fix: fall back to plain text when RSS item content triggers RecursionError (#2333)

* fix: fall back to plain text when RSS item content triggers RecursionError

Deeply nested HTML inside an RSS item's description or content:encoded
previously hit the broad except in _parse_content and returned the raw
unconverted HTML, silently embedding HTML tags in the markdown output.

Catch RecursionError specifically and fall back to BeautifulSoup's
iterative get_text(), mirroring the HTML converter fix from #1644.
Includes a deterministic regression test.

* Honor strict=true, similar to HTML

Author: shoemoney

packages/markitdown/src/markitdown/converters/_rss_converter.py

@@ -1,3 +1,5 @@
+import warnings
+
 from defusedxml import minidom
 from xml.dom.minidom import Document, Element
 from typing import BinaryIO, Any, Union
@@ -87,18 +89,23 @@ def convert(
         stream_info: StreamInfo,
         **kwargs: Any,  # Options to pass to the converter
     ) -> DocumentConverterResult:
+        # Pop our own keyword before forwarding the rest to markdownify.
+        # strict=True raises RecursionError instead of falling back to plain text.
+        strict: bool = kwargs.pop("strict", False)
         self._kwargs = kwargs
         doc = minidom.parse(file_stream)
         feed_type = self._feed_type(doc)
 
         if feed_type == "rss":
-            return self._parse_rss_type(doc)
+            return self._parse_rss_type(doc, strict=strict)
         elif feed_type == "atom":
-            return self._parse_atom_type(doc)
+            return self._parse_atom_type(doc, strict=strict)
         else:
             raise ValueError("Unknown feed type")
 
-    def _parse_atom_type(self, doc: Document) -> DocumentConverterResult:
+    def _parse_atom_type(
+        self, doc: Document, *, strict: bool = False
+    ) -> DocumentConverterResult:
         """Parse the type of an Atom feed.
 
         Returns None if the feed type is not recognized or something goes wrong.
@@ -121,16 +128,18 @@ def _parse_atom_type(self, doc: Document) -> DocumentConverterResult:
             if entry_updated:
                 md_text += f"Updated on: {entry_updated}\n"
             if entry_summary:
-                md_text += self._parse_content(entry_summary)
+                md_text += self._parse_content(entry_summary, strict=strict)
             if entry_content:
-                md_text += self._parse_content(entry_content)
+                md_text += self._parse_content(entry_content, strict=strict)
 
         return DocumentConverterResult(
             markdown=md_text,
             title=title,
         )
 
-    def _parse_rss_type(self, doc: Document) -> DocumentConverterResult:
+    def _parse_rss_type(
+        self, doc: Document, *, strict: bool = False
+    ) -> DocumentConverterResult:
         """Parse the type of an RSS feed.
 
         Returns None if the feed type is not recognized or something goes wrong.
@@ -158,21 +167,34 @@ def _parse_rss_type(self, doc: Document) -> DocumentConverterResult:
             if pubDate:
                 md_text += f"Published on: {pubDate}\n"
             if description:
-                md_text += self._parse_content(description)
+                md_text += self._parse_content(description, strict=strict)
             if content:
-                md_text += self._parse_content(content)
+                md_text += self._parse_content(content, strict=strict)
 
         return DocumentConverterResult(
             markdown=md_text,
             title=channel_title,
         )
 
-    def _parse_content(self, content: str) -> str:
+    def _parse_content(self, content: str, *, strict: bool = False) -> str:
         """Parse the content of an RSS feed item"""
         try:
             # using bs4 because many RSS feeds have HTML-styled content
             soup = BeautifulSoup(content, "html.parser")
             return _CustomMarkdownify(**self._kwargs).convert_soup(soup)
+        except RecursionError:
+            if strict:
+                raise
+            # Deeply nested item content can exceed Python's recursion limit
+            # during markdownify's recursive DOM traversal.  Fall back to
+            # BeautifulSoup's iterative get_text() so the caller still gets
+            # usable plain-text content instead of raw HTML.
+            warnings.warn(
+                "RSS item content is too deeply nested for markdown conversion "
+                "(RecursionError). Falling back to plain-text extraction.",
+                stacklevel=2,
+            )
+            return BeautifulSoup(content, "html.parser").get_text("\n", strip=True)
         except BaseException as _:
             return content
 

packages/markitdown/tests/test_module_misc.py

@@ -9,6 +9,7 @@
 
 from markitdown._uri_utils import parse_data_uri, file_uri_to_path
 from markitdown._markitdown import _get_content_disposition_filename
+from markitdown.converters import RssConverter
 
 from markitdown import (
     MarkItDown,
@@ -434,6 +435,7 @@ def test_deeply_nested_html_fallback() -> None:
             # Should have emitted a warning about the fallback
             recursion_warnings = [x for x in w if "deeply nested" in str(x.message)]
             assert len(recursion_warnings) > 0
+
     finally:
         sys.setrecursionlimit(original_limit)
 
@@ -444,6 +446,96 @@ def test_deeply_nested_html_fallback() -> None:
     assert "<p>" not in result.markdown
 
 
+def test_deeply_nested_rss_item_fallback() -> None:
+    """Deeply nested HTML inside an RSS item should fall back to plain-text
+    extraction instead of silently embedding raw unconverted HTML in the
+    markdown output (same failure class as the HTML converter fix in #1644).
+
+    Note: This test uses sys.setrecursionlimit to guarantee a RecursionError
+    regardless of the host environment's default limit, making it deterministic
+    across different platforms and CI configurations.
+    """
+    import sys
+    import warnings
+
+    markitdown = MarkItDown()
+
+    # Use a small recursion limit so the test is environment-independent.
+    # We restore the original limit in a finally block to avoid side-effects.
+    original_limit = sys.getrecursionlimit()
+    low_limit = 200  # well below markdownify's traversal depth for depth=500
+
+    # Build an RSS item whose content is deeply nested HTML
+    depth = 500
+    item_html = ""
+    for _ in range(depth):
+        item_html += '<div style="margin-left:10px">'
+    item_html += "<p>Deep feed content with <b>bold text</b></p>"
+    for _ in range(depth):
+        item_html += "</div>"
+
+    rss = (
+        '<?xml version="1.0" encoding="UTF-8"?>'
+        '<rss version="2.0" '
+        'xmlns:content="http://purl.org/rss/1.0/modules/content/">'
+        "<channel>"
+        "<title>Test Feed</title>"
+        "<description>A test feed</description>"
+        "<item>"
+        "<title>Deep Item</title>"
+        f"<content:encoded><![CDATA[{item_html}]]></content:encoded>"
+        "</item>"
+        "</channel>"
+        "</rss>"
+    )
+    atom = (
+        '<?xml version="1.0" encoding="UTF-8"?>'
+        '<feed xmlns="http://www.w3.org/2005/Atom">'
+        "<title>Test Feed</title>"
+        "<entry>"
+        "<title>Deep Entry</title>"
+        f'<content type="html"><![CDATA[{item_html}]]></content>'
+        "</entry>"
+        "</feed>"
+    )
+
+    try:
+        sys.setrecursionlimit(low_limit)
+        with warnings.catch_warnings(record=True) as w:
+            warnings.simplefilter("always")
+            result = markitdown.convert_stream(
+                io.BytesIO(rss.encode("utf-8")),
+                file_extension=".rss",
+            )
+
+            # Should have emitted a warning about the fallback
+            recursion_warnings = [x for x in w if "deeply nested" in str(x.message)]
+            assert len(recursion_warnings) > 0
+
+        # strict=True should expose the conversion failure rather than applying
+        # the plain-text fallback.
+        with pytest.raises(RecursionError):
+            RssConverter().convert(
+                io.BytesIO(rss.encode("utf-8")),
+                StreamInfo(extension=".rss"),
+                strict=True,
+            )
+        with pytest.raises(RecursionError):
+            RssConverter().convert(
+                io.BytesIO(atom.encode("utf-8")),
+                StreamInfo(extension=".atom"),
+                strict=True,
+            )
+    finally:
+        sys.setrecursionlimit(original_limit)
+
+    # The output should contain the text content, not raw HTML
+    assert "Deep feed content" in result.markdown
+    assert "bold text" in result.markdown
+    assert "<div" not in result.markdown
+    assert "<p>" not in result.markdown
+
+
 def test_doc_rlink() -> None:
     # Test for: CVE-2025-11849
     markitdown = MarkItDown()
@@ -644,7 +736,7 @@ def test_json_with_late_non_ascii_character(tmp_path) -> None:
             "title": "Example record",
             "abstract": "This is sample test. " * 500,
         },
-        "notes": "non-ASCII character: è",
+        "notes": "non-ASCII character: è",
     }
     json_path = tmp_path / "input.json"
     json_path.write_text(
@@ -653,7 +745,7 @@ def test_json_with_late_non_ascii_character(tmp_path) -> None:
 
     result = MarkItDown().convert(str(json_path))
 
-    assert "non-ASCII character: è" in result.text_content
+    assert "non-ASCII character: è" in result.text_content
 
 
 if __name__ == "__main__":