[bug] CRuby namespace shadowing bug

[johnnyshields]

The following test case illustrates a bug related to xmlns shadowing on MRI that doesn't happen JRuby. This bug changes the semantic meaning of the output XML.

  it 'redeclares xmlns definitions when shadowed' do
    doc = Nokogiri::XML::Builder.new do |xml|
      xml['dnd'].adventure('xmlns:dnd' => 'http://www.w3.org/dungeons#') do
        xml['dnd'].party('xmlns:dnd' => 'http://www.w3.org/dragons#') do
          xml['dnd'].members('xmlns:dnd' => 'http://www.w3.org/dragons#') do
            xml['dnd'].character('xmlns:dnd' => 'http://www.w3.org/dungeons#') do
              xml['dnd'].name('Nigel', 'xmlns:dnd' => 'http://www.w3.org/dungeons#')
            end
          end
        end
      end
    end.doc.to_xml(save_with: Nokogiri::XML::Node::SaveOptions::AS_XML)

    assert_includes(doc, '<dnd:adventure xmlns:dnd="http://www.w3.org/dungeons#">')
    assert_includes(doc, '<dnd:party xmlns:dnd="http://www.w3.org/dragons#">')
    assert_includes(doc, '<dnd:members>')
    # TODO: The MRI (non-JRuby) behavior here is incorrect.
    # See: https://github.com/sparklemotion/nokogiri/issues/3458
    if Nokogiri.jruby?
      assert_includes(doc, '<dnd:character xmlns:dnd="http://www.w3.org/dungeons#">')
    else
      assert_includes(doc, '<dnd:character>')
    end
    assert_includes(doc, '<dnd:name>Nigel</dnd:name>')
  end

[flavorjones]

Simplifying this test as much as I can:

  it "foo" do
    doc = Nokogiri::XML(<<~XML)
      <dnd:adventure xmlns:dnd="http://www.w3.org/dungeons#">
        <dnd:party xmlns:dnd="http://www.w3.org/dragons#">
          <dnd:members>
          </dnd:members>
        </dnd:party>
      </dnd:adventure>      
    XML

    dungeons_ns = doc.root.namespace_definitions.find { |ns| ns.prefix == "dnd" }
    parent = doc.xpath("//ns:members", ns: "http://www.w3.org/dragons#").first

    node = doc.create_element("character")
    node.namespace = dungeons_ns
    parent.add_child(node)

    # passes, indicating it's not a reparenting issues
    assert_equal("http://www.w3.org/dungeons#", node.namespace.href)
    # fails, maybe this is a serialization issue? not sure.
    assert_includes(doc.to_xml, %Q{<dnd:character xmlns:dnd="http://www.w3.org/dungeons#"/>})
  end

This seems to indicate that the tree is constructed properly, but it serializes incorrectly. I'm not sure, though, since when Nokogiri parses an input string containing all these elements, everything is fine. Some combination of reparenting and serialization, maybe.

[flavorjones]

Yeah, ok, this is a bug in the builder. Polishing a fix now.

No, this should actually be fixed in the node reparenting function. Working on it.

[flavorjones]

OK, just to clarify ...

1. the bug in the original test is in Builder, and I can fix it there.
2. my attempt to simplify the test demonstrates a second bug, which I'll open a separate issue for

[flavorjones]

OK, I've got a fix for Builder which simplifies and makes the behavior more correct; but am now running into further bugs in the Java implementation which I think are related to the namespace cache (which I want to delete because it's buggy, see #2543).

This is just one reason why I am not crazy about continuing to support the JRuby implementation. Sigh.

[flavorjones]

See #3459 for tracking the secondary bug I found above.