Skip to content

RAGFlow is an agent-native database: a Retrieval-Augmented Generation (RAG) engine that grew into context infrastructure.

fix(parser): accept longer closing fences during table extraction (#20028)

tri4tanaf4014e6037de
diff --git a/deepdoc/parser/markdown_parser.py b/deepdoc/parser/markdown_parser.py
index 3f4583d783..92b15bfb75 100644
--- a/deepdoc/parser/markdown_parser.py
+++ b/deepdoc/parser/markdown_parser.py
@@ -43,7 +43,8 @@ class RAGFlowMarkdownParser:
# MarkdownElementExtractor already shields fences (see _fenced_code_ranges), so
# keep table extraction consistent with it.
fence_pattern = re.compile(
- r"^[ \t]{0,3}(`{3,}|~{3,})[^\n]*\n.*?(?:^[ \t]{0,3}\1[ \t]*$|\Z)",
+ r"^[ \t]{0,3}(?:(?P<backticks>`{3,})|(?P<tildes>~{3,}))[^\n]*\n"
+ r".*?(?:^[ \t]{0,3}(?(backticks)(?P=backticks)`*|(?P=tildes)~*)[ \t]*$|\Z)",
re.MULTILINE | re.DOTALL,
)
diff --git a/test/unit_test/deepdoc/parser/test_markdown_parser.py b/test/unit_test/deepdoc/parser/test_markdown_parser.py
index 1395a80978..7d5ed0ca19 100644
--- a/test/unit_test/deepdoc/parser/test_markdown_parser.py
+++ b/test/unit_test/deepdoc/parser/test_markdown_parser.py
@@ -64,6 +64,18 @@ def markdown_parser_module(monkeypatch):
@pytest.mark.p2
class TestMarkdownElementExtractorFences:
+ @pytest.mark.parametrize("marker", ["`", "~"])
+ @pytest.mark.parametrize("table", ["| Product | Price |\n| --- | --- |\n| Widget | 10 |\n", '<table border="1"><tr><td>Widget</td></tr></table>\n'])
+ def test_table_after_longer_closing_fence(self, markdown_parser_module, marker, table):
+ fenced = f"{marker * 3}text\nexample\n{marker * 4}\n"
+
+ remainder, tables = markdown_parser_module.RAGFlowMarkdownParser().extract_tables_and_remainder(fenced + "\n" + table)
+
+ assert len(tables) == 1
+ assert "Widget" in tables[0]
+ assert fenced in remainder
+ assert "Widget" not in remainder
+
def test_custom_delimiter_preserves_backtick_fence(self, markdown_element_extractor):
text = "# Title\n```python\nprint('a')\nprint('b')\n```\nAfter"