fix(data-process): 合并文档结构短切片

backend/tests/test_data_process_api.py

@@ -1402,7 +1402,6 @@ def test_every_unstructured_preprocess_option_changes_preview_behavior() -> None
     )
     assert without_merge[0]["token_count"] < 5
     assert with_merge[0]["token_count"] >= 5
-
     mojibake = "这是无法可靠读取的内容，锟斤拷锟斤拷锟斤拷，需要预先过滤。"
     assert len(_preview_task(mojibake, options=[])) == 1
     assert _preview_task(mojibake, options=["filter_low_quality"]) == []
@@ -1451,6 +1450,25 @@ def test_every_unstructured_preprocess_option_changes_preview_behavior() -> None
     assert "[EMAIL]" in masked["edited_content"]
 
 
+def test_merge_short_content_applies_across_adjacent_structure_sections() -> None:
+    content = "\n".join(f"{index}. 小节{index}\n内容{index}。" for index in range(1, 9))
+    items = _preview_task(
+        content,
+        options=["merge_short_content"],
+        config={
+            "chunk_method": "structure",
+            "chunk_size": 40,
+            "chunk_overlap": 0,
+            "min_chunk_size": 20,
+        },
+    )
+
+    assert len(items) == 2
+    assert all(20 <= item["token_count"] <= 40 for item in items)
+    assert items[0]["source_start_line"] == 1
+    assert items[0]["source_end_line"] == 8
+
+
 def test_stored_binary_document_text_is_not_reparsed_as_binary() -> None:
     for file_format in ("pdf", "docx", "pptx"):
         items = _preview_task(