documents with space prior to xml preamble fail to compare on jruby

[coffeejunk]

I found this relatively obscure bug when using equivalent-xml with jruby, which I think is a bug in nokogiri:

when comparing to xml documents where one of them starts with a string before the xml preamble, nokogiri ignores the following elements in the document.

jruby-1.7.0 :001 > require 'nokogiri'
 => true 
jruby-1.7.0 :002 > doc1 = Nokogiri::XML(" <?xml version='1.0' encoding='utf-8' ?><first \>")
 => #<Nokogiri::XML::Document:0x7fa name="document"> 
jruby-1.7.0 :003 > doc2 = Nokogiri::XML("<?xml version='1.0' encoding='utf-8' ?><first \>")
 => #<Nokogiri::XML::Document:0x7fe name="document" children=[#<Nokogiri::XML::Element:0x7fc name="first">]>

this works perfectly fine in ruby 1.8, 1.9, rbx1.8 and rbx1.9:

1.9.3-p286 :001 > require 'nokogiri'
 => true 
1.9.3-p286 :002 > doc1 = Nokogiri::XML(" <?xml version='1.0' encoding='utf-8' ?><first \>")
 => #<Nokogiri::XML::Document:0x3fd36c8e30bc name="document" children=[#<Nokogiri::XML::ProcessingInstruction:0x3fd36c8e2b80 name="xml">, #<Nokogiri::XML::Element:0x3fd36c8e2964 name="first">]> 
1.9.3-p286 :003 > doc2 = Nokogiri::XML("<?xml version='1.0' encoding='utf-8' ?><first \>")
 => #<Nokogiri::XML::Document:0x3fd36c8d7f28 name="document" children=[#<Nokogiri::XML::Element:0x3fd36c8d776c name="first">]>

weird however is, that this works as originally expected with all ruby implementations when the first element is not the xml preamble:

jruby-1.7.0 :004 > doc1 = Nokogiri::XML("<bar><foo /></bar>")
 => #<Nokogiri::XML::Document:0x804 name="document" children=[#<Nokogiri::XML::Element:0x802 name="bar" children=[#<Nokogiri::XML::Element:0x800 name="foo">]>]> 
jruby-1.7.0 :005 > doc2 = Nokogiri::XML(" <bar><foo /></bar>")
 => #<Nokogiri::XML::Document:0x80a name="document" children=[#<Nokogiri::XML::Element:0x808 name="bar" children=[#<Nokogiri::XML::Element:0x806 name="foo">]>]> 

1.9.3-p286 :004 > doc1 = Nokogiri::XML("<bar><foo /></bar>")
 => #<Nokogiri::XML::Document:0x3fd36c8cef90 name="document" children=[#<Nokogiri::XML::Element:0x3fd36c8ceb30 name="bar" children=[#<Nokogiri::XML::Element:0x3fd36c8ce8d8 name="foo">]>]> 
1.9.3-p286 :005 > doc2 = Nokogiri::XML(" <bar><foo /></bar>")
 => #<Nokogiri::XML::Document:0x3fd36c8c8848 name="document" children=[#<Nokogiri::XML::Element:0x3fd36c8c8280 name="bar" children=[#<Nokogiri::XML::Element:0x3fd36c8c7b50 name="foo">]>]>

Although I assume that this is a bug in nokogiri, I published two test cases, to my fork of the repository, showing off the odd behaviour because I think this should be a know issue.

  # this passes on jruby
  it "should ignore leading whitespace #1" do
    doc1 = Nokogiri::XML("<bar><foo /></bar>")
    doc2 = Nokogiri::XML(" <bar><foo /></bar>")
    doc1.should be_equivalent_to(doc2)
  end

  # this fails on jruby
  it "should ignore leading whitespace #2" do
    doc1 = Nokogiri::XML("<?xml version='1.0' encoding='utf-8' ?><foo />")
    doc2 = Nokogiri::XML(" <?xml version='1.0' encoding='utf-8' ?><foo />")
    doc1.should be_equivalent_to(doc2)
  end

Failures:

  1) EquivalentXml should ignore leading whitespace #2
     Failure/Error: doc1.should be_equivalent_to(doc2)
       expected:
       <?xml version="1.0"?>

       got:
       <?xml version="1.0"?>
       <foo/>
     # ./spec/equivalent-xml_spec.rb:123:in `(root)'

the complete output is visible on travis ci

[mbklein]

Thanks for reporting this, and for reporting it upstream to the Nokogiri folks.

[coffeejunk]

just as an FYI this bug has been fixed with nokogiri 1.5.6 (https://travis-ci.org/coffeejunk/equivalent-xml/jobs/3773113)