Prefixed element inconsistency between CRuby and JRuby

[benlangfeld]

Affected ruby versions

• JRuby
• Ruby 1.9.3 (probably more)

Affected nokogiri versions

• 1.5.x (including 1.5.5)

Test case

require 'nokogiri'

Foo = Class.new Nokogiri::XML::Node

shared_examples_for 'prefixed attributes' do
  let(:doc) { Nokogiri::XML::Document.new }

  subject { described_class.new 'foo', doc }

  it "should write attributes with prefixes and read them back" do
    subject[:'xml:lang'] = 'en-GB'
    subject[:'xml:lang'].should be == 'en-GB'
    subject[:'lang'].should be nil
  end
end

describe Nokogiri::XML::Node do
  it_should_behave_like 'prefixed attributes'
end

describe Foo do
  it_should_behave_like 'prefixed attributes'
end

Result

{11:11}[ruby-1.9.3]~/Desktop/nokogiri ben% be rspec spec/nokogiri_spec.rb
FF

Failures:

  1) Nokogiri::XML::Node it should behave like prefixed attributes should write attributes with prefixes and read them back
     Failure/Error: subject[:'xml:lang'].should be == 'en-GB'
       expected: == "en-GB"
            got:    nil
     Shared Example Group: "prefixed attributes" called from ./spec/nokogiri_spec.rb:18
     # ./spec/nokogiri_spec.rb:12:in `block (2 levels) in <top (required)>'

  2) Foo it should behave like prefixed attributes should write attributes with prefixes and read them back
     Failure/Error: subject[:'xml:lang'].should be == 'en-GB'
       expected: == "en-GB"
            got:    nil
     Shared Example Group: "prefixed attributes" called from ./spec/nokogiri_spec.rb:22
     # ./spec/nokogiri_spec.rb:12:in `block (2 levels) in <top (required)>'

Finished in 0.00519 seconds
2 examples, 2 failures

Failed examples:

rspec ./spec/nokogiri_spec.rb:10 # Nokogiri::XML::Node it should behave like prefixed attributes should write attributes with prefixes and read them back
rspec ./spec/nokogiri_spec.rb:10 # Foo it should behave like prefixed attributes should write attributes with prefixes and read them back

{11:11}[jruby-1.6.7.2]~/Desktop/nokogiri ben% be rspec spec/nokogiri_spec.rb
..

Finished in 0.241 seconds
2 examples, 0 failures
bundle exec rspec spec/nokogiri_spec.rb  26.21s user 1.06s system 258% cpu 10.569 total

I believe JRuby is behaving correctly here, and the C version is the one at fault.

[benlangfeld]

I'd like to garner some sort of consensus on the approach here before I attempt to make a fix. @flavorjones, @tenderlove: Would you agree that the behavior of JRuby is the correct one here?

[benlangfeld]

I note that in this case, Ox follows the same semantics as Nokogiri on JRuby.

[benlangfeld]

Things that behave like #['lang']:

• libxml-ruby

Things that behave like #['xml:lang']:

• Ox
• Hpricot
• XPath

In my mind, that brings it hands down in favour of qualifying namespace prefixes in attribute accessors. I shall prepare a patch.

[benlangfeld]

Patch at #726 ready to be merged. Can someone take a look?