nokobug.rb

require 'nokogiri'

good_html = "<div> <span>first</span>        <span>second</span> </div>"
bad_html  = "<div> <span><a>first</a></span> <span>second</span> </div>"

# This worked under libxml2 2.8.0 (default in Ubuntu 12.10), but doesn't work anymore with libxml2 2.9.0 (under Ubuntu 13.04).
puts Nokogiri::HTML(good_html).search('div span:nth-child(2)').first # => <span>second</span>
puts Nokogiri::HTML(bad_html).search('div span:nth-child(2)').first # => nil

# I found two workarounds. One is to use ">" for more precise selection:
puts Nokogiri::HTML(good_html).search('div > span:nth-child(2)').first # => <span>second</span>
puts Nokogiri::HTML(bad_html).search('div > span:nth-child(2)').first # => <span>second</span>

# Another workaround is to use nth-of-type instead of nth-child:
puts Nokogiri::HTML(good_html).search('div span:nth-of-type(2)').first # => <span>second</span>
puts Nokogiri::HTML(bad_html).search('div span:nth-of-type(2)').first # => <span>second</span>