RSS/Atom converter silently truncates content to the first child node, dropping CDATA sections #2461

The Bug

RssConverter._get_data_by_tag_name (packages/markitdown/src/markitdown/converters/_rss_converter.py, lines 308-321), the shared helper used to extract title, description, pubDate, content:encoded (RSS) and title, subtitle, updated, summary/content (Atom), only reads the first child node of a matched element instead of concatenating all of its text-bearing children:

def _get_data_by_tag_name(self, element, tag_name):
    """Get data from first child element with the given tag name.
    Returns None when no such element is found.
    """
    nodes = element.getElementsByTagName(tag_name)
    if not nodes:
        return None
    fc = nodes[0].firstChild
    if fc:
        if hasattr(fc, "data"):
            return fc.data
    return None

In xml.dom.minidom, an element's children are not automatically merged. Any element that mixes plain text with a <![CDATA[...]]> section, or just has whitespace indentation before a CDATA block (extremely common in pretty-printed feeds), produces multiple sibling child nodes. Only the first one is ever read - everything after it is silently dropped, with no error or warning, and the leftover whitespace/text is still truthy so calling code has no signal anything went wrong.

Steps to Reproduce

Case 1, whitespace-indented CDATA (the more dangerous variant - the result is truthy and looks superficially fine):

from xml.dom import minidom

xml_bytes = b"""<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0">
  <channel>
    <title>
      <![CDATA[My Real Channel Title]]>
    </title>
    <description>Feed description</description>
  </channel>
</rss>
"""
doc = minidom.parseString(xml_bytes)

def get_data_by_tag_name(element, tag_name):
    nodes = element.getElementsByTagName(tag_name)
    if not nodes:
        return None
    fc = nodes[0].firstChild
    if fc and hasattr(fc, "data"):
        return fc.data
    return None

channel = doc.getElementsByTagName("channel")[0]
print(repr(get_data_by_tag_name(channel, "title")))

Actual: '\n ' - the real title "My Real Channel Title" (held in the CDATASection sibling) is completely lost. _parse_rss_type's if channel_title: md_text += f"# {channel_title}\n" still fires (the whitespace string is truthy) and emits a heading containing only whitespace.

Expected: "My Real Channel Title".

Case 2, mixed text/CDATA content:

<description>Intro text before html. <![CDATA[<p>Rich HTML fragment that matters.</p>]]> Trailing text after html too.</description>

Actual: 'Intro text before html. ' - both the CDATA HTML fragment and the trailing sentence are dropped.
Expected: all three pieces concatenated.

Fix direction

Concatenate every text-bearing child instead of returning only firstChild.data, e.g.:

from xml.dom.minidom import Node

def _get_data_by_tag_name(self, element, tag_name):
    nodes = element.getElementsByTagName(tag_name)
    if not nodes:
        return None
    parts = [
        child.data
        for child in nodes[0].childNodes
        if child.nodeType in (Node.TEXT_NODE, Node.CDATA_SECTION_NODE)
    ]
    return "".join(parts).strip() if parts else None

Duplicate check

Searched RSS+CDATA, firstChild, RSS+truncat, getElementsByTagName, nodeValue, silently+truncat (all 0 results), plus broader RSS+description (12 hits) and rss+converter (4 hits) whose bodies were read in full - all about different mechanisms: double-HTML-parsing losing literal angle brackets (#2373-2375), Atom XHTML preservation (#2297), a RecursionError fallback (#2333, already implemented), an UnboundLocalError on missing title (#1946, #1989, #2164), RSS metadata leakage plus an unrelated pptx fix (#1785), and two multi-bug grab-bag PRs (#2329, #1723). None touch firstChild, CDATA, or multi-child-node text extraction.