From ac115d96fbe1dc1730b3839091738b71a17e694c Mon Sep 17 00:00:00 2001
From: iomgaa <iomgaaycz@gmail.com>
Date: Tue, 14 Jul 2026 15:44:43 -0400
Subject: [PATCH] test: cover backfill params end-to-end (mutation isolation +
 seq_offset)

---
 tests/integration/test_pipeline_v2.py | 84 +++++++++++++++++++++++++++
 1 file changed, 84 insertions(+)

diff --git a/tests/integration/test_pipeline_v2.py b/tests/integration/test_pipeline_v2.py
index fbaf5dc..f089245 100644
--- a/tests/integration/test_pipeline_v2.py
+++ b/tests/integration/test_pipeline_v2.py
@@ -580,6 +580,90 @@ class TestPipelineV2:
         if result.accepted:
             assert len(result.heavy_sampled) == len(result.accepted)
 
+    @pytest.mark.asyncio
+    async def test_backfill_params_do_not_mutate_caller_objects(
+        self, tree, default_config, store, tmp_path
+    ):
+        """补生成三参：传入的 used_node_ids / embed_pool 对象不被就地修改。
+
+        内部应复制而非别名 —— run 完之后调用方传入的容器长度/内容保持原样。
+        """
+        vlm = MockVLM([_make_candidate_json(f"Q{i}?") for i in range(20)])
+        llm = MockLLM([_make_gate_pass_response()] * 100)
+
+        config = PipelineConfig(
+            per_type=2,
+            retry_limit=2,
+            heavy_sample_rate=0.0,
+            dedup_threshold=0.85,
+            concurrency=1,
+            seed=42,
+            output_dir=tmp_path / "out",
+        )
+
+        initial_used_node_ids = {"some_node"}
+        seed_vec = _mock_embed_fn("seed embedding text")
+        initial_embed_pool = [seed_vec]
+
+        result = await run_pipeline_v2(
+            video_ids=["vid_L1_000"],
+            trees={"vid_L1_000": tree},
+            vlm=vlm,
+            llm=llm,
+            embed_fn=_mock_embed_fn,
+            store=store,
+            config=config,
+            task_types=["Action Recognition"],
+            initial_used_node_ids=initial_used_node_ids,
+            initial_embed_pool=initial_embed_pool,
+        )
+
+        # 至少接受一题，确保内部确实往副本里追加了 node/embedding
+        assert len(result.accepted) > 0
+
+        # 调用方对象未被 mutation：集合仍只含原始节点
+        assert initial_used_node_ids == {"some_node"}
+        # embed_pool 仍只含最初的一个向量，且内容未变
+        assert len(initial_embed_pool) == 1
+        assert np.array_equal(initial_embed_pool[0], seed_vec)
+
+    @pytest.mark.asyncio
+    async def test_seq_offset_continues_question_ids(
+        self, tree, default_config, store, tmp_path
+    ):
+        """补生成三参：seq_offset 端到端续编 question_id 的 seq，避开旧 run 号段。"""
+        vlm = MockVLM([_make_candidate_json(f"Q{i}?") for i in range(20)])
+        llm = MockLLM([_make_gate_pass_response()] * 100)
+
+        config = PipelineConfig(
+            per_type=2,
+            retry_limit=2,
+            heavy_sample_rate=0.0,
+            dedup_threshold=0.85,
+            concurrency=1,
+            seed=42,
+            output_dir=tmp_path / "out",
+        )
+
+        result = await run_pipeline_v2(
+            video_ids=["vid_L1_000"],
+            trees={"vid_L1_000": tree},
+            vlm=vlm,
+            llm=llm,
+            embed_fn=_mock_embed_fn,
+            store=store,
+            config=config,
+            task_types=["Action Recognition"],
+            seq_offset=30,
+        )
+
+        assert len(result.accepted) > 0
+        # question_id 格式 "{video_id}_{task_type}_{seq:04d}"，seq 是最后一段 4 位数字
+        seqs = [int(q.question_id.rsplit("_", 1)[1]) for q in result.accepted]
+        assert all(s > 30 for s in seqs), seqs
+        # per_type=2 → seq 从 31 起续编，不与 0-30 撞
+        assert min(seqs) == 31
+
     @pytest.mark.asyncio
     async def test_store_records_all(self, tree, default_config, store, tmp_path):
         """验证 store 中记录了每道题的生成与门控结果。"""