Skip to content

Build and run agents you can see, understand and trust.

fix(rag): preserve image-only and header-only Excel sheets (#2666)

JasonHd60d58ec7429
diff --git a/src/agentscope/rag/_parser/_excel.py b/src/agentscope/rag/_parser/_excel.py
index 4f0d80939..3288ffb71 100644
--- a/src/agentscope/rag/_parser/_excel.py
+++ b/src/agentscope/rag/_parser/_excel.py
@@ -134,7 +134,8 @@ class ExcelParser(ParserBase):
Each sheet is scanned for tabular data and (optionally) images.
Tables are rendered as Markdown pipe-tables or JSON arrays; images
- are emitted as standalone :class:`DataBlock` sections.
+ are emitted as standalone :class:`DataBlock` sections, including on
+ sheets without cell values. Header-only tables are preserved.
When ``separate_sheet=True`` each sheet becomes a batch of
sections that never intermix with other sheets, making it possible
@@ -310,24 +311,23 @@ class ExcelParser(ParserBase):
logger.warning("Failed to parse sheet '%s': %s", sheet_name, e)
return sheet_sections
- if df.empty:
- return sheet_sections
-
- table_data = _extract_table_data(df)
-
- if self.table_format == "markdown":
- table_text = self._table_to_markdown(table_data, sheet_name)
- else:
- table_text = self._table_to_json(table_data, sheet_name)
+ # A header-only sheet is "empty" to pandas but still has columns
+ if len(df.columns) > 0:
+ table_data = _extract_table_data(df)
- if table_text:
- sheet_sections.append(
- Section(
- content=TextBlock(text=table_text),
- source=filename,
- metadata={"sheet": sheet_name},
- ),
- )
+ if self.table_format == "markdown":
+ table_text = self._table_to_markdown(table_data, sheet_name)
+ else:
+ table_text = self._table_to_json(table_data, sheet_name)
+
+ if table_text:
+ sheet_sections.append(
+ Section(
+ content=TextBlock(text=table_text),
+ source=filename,
+ metadata={"sheet": sheet_name},
+ ),
+ )
if self.include_image and workbook is not None:
try:
diff --git a/tests/rag_parser_test.py b/tests/rag_parser_test.py
index 100cdb2b5..9ef6b81c5 100644
--- a/tests/rag_parser_test.py
+++ b/tests/rag_parser_test.py
@@ -1022,6 +1022,76 @@ class PPTParserTest(IsolatedAsyncioTestCase):
class ExcelParserTest(IsolatedAsyncioTestCase):
"""Behavioural coverage for :class:`ExcelParser`."""
+ async def test_header_only_sheet(self) -> None:
+ """A sheet with only a header row is kept as a table."""
+ xlsx_bytes = _make_xlsx_simple({"Data": [["Revenue", "Year"]]})
+ sections = await ExcelParser().parse(xlsx_bytes, "header.xlsx")
+
+ self.assertListEqual(
+ [s.model_dump() for s in sections],
+ [
+ {
+ "content": {
+ "type": "text",
+ "text": (
+ "Sheet: Data\n"
+ "| Revenue | Year |\n"
+ "| --- | --- |\n"
+ ),
+ "id": AnyString(),
+ "created_at": AnyString(),
+ "finished_at": None,
+ },
+ "source": "header.xlsx",
+ "metadata": {},
+ },
+ ],
+ )
+
+ async def test_blank_sheet_has_no_sections(self) -> None:
+ """A blank sheet produces neither a table nor images."""
+ xlsx_bytes = _make_xlsx_simple({"Blank": []})
+ sections = await ExcelParser().parse(xlsx_bytes, "blank.xlsx")
+
+ self.assertListEqual(sections, [])
+
+ async def test_image_only_sheet(self) -> None:
+ """Images on a sheet without cell values are still extracted."""
+ from openpyxl import Workbook
+ from openpyxl.drawing.image import Image
+
+ workbook = Workbook()
+ workbook.active.title = "Images"
+ workbook.active.add_image(Image(io.BytesIO(_PNG_PIXEL)), "A3")
+ buffer = io.BytesIO()
+ workbook.save(buffer)
+ workbook.close()
+
+ parser = ExcelParser(include_image=True)
+ sections = await parser.parse(buffer.getvalue(), "images.xlsx")
+
+ self.assertListEqual(
+ [s.model_dump() for s in sections],
+ [
+ {
+ "content": {
+ "type": "data",
+ "id": AnyString(),
+ "source": {
+ "type": "base64",
+ "data": _PNG_PIXEL_B64,
+ "media_type": "image/png",
+ },
+ "name": "images.xlsx",
+ "created_at": AnyString(),
+ "finished_at": None,
+ },
+ "source": "images.xlsx",
+ "metadata": {"sheet": "Images", "media_type": "image/png"},
+ },
+ ],
+ )
+
async def test_single_sheet_markdown(self) -> None:
"""A single-sheet workbook produces one text Section with
Markdown table."""