Fix sphinx-doc#9330: versionchanged causes error during pdf build

The versionchanged directive breaks doctree if its contents start with
non-paragraph element (ex. lists, tables, and so on).  It causes build
error in LaTeX.

This inserts a paragraph if the first child of the contents is not a
paragraph not to break doctree.

CHANGES

@@ -71,6 +71,8 @@ Bugs fixed
   :confval:`man_make_section_directory` is not correct
 * #9306: Linkcheck reports broken link when remote server closes the connection
   on HEAD request
+* #9330: changeset domain: :rst:dir:`versionchanged` with contents being a list
+  will cause error during pdf build
 * #9280: py domain: "exceptions" module is not displayed
 * #9224: ``:param:`` and ``:type:`` fields does not support a type containing
   whitespace (ex. ``Dict[str, str]``)

sphinx/domains/changeset.py

@@ -74,8 +74,10 @@ def run(self) -> List[Node]:
         if self.content:
             self.state.nested_parse(self.content, self.content_offset, node)
         classes = ['versionmodified', versionlabel_classes[self.name]]
-        if len(node):
-            if isinstance(node[0], nodes.paragraph) and node[0].rawsource:
+        if len(node) > 0 and isinstance(node[0], nodes.paragraph):
+            # the content starts with a paragraph
+            if node[0].rawsource:
+                # make the first paragraph translatable
                 content = nodes.inline(node[0].rawsource, translatable=True)
                 content.source = node[0].source
                 content.line = node[0].line
@@ -84,10 +86,16 @@ def run(self) -> List[Node]:
 
             para = cast(nodes.paragraph, node[0])
             para.insert(0, nodes.inline('', '%s: ' % text, classes=classes))
+        elif len(node) > 0:
+            # the content does not starts with a paragraph
+            para = nodes.paragraph('', '',
+                                   nodes.inline('', '%s: ' % text, classes=classes),
+                                   translatable=False)
+            node.insert(0, para)
         else:
+            # the content is empty
             para = nodes.paragraph('', '',
-                                   nodes.inline('', '%s.' % text,
-                                                classes=classes),
+                                   nodes.inline('', '%s.' % text, classes=classes),
                                    translatable=False)
             node.append(para)