fix(data-process): 修正任务详情数据契约

backend/tests/test_data_process_store.py

@@ -274,6 +274,10 @@ class _TaskDetailConnection:
         assert "dataset.source_task_id=task.id" in normalized
         assert "dataset.source_task_id IS NULL AND dataset.task_id=task.id" in normalized
         assert "dataset.deleted_at IS NULL" in normalized
+        assert "dataset.metadata::jsonb" not in normalized
+        assert "WHEN 'train' THEN 'train'" in normalized
+        assert "WHEN 'val' THEN 'validation'" in normalized
+        assert "WHEN 'test' THEN 'test'" in normalized
         task_id = params[0]
         visible = [
             dataset
@@ -364,6 +368,61 @@ class _TaskListStore(DataProcessStore):
         yield self._conn
 
 
+class _StartGenerationConnection:
+    def __init__(self) -> None:
+        self.task = {
+            **_regeneration_task(
+                status="pending",
+                output_dataset_id=None,
+                output_count=28,
+            ),
+            "generation_run_id": None,
+        }
+        self.results = [{"id": "old-result"}]
+
+    def execute(self, sql: str, params: Any = None) -> _Result:
+        normalized = " ".join(sql.split())
+        if normalized.startswith("SELECT COUNT(*) AS count FROM data_process_preview_items"):
+            return _Result(row={"count": 1})
+        if normalized.startswith("DELETE FROM data_process_results"):
+            self.results.clear()
+            return _Result()
+        assert normalized.startswith("UPDATE data_process_tasks SET status='running'")
+        assert "output_count=0" in normalized
+        self.task.update(
+            {
+                "status": "running",
+                "progress": 30,
+                "output_count": 0,
+                "failure_reason": None,
+                "started_at": params[0],
+                "completed_at": None,
+                "filtered_count": 0,
+                "duplicate_count": 0,
+                "error_count": 0,
+                "generation_run_id": params[1],
+                "updated_at": params[2],
+            }
+        )
+        return _Result(row=dict(self.task))
+
+
+class _StartGenerationStore(DataProcessStore):
+    def __init__(self, conn: _StartGenerationConnection) -> None:
+        self._conn = conn
+
+    @contextmanager
+    def connect(self) -> Iterator[_StartGenerationConnection]:
+        yield self._conn
+
+    def _task_in_connection(
+        self, conn: Any, task_id: str, *, for_update: bool = False
+    ) -> dict[str, Any]:
+        assert task_id == "task-1"
+        assert for_update is True
+        return dict(self._conn.task)
+
+
 def test_decode_row_serializes_postgres_numeric_values_as_json_numbers() -> None:
     decoded = _decode_row(
         {
@@ -522,6 +581,16 @@ def test_list_tasks_exposes_source_and_generation_counts() -> None:
     assert item["output_dataset_id"] is None
 
 
+def test_start_generation_clears_previous_output_count() -> None:
+    conn = _StartGenerationConnection()
+
+    task = _StartGenerationStore(conn).start_generation("task-1")
+
+    assert task["status"] == "running"
+    assert task["output_count"] == 0
+    assert conn.results == []
+
+
 def _legacy_published_datasets(task_id: str = "task-1") -> list[dict[str, Any]]:
     specs = (
         ("dataset_train", "制度问答-训练集", "train", "train", 22),
@@ -585,6 +654,8 @@ def test_prepare_regeneration_preserves_outputs_sources_and_generation_only_prev
 
 def test_prepare_regeneration_backfills_and_keeps_legacy_task_datasets_visible() -> None:
     legacy_datasets = _legacy_published_datasets()
+    # 历史 metadata 可能不是合法 JSON，详情查询不能再依赖 metadata::jsonb。
+    legacy_datasets[0]["metadata"] = "{legacy-invalid-json"
     deleted_dataset = {
         **legacy_datasets[0],
         "id": "dataset_deleted",