12 @@ def _detach_terminal_generation_for_edit(
return False
stored_metadata = json.loads(row["metadata_json"]) if row["metadata_json"] else {}
incoming_metadata = message.get("metadata")
+ # Not generationSettled: a settle whose `incomplete` disagrees with the producer's is refused,
+ # so a finished reply would never become editable (#13938). The run status above decides.
if (
not isinstance(stored_metadata, dict)
or stored_metadata.get("serverManaged") is not True
or stored_metadata.get("generationRunId") != row["run_id"]
- or stored_metadata.get("generationSettled") is not True
or isinstance(incoming_metadata, dict)
and incoming_metadata.get("serverManaged") is True
or str(message.get("role")) != str(row["role"])
diff --git a/studio/backend/tests/test_chat_generation_runs.py b/studio/backend/tests/test_chat_generation_runs.py
index 0466d01e5c5..e05bd8bbb60 100644
--- a/studio/backend/tests/test_chat_generation_runs.py
+++ b/studio/backend/tests/test_chat_generation_runs.py
@@ -417,6 +417,57 @@ def test_explicit_edit_keeps_display_metadata_but_not_the_run_claim(chat_home):
assert runs_db.get_run("run-1", "alice") is None
+ "finish_reason, tab_reason",
+ [("stop", "empty"), ("length", "context_window")],
+)
+def test_finished_generation_whose_settle_was_refused_can_still_be_edited(
+ chat_home, finish_reason, tab_reason
+):
+ """#13938: a settle whose `incomplete` disagrees with the producer's is refused, so the row
+ never gets generationSettled; the run is finished, so an edit must still detach it."""
+ _create()
+ token = runs_db.get_worker_token("run-1")
+ assert runs_db.mark_running("run-1", token)
+ assert runs_db.finish_run(
+ "run-1", worker_token = token, status = "completed", finish_reason = finish_reason
+ )
+ run = runs_db.get_run("run-1", "alice")
+ stored = studio_db.get_chat_message("thread-1", "assistant-1")
+ settle = json.loads(json.dumps(stored))
+ settle["metadata"].update(
+ {
+ "generationSeq": run["lastEventSeq"],
+ "generationStatus": "completed",
+ "generationSettled": True,
+ "incomplete": {"reason": tab_reason},
+ }
+ )
+ studio_db.sync_chat_messages("thread-1", [settle])
+ stored = studio_db.get_chat_message("thread-1", "assistant-1")
+ assert stored["metadata"].get("generationSettled") is not True, "the settle landed"
+
+ edited = {key: value for key, value in stored.items() if key != "metadata"}
+ edited["content"] = [{"type": "text", "text": "edited"}]
+ edited["metadata"] = {"incomplete": {"reason": tab_reason}}
+ saved = studio_db.upsert_chat_message(edited, allow_generation_edit = True)
+ assert saved["content"] == [{"type": "text", "text": "edited"}]
+ assert runs_db.get_run("run-1", "alice") is None
+ _assert_protected(settle)
+
+
+def test_running_generation_still_cannot_be_explicitly_edited(chat_home):
+ _create()
+ token = runs_db.get_worker_token("run-1")
+ assert runs_db.mark_running("run-1", token)
+ stored = studio_db.get_chat_message("thread-1", "assistant-1")
+ edited = {key: value for key, value in stored.items() if key != "metadata"}
+ edited["content"] = [{"type": "text", "text": "edited"}]
+ with pytest.raises(studio_db.ChatMessageProtectedError):
+ studio_db.upsert_chat_message(edited, allow_generation_edit = True)
+ assert runs_db.get_run("run-1", "alice")["status"] == "running"
+
+
def test_batched_events_have_gapless_cursor_and_terminal_flush(chat_home):
run, _created = _create()
worker_token = runs_db.get_worker_token("run-1")