docs: automarkup.py: Fix invalid HTML link output and broken URI fragments

Since commit d18b017 ("docs: Add automatic cross-reference for
documentation pages"), references that were already explicitly defined
with "ref:" and referred to other pages with a path have been doubled.
This is reported as the following error by Firefox:

  Start tag "a" seen but an element of the same type was already open.
  End tag "a" violates nesting rules.

As well as the invalid HTML, this also obscures the URI fragment links
to subsections because the second link overrides the first. For example
on the page admin-guide/hw-vuln/mds.html the last link should be to the
"Default Mitigations" subsection using a # URI fragment:

  admin-guide/hw-vuln/l1tf.html#default-mitigations

But it is obsured by a second link to the whole page:

  admin-guide/hw-vuln/l1tf.html

The full HTML with the double <a> tags looks like this:

  <a class="reference internal" href="l1tf.html#default-mitigations">
    <span class="std std-ref">
      <a class="reference internal" href="l1tf.html">
        <span class="doc">L1TF - L1 Terminal Fault</span>
      </a>
    </span>
  </a>

After this commit, there is only a single link:

  <a class="reference internal" href="l1tf.html#default-mitigations">
    <span class="std std-ref">Documentation/admin-guide/hw-vuln//l1tf.rst</span>
  </a>

Now that the second link is removed, the browser correctly jumps to the
default-mitigations subsection when clicking the link.

The fix is to check that nodes in the document to be modified are not
already references. A reference is counted as any text that is a
descendant of a reference type node. Only plain text should be converted
to new references, otherwise the doubling occurs.

Testing
=======

 * Test that the build stdout is the same (ignoring ordering), and that
   no new warnings are printed.

 * Diff all .html files and check that the only modifications occur
   to the bad double links.

 * The auto linking of bare references to pages without "ref:" is still
   working.

Fixes: d18b017 ("docs: Add automatic cross-reference for documentation pages")
Reviewed-by: Nícolas F. R. A. Prado <[email protected]>
Signed-off-by: James Clark <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jonathan Corbet <[email protected]>

Author: James-A-Clark

Documentation/sphinx/automarkup.py

@@ -271,19 +271,30 @@ def get_c_namespace(app, docname):
 def auto_markup(app, doctree, name):
     global c_namespace
     c_namespace = get_c_namespace(app, name)
+    def text_but_not_a_reference(node):
+        # The nodes.literal test catches ``literal text``, its purpose is to
+        # avoid adding cross-references to functions that have been explicitly
+        # marked with cc:func:.
+        if not isinstance(node, nodes.Text) or isinstance(node.parent, nodes.literal):
+            return False
+
+        child_of_reference = False
+        parent = node.parent
+        while parent:
+            if isinstance(parent, nodes.Referential):
+                child_of_reference = True
+                break
+            parent = parent.parent
+        return not child_of_reference
+
     #
     # This loop could eventually be improved on.  Someday maybe we
     # want a proper tree traversal with a lot of awareness of which
     # kinds of nodes to prune.  But this works well for now.
     #
-    # The nodes.literal test catches ``literal text``, its purpose is to
-    # avoid adding cross-references to functions that have been explicitly
-    # marked with cc:func:.
-    #
     for para in doctree.traverse(nodes.paragraph):
-        for node in para.traverse(nodes.Text):
-            if not isinstance(node.parent, nodes.literal):
-                node.parent.replace(node, markup_refs(name, app, node))
+        for node in para.traverse(condition=text_but_not_a_reference):
+            node.parent.replace(node, markup_refs(name, app, node))
 
 def setup(app):
     app.connect('doctree-resolved', auto_markup)