MRI Ruby vs. JRuby XML child namespace output differences

[johnnyshields]

The Problem

JRuby includes redundant xmlns:* attributes on child notes, even when the parent node already has the namespace. MRI correctly omits redundant xmlns:* attributes on child nodes.

Who cares?

I'm working to migrate RubySaml to purely use Nokogiri. While technically this is not a major functional issue, it is making my life difficult with test assertions about the structure of the XML.

Show me

Here's a simple script which illustrates the problem:

require 'nokogiri'

Nokogiri::XML::Builder.new do |xml|
  xml['ds'].Signature('xmlns:ds' => 'http://www.w3.org/2000/09/xmldsig#') do
    xml['ds'].SignedInfo do
    end
  end
end.doc.to_xml(save_with: Nokogiri::XML::Node::SaveOptions::AS_XML)

# MRI
# "<?xml version=\"1.0\"?>\n<ds:Signature xmlns:ds=\"http://www.w3.org/2000/09/xmldsig#\"><ds:SignedInfo/></ds:Signature>\n"

# JRuby
# "<?xml version=\"1.0\"?>\n<ds:Signature xmlns:ds=\"http://www.w3.org/2000/09/xmldsig#\"><ds:SignedInfo xmlns:ds=\"http://www.w3.org/2000/09/xmldsig#\"/></ds:Signature>"

[flavorjones]

Hi @johnnyshields. I personally have no plans to fix bugs or add features to the JRuby implementation at this point, though I am willing to review and merge pull requests from contributors.

Can you help me understand better why this is an issue for you? Naively, I would think that you could simply avoid making assertions on the attributes that the JRuby backend is unnecessarily adding. If you can, please show an entire test case that demonstrates why it's difficult?

[flavorjones]

For context, there are a lot of differences between the JRuby and CRuby implementations; but specifically namespaces are one of those things that are different, but not wrong.

See #1875 for a similar case where JRuby serializes differently than CRuby, but the document generated is semantically correct. In a perfect world, the libraries would behave the same, and so I'm sorry if this is making things difficult for you as a user of Nokogiri.

We've historically declined to try to make the two backends serialize byte-for-byte identically, and this is reflected in the https://nokogiri.org/#guiding-principles of the project:

be a thin-as-reasonable layer on top of the underlying parsers, and don't attempt to fix behavioral differences between the parsers

In this case, since the serialized document is not incorrect, I'm going to decline to try to change the behavior. But if you show me your test case I can try to help design a robust way to test the behavior that you care about!

[johnnyshields]

@flavorjones thanks for taking a look. The example in the original post shows the issue; on JRuby, Nokogiri re-adds an explicit xmlns attribute on every child node, even though it is already present on a parent node.

Can you help me understand better why this is an issue for you?

At a minimum, this is behavior is redundant and unnecessary. Technically speaking it is "not incorrect" in terms of "it theoretically shouldn't affect a third-party consumer's ability to parse the XML", however, given that my particular library (RubySaml) has 1000s of test cases which use Regexps to make assertions about the structure of the XML, it's quite a pain when Nokogiri on MRI and REXML match, but Nokogiri on JRuby does something significantly different.

please show an entire test case that demonstrates why it's difficult?

Here's an example of a Minitest test case that passes on MRI and fails on JRuby.

it "adds the signature value" do
  xml_doc = Nokogiri::XML::Builder.new do |xml|
    xml['ds'].Signature('xmlns:ds' => 'http://www.w3.org/2000/09/xmldsig#') do
      xml['ds'].SignatureValue("foobar")
    end
  end.doc.to_xml(save_with: Nokogiri::XML::Node::SaveOptions::AS_XML)

  assert_match(%r{<ds:SignatureValue>([a-zA-Z0-9/+]+=?=?)</ds:SignatureValue>}, xml_doc)
end

don't attempt to fix behavioral differences between the parsers

Strictly speaking this appears to be a "renderer" (output) issue rather than a "parser" (ingestion) issue.

If the fix isn't obvious I'm happy to dig into the JRuby code when bandwidth allows.

[flavorjones]

on JRuby, Nokogiri re-adds an explicit xmlns attribute on every child node

Yup, I understand.

Can you help me understand better why this is an issue for you?

RubySaml has 1000s of test cases which use Regexps to make assertions

Got it.

I'm happy to dig into the JRuby code

Thanks for your PR, let's continue the conversation there.