ruby/test/rss/test_taxonomy.rb

173 строки
4.9 KiB
Ruby

require "cgi"
require "rss-testcase"
require "rss/1.0"
require "rss/2.0"
require "rss/taxonomy"
module RSS
class TestTaxonomy < TestCase
def setup
@prefix = "taxo"
@uri = "http://purl.org/rss/1.0/modules/taxonomy/"
@dc_prefix = "dc"
@dc_uri = "http://purl.org/dc/elements/1.1/"
@ns = {
@prefix => @uri,
@dc_prefix => @dc_uri,
}
@topics_parents = %w(channel item)
@topics_lis = [
"http://meerkat.oreillynet.com/?c=cat23",
"http://meerkat.oreillynet.com/?c=47",
"http://dmoz.org/Computers/Data_Formats/Markup_Languages/XML/",
]
@topics_node = "<#{@prefix}:topics>\n"
@topics_node << " <rdf:Bag>\n"
@topics_lis.each do |value|
resource = CGI.escapeHTML(value)
@topics_node << " <rdf:li resource=\"#{resource}\"/>\n"
end
@topics_node << " </rdf:Bag>\n"
@topics_node << "</#{@prefix}:topics>"
@topic_topics_lis = \
[
"http://meerkat.oreillynet.com/?c=cat23",
"http://dmoz.org/Computers/Data_Formats/Markup_Languages/SGML/",
"http://dmoz.org/Computers/Programming/Internet/",
]
@topic_contents = \
[
{
:link => "http://meerkat.oreillynet.com/?c=cat23",
:title => "Data: XML",
:description => "A Meerkat channel",
},
{
:link => "http://dmoz.org/Computers/Data_Formats/Markup_Languages/XML/",
:title => "XML",
:subject => "XML",
:description => "DMOZ category",
:topics => @topic_topics_lis,
}
]
@topic_nodes = @topic_contents.collect do |info|
link = info[:link]
rv = "<#{@prefix}:topic rdf:about=\"#{link}\">\n"
info.each do |name, value|
case name
when :topics
rv << " <#{@prefix}:topics>\n"
rv << " <rdf:Bag>\n"
value.each do |li|
resource = CGI.escapeHTML(li)
rv << " <rdf:li resource=\"#{resource}\"/>\n"
end
rv << " </rdf:Bag>\n"
rv << " </#{@prefix}:topics>\n"
else
prefix = (name == :link ? @prefix : @dc_prefix)
rv << " <#{prefix}:#{name}>#{value}</#{prefix}:#{name}>\n"
end
end
rv << "</#{@prefix}:topic>"
end
@rss_source = make_RDF(<<-EOR, @ns)
#{make_channel(@topics_node)}
#{make_image()}
#{make_item(@topics_node)}
#{make_textinput()}
#{@topic_nodes.join("\n")}
EOR
@rss = Parser.parse(@rss_source)
end
def test_parser
assert_nothing_raised do
Parser.parse(@rss_source)
end
assert_too_much_tag("topics", "channel") do
Parser.parse(make_RDF(<<-EOR, @ns))
#{make_channel(@topics_node * 2)}
#{make_item()}
EOR
end
assert_too_much_tag("topics", "item") do
Parser.parse(make_RDF(<<-EOR, @ns))
#{make_channel()}
#{make_item(@topics_node * 2)}
EOR
end
end
def test_accessor
topics = @rss.channel.taxo_topics
assert_equal(@topics_lis.sort,
topics.Bag.lis.collect {|li| li.resource}.sort)
assert_equal(@topics_lis.sort, topics.resources.sort)
assert_equal(@rss.taxo_topics.first, @rss.taxo_topic)
@topic_contents.each_with_index do |info, i|
topic = @rss.taxo_topics[i]
info.each do |name, value|
case name
when :link
assert_equal(value, topic.about)
assert_equal(value, topic.taxo_link)
when :topics
assert_equal(value.sort, topic.taxo_topics.resources.sort)
else
assert_equal(value, topic.__send__("dc_#{name}"))
end
end
end
end
def test_to_s
@topics_parents.each do |parent|
meth = "taxo_topics_element"
assert_equal(@topics_node, @rss.__send__(parent).__send__(meth))
end
@topic_nodes.each_with_index do |node, i|
expected_xml = taxo_xmlns_container(node)
expected = REXML::Document.new(expected_xml).root.elements[1]
actual_xml = taxo_xmlns_container(@rss.taxo_topics[i].to_s(true, ""))
actual = REXML::Document.new(actual_xml).root.elements[1]
expected_elems = expected.reject {|x| x.is_a?(REXML::Text)}
actual_elems = actual.reject {|x| x.is_a?(REXML::Text)}
expected_elems.sort! {|x, y| x.name <=> y.name}
actual_elems.sort! {|x, y| x.name <=> y.name}
assert_equal(expected_elems.collect {|x| x.to_s},
actual_elems.collect {|x| x.to_s})
assert_equal(expected.attributes.sort, actual.attributes.sort)
end
end
private
def taxo_xmlns_container(content)
xmlns_container({
@prefix => @uri,
"dc" => "http://purl.org/dc/elements/1.1/",
"rdf" => "http://www.w3.org/1999/02/22-rdf-syntax-ns#",
},
content)
end
end
end