require_relative "rexml_test_utils" require "rexml/document" class XPathTester < Test::Unit::TestCase include REXMLTestUtils include REXML SOURCE = <<-EOF

EOF JENI_TENNISON = <<-EOJT EOJT def setup @@doc = Document.new(SOURCE) unless defined? @@doc @@jeni = Document.new( JENI_TENNISON ) unless defined? @@jeni end def each_test( element, xpath ) count = 0 XPath::each( element, xpath ) { |child| count += 1 yield child if block_given? } count end def test_descendant doc = Document.new("") p = XPath.match( doc, "//c" ) assert_equal( 2, p.size ) p = XPath.first( @@doc, "//p" ) assert_equal "p", p.name c = each_test( @@doc, "//c" ) { |child| assert_equal "c", child.name } assert_equal 5, c c = each_test( @@doc.root, "b//c" ) { |child| assert_equal "c", child.name } assert_equal 2, c doc = Document.new( "" ) # //para[1] : all descendants which are the first para child of their parent assert_equal( 4, XPath.match( doc, "//z[1]" ).size ) # /descendant::para[1] : the first descendant para element assert_equal( 1, XPath.match( doc, "/descendant::z[1]" ).size ) end def test_root source = "" doc = Document.new( source ) assert_equal doc, doc.root_node assert_equal "a", XPath::first( doc, "/" ).elements[1].name end def test_abbreviated_simple_child assert_equal "a", XPath::first(@@doc, "a").name end def test_child c = XPath::first( @@doc, "a/b/c" ) assert_equal "c", c.name assert_equal "3", XPath::first(@@doc, "a/b/c").attributes["id"] end def test_root_child assert_equal "a", XPath::first(@@doc, "/a").name c = XPath::first( @@doc, "a/b/c" ) assert_equal "a", XPath::first(c, "/a").name end def test_root_children c = XPath::first( @@doc, "a/b/c" ) assert_equal "2", XPath::first(c, "/a/b").attributes["id"] end def test_abbreviated_step c = XPath::first( @@doc, "a/b/c" ) assert_equal("c", c.name) assert_equal("a", XPath::first(@@doc.root, ".").name) assert_equal("b", XPath::first(c, "..").name) assert_equal("a", XPath::first(@@doc, "a/b/..").name) doc = REXML::Document.new(File.new(fixture_path("project.xml"))) c = each_test(doc.root, "./Description" ) { |child| assert_equal("Description",child.name) } assert_equal 1, c end # Things that aren't tested elsewhere def test_predicates assert_equal "12", XPath::first(@@doc, "a/e/f[3]").attributes["id"] assert_equal "13", XPath::first(@@doc, "a/e/f[3]/g").attributes["id"] assert_equal "14", XPath::first(@@doc, "a/e/f[@a='d'][2]").attributes["id"] assert_equal "14", XPath::first(@@doc, "a/e/f[@a='d'][@id='14']").attributes["id"] assert_equal "a", XPath::first( @@doc, "*[name()='a' and @id='1']" ).name c=each_test( @@doc, "//*[name()='f' and @a='d']") { |i| assert_equal "f", i.name } assert_equal 2, c c=each_test( @@doc, "//*[name()='m' or @a='d']") { |i| assert ["m","f"].include?(i.name) } assert_equal 3, c assert_equal "b", XPath::first( @@doc, "//b[@x]" ).name end def test_node_type doc = Document.new "text" #res = XPath::first(doc.root, "text()") #assert_equal "text", res.to_s #res = XPath::first(doc, "*") #assert_equal "a", res.name assert_equal( :processing_instruction, XPath::first(doc.root, "processing-instruction()").node_type) assert_equal( :comment, XPath::first(doc.root, "comment()").node_type) end def test_functions # trivial text() test # confuse-a-function source = "more dumb text" doc = Document.new source res = "" #XPath::each(doc.root, "text()") {|val| res << val.to_s} #assert_equal "more text", res #res = XPath::first(doc.root, "b[last()]") #assert_equal '3', res.attributes['id'] res = XPath::first(doc.root, "b[position()=2]") assert_equal '2', res.attributes['id'] res = XPath::first(doc.root, "*[name()='c']") assert_equal "c", res.name end def no_test_ancestor doc = REXML::Document.new(File.new(fixture_path("testsrc.xml"))) doc.elements.each("//item") { |el| print el.name if el.attributes['x'] puts " -- "+el.attributes['x'] else puts end } doc.elements.each("//item/ancestor::") { |el| print el.name if el.attributes['x'] puts " -- "+el.attributes['x'] else puts end } end # Here are some XPath tests that were originally submitted by ... # The code has changed some, but the logic and the source documents are the # same. # This method reads a document from a file, and then a series of xpaths, # also from a file. It then checks each xpath against the source file. def test_more xmlsource = fixture_path("testsrc.xml") xpathtests = fixture_path("xp.tst") doc = REXML::Document.new(File.new(xmlsource)) #results = "" results = REXML::Document.new results.add_element "test-results" for line in File.new(xpathtests) line.strip! begin doc.root #puts "#"*80 #print "\nDoing #{line} " ; $stdout.flush doc.elements.each(line) do |el| #print "." ; $stdout.flush results.root << el.clone #results << el.to_s end #ObjectSpace.garbage_collect GC::start rescue Exception => z #puts "\n'#{line}' failed" fail("Error on line #{line}:\n#{z.message}\n"+z.backtrace[0,10].join("\n")) #results.root.add_element( "error", {"path"=>line}).text = z.message+"\n"+z.backtrace[0,10].join("\n") #results << ""+z.message+"" end end end def test_axe_descendant assert_equal "f", XPath::first( @@doc, "descendant::f").name end def test_axe_parent q = XPath.first( @@doc, "a/d/c/parent::*/q" ) assert_equal 19, q.attributes["id"].to_i end def test_abbreviated_attribute assert_equal 'a', XPath::first( @@doc, "a[@id='1']" ).name c = XPath::first( @@doc, "a/b/c[@id='4']" ) assert_equal 'c', c.name assert_equal '4', c.attributes['id'] result = XPath::first( @@doc, "descendant::f[@a='c']") assert_equal "11", result.attributes['id'] assert_equal "11", XPath::first(@@doc, "a/e/f[@a='c']").attributes["id"] assert_equal "11", XPath::first(@@doc, "a/e/*[@a='c']").attributes["id"] end def test_axe_self c = XPath::first( @@doc, "a/b/c" ) assert c assert_equal "c", c.name assert_equal "c", XPath::first( c, "self::node()" ).name end def test_axe_ancestor doc = REXML::Document.new " " d = XPath.first( doc, "//d" ) assert_equal "d", d.name b = each_test( d, "ancestor::b" ) { |el| assert((1..2) === el.attributes['id'].to_i, "Expected #{el.attributes['id']} to be either 1 or 2" ) } assert_equal 2, b end def test_axe_child m = XPath.first( @@doc, "a/child::m" ) assert_equal 15, m.attributes['id'].to_i end def test_axe_attribute a = XPath.first( @@doc, "a/attribute::id" ) assert_equal "1", a.value a = XPath.first( @@doc, "a/e/f[@id='14']/attribute::a" ) assert_equal "d", a.value end def test_axe_sibling doc = Document.new "" first_f = XPath.first( doc, "a/e/f" ) assert first_f assert_equal '10', first_f.attributes['id'] next_f = XPath.first( doc, "a/e/f/following-sibling::node()" ) assert_equal '11', next_f.attributes['id'] b = XPath.first( doc, "a/e/preceding-sibling::node()" ) assert_equal 'b', b.name end def test_lang doc = Document.new(File.new(fixture_path("lang0.xml"))) #puts IO.read( "test/lang.xml" ) #puts XPath.match( doc, "//language/*" ).size c = each_test( doc, "//language/*" ) { |element| #puts "#{element.name}: #{element.text}" } assert_equal 4, c end def test_namespaces_1 source = <<-EOF this bar that bar EOF doc = Document.new source XPath.each( doc, "//bar" ) { fail "'bar' should match nothing in this case" } namespace = {"t"=>"this"} results = XPath.first( doc, "//t:bar", namespace ) assert_equal "this bar", results.text end def test_namespaces_2 source = <<-EOF this bar that bar EOF doc = Document.new source res = XPath::first(doc, "//*[local_name()='bar']") assert res, "looking for //*[name()='bar']" assert_equal 'this', res.namespace res = XPath::first(doc.root, "*[namespace_uri()='that']") assert_equal 'that bar', res.text end def test_complex next_f = XPath.first( @@doc, "a/e/f[@id='11']/following-sibling::*" ) assert_equal 12, next_f.attributes['id'].to_i prev_f = XPath.first( @@doc, "a/e/f[@id='11']/preceding-sibling::*" ) assert_equal 10, prev_f.attributes['id'].to_i c = each_test( @@doc, "descendant-or-self::*[@x='y']" ) assert_equal 2, c end def test_grouping t = XPath.first( @@doc, "a/d/*[name()='d' and (name()='f' or name()='q')]" ) assert_nil t t = XPath.first( @@doc, "a/d/*[(name()='d' and name()='f') or name()='q']" ) assert_equal 'q', t.name end def test_preceding d = Document.new "" start = XPath.first( d, "/a/b[@id='1']" ) assert_equal 'b', start.name c = XPath.first( start, "preceding::c" ) assert_equal '2', c.attributes['id'] c1, c0 = XPath.match( d, "/a/b/c[@id='2']/preceding::node()" ) assert_equal '1', c1.attributes['id'] assert_equal '0', c0.attributes['id'] c2, c1, c0, b, b2, b0 = XPath.match( start, "preceding::node()" ) assert_equal 'c', c2.name assert_equal 'c', c1.name assert_equal 'c', c0.name assert_equal 'b', b.name assert_equal 'b', b2.name assert_equal 'b', b0.name assert_equal '2', c2.attributes['id'] assert_equal '1', c1.attributes['id'] assert_equal '0', c0.attributes['id'] assert b.attributes.empty? assert_equal '2', b2.attributes['id'] assert_equal '0', b0.attributes['id'] d = REXML::Document.new("") matches = REXML::XPath.match(d, "/a/d/preceding::node()") assert_equal("c", matches[0].name) assert_equal("b", matches[1].name) s = "" d = REXML::Document.new(s) c = REXML::XPath.match( d, "//c[@id = '5']") cs = REXML::XPath.match( c, "preceding::c" ) assert_equal( 4, cs.length ) end def test_following d = Document.new "" start = XPath.first( d, "/a/b[@id='0']" ) assert_equal 'b', start.name c = XPath.first( start, "following::c" ) assert_equal '1', c.attributes['id'] s = "" d = Document.new(s) c = XPath.first(d, '/a/b/c') assert_equal 'c', c.name res = XPath.match( c, 'following::*' ) assert_equal 6, res.size res = XPath.match( c, 'following::i' ) assert_equal 2, res.size end # The following three paths were provided by # Jeni Tennison # a consultant who is also an XSL and XPath expert #def test_child_cubed # els = @@jeni.elements.to_a("*****") # assert_equal 3, els.size #end #def test_div_2 # results = doc.elements.to_a("/ div 2") #end #def test_nested_predicates # puts @@jeni.root.elements[1].elements[1].name # results = @@jeni.root.elements[1].elements[1].elements.to_a("../following-sibling::*[*[name() = name(current())]]") # puts results #end # Contributed by Mike Stok def test_starts_with source = <<-EOF a@b.c http://www.foo.com EOF doc = Document.new source mailtos = doc.elements.to_a("//a[starts-with(@href, 'mailto:')]") assert_equal 1, mailtos.size assert_equal "mailto:a@b.c", mailtos[0].attributes['href'] ailtos = doc.elements.to_a("//a[starts-with(@href, 'ailto:')]") assert_equal 0, ailtos.size end def test_toms_text_node file = "ABCDEF" doc = Document.new(file) assert_equal 'A', XPath.first(doc[0], 'text()').to_s assert_equal 'AF', XPath.match(doc[0], 'text()').collect { |n| n.to_s }.join('') assert_equal 'B', XPath.first(doc[0], 'b/text()').to_s assert_equal 'D', XPath.first(doc[0], '//d/text()').to_s assert_equal 'ABCDEF', XPath.match(doc[0], '//text()').collect {|n| n.to_s }.join('') end def test_string_length doc = Document.new <<-EOF EOF assert doc, "create doc" set = doc.elements.to_a("//*[string-length(name()) = 3]") assert_equal 2, set.size, "nodes with names length = 3" set = doc.elements.to_a("//*[string-length(name()) < 3]") assert_equal 2, set.size, "nodes with names length < 3" set = doc.elements.to_a("//*[string-length(name()) > 3]") assert_equal 3, set.size, "nodes with names length > 3" end # Test provided by Mike Stok def test_contains source = <<-EOF a@b.c http://www.foo.com EOF doc = Document.new source [ #['o', 2], ['foo', 1], ['bar', 0]].each { |search, expected| set = doc.elements.to_a("//a[contains(@href, '#{search}')]") assert_equal expected, set.size } end # Mike Stok and Sean Russell def test_substring # examples from http://www.w3.org/TR/xpath#function-substring doc = Document.new('') Document.new("") #puts XPath.first(d, 'node()[0 + 1]') #d = Document.new("") #puts XPath.first(d, 'a[0 mod 0]') [ [1.5, 2.6, '234'], [0, 3, '12'], [0, '0 div 0', ''], [1, '0 div 0', ''], ['-42', '1 div 0', '12345'], ['-1 div 0', '1 div 0', ''] ].each { |start, length, expected| set = doc.elements.to_a("//test[substring(@string, #{start}, #{length}) = '#{expected}']") assert_equal 1, set.size, "#{start}, #{length}, '#{expected}'" } end def test_translate source = <<-EOF EOF doc = Document.new(source) [ ['bar', 'abc', 'ABC', 'w3c one'], ['--aaa--','abc-','ABC', 'w3c two'], ['lead', 'dear language', 'doll groover', 'alchemy'], ['A Space Odissei', 'i', 'y', 'vbxml one'], ['abcdefg', 'aceg', 'ACE', 'vbxml two'], ].each { |arg1, arg2, arg3, name| translate = "translate('#{arg1}', '#{arg2}', '#{arg3}')" set = doc.elements.to_a("//case[@result = #{translate}]") assert_equal 1, set.size, translate assert_equal name, set[0].attributes['name'] } end def test_math d = Document.new( '' ) assert XPath.first( d.root, 'node()[1]' ) assert_equal 'b', XPath.first( d.root, 'node()[1]' ).name assert XPath.first( d.root, 'node()[0 + 1]' ) assert_equal 'b', XPath.first( d.root, './node()[0 + 1]' ).name assert XPath.first( d.root, 'node()[1 + 1]' ) assert_equal 'c', XPath.first( d.root, './node()[1 + 1]' ).name assert XPath.first( d.root, 'node()[4 div 2]' ) assert_equal 'c', XPath.first( d.root, './node()[4 div 2]' ).name assert XPath.first( d.root, 'node()[2 - 1]' ) assert_equal 'b', XPath.first( d.root, './node()[2 - 1]' ).name assert XPath.first( d.root, 'node()[5 mod 2]' ) assert_equal 'b', XPath.first( d.root, './node()[5 mod 2]' ).name assert XPath.first( d.root, 'node()[8 mod 3]' ) assert_equal 'c', XPath.first( d.root, './node()[8 mod 3]' ).name assert XPath.first( d.root, 'node()[1 * 2]' ) assert_equal 'c', XPath.first( d.root, './node()[1 * 2]' ).name assert XPath.first( d.root, 'node()[2 + -1]' ) assert_equal 'b', XPath.first( d.root, './node()[2 + -1]' ).name end def test_name assert_raise( UndefinedNamespaceException, "x should be undefined" ) { REXML::Document.new("") } d = REXML::Document.new("") assert_equal 1, d.root.elements.to_a('*[name() = "b"]').size assert_equal 1, d.elements.to_a('//*[name() = "x:b"]').size end def test_local_name d = REXML::Document.new("") assert_equal 2, d.root.elements.to_a('*[local_name() = "b"]').size assert_equal 2, d.elements.to_a('//*[local_name() = "b"]').size end def test_comparisons source = "" doc = REXML::Document.new(source) # NOTE TO SER: check that number() is required assert_equal 2, REXML::XPath.match(doc, "//b[number(@id) > 1]").size assert_equal 3, REXML::XPath.match(doc, "//b[number(@id) >= 1]").size assert_equal 1, REXML::XPath.match(doc, "//b[number(@id) <= 1]").size assert_equal 1, REXML::XPath.match(doc, "//b[number(@id) = (1 * 1)]").size assert_equal 1, REXML::XPath.match(doc, "//b[number(@id) = (1 mod 2)]").size assert_equal 1, REXML::XPath.match(doc, "//b[number(@id) = (4 div 2)]").size end # Contributed by Kouhei def test_substring_before doc = Document.new("") assert_equal("a", doc.root.elements.to_a("*[name()=substring-before('abc', 'b')]")[0].name) assert_equal("c", doc.root.elements.to_a("*[name()=substring-after('abc', 'b')]")[0].name) end def test_spaces doc = Document.new(" ") assert_equal( 1, REXML::XPath.match(doc, "//*[local-name()='c' and @id='b']").size ) assert_equal( 1, REXML::XPath.match(doc, "//*[ local-name()='c' and @id='b' ]").size ) assert_equal( 1, REXML::XPath.match(doc, "//*[ local-name() = 'c' and @id = 'b' ]").size ) assert_equal( 1, REXML::XPath.match(doc, '/a/c[@id]').size ) assert_equal( 1, REXML::XPath.match(doc, '/a/c[(@id)]').size ) assert_equal( 1, REXML::XPath.match(doc, '/a/c[ @id ]').size ) assert_equal( 1, REXML::XPath.match(doc, '/a/c[ (@id) ]').size ) assert_equal( 1, REXML::XPath.match(doc, '/a/c[( @id )]').size ) assert_equal( 1, REXML::XPath.match(doc.root, '/a/c[ ( @id ) ]').size ) assert_equal( 1, REXML::XPath.match(doc, '/a/c [ ( @id ) ] ').size ) assert_equal( 1, REXML::XPath.match(doc, ' / a / c [ ( @id ) ] ').size ) end def test_text_nodes # source = " # #test #" source = "test" d = REXML::Document.new( source ) r = REXML::XPath.match( d, %q{/root/child[text()="test"]} ) assert_equal( 1, r.size ) assert_equal( "child", r[0].name ) assert_equal( "test", r[0].text ) end def test_auto_string_value source = "Introduction" d = REXML::Document.new( source ) #r = REXML::XPath.match( d, %q{/root[title="Introduction"]} ) #assert_equal( 1, r.size ) source = "test" d = REXML::Document.new( source ) r = REXML::XPath.match( d, %q{/a[c='test']} ) assert_equal( 1, r.size ) r = REXML::XPath.match( d, %q{a[c='test']} ) assert_equal( 1, r.size ) r = d.elements["/a[c='test']"] assert_not_nil( r ) r = d.elements["a[c='test']"] assert_not_nil( r ) r = d.elements["a[c='xtest']"] assert_nil( r ) r = REXML::XPath.match( d, %q{a[c='xtest']} ) assert_equal( 0, r.size ) end def test_ordering source = "" d = REXML::Document.new( source ) r = REXML::XPath.match( d, %q{/a/*/*[1]} ) assert_equal( 1, r.size ) r.each { |el| assert_equal( '1', el.attribute('id').value ) } end def test_descendant_or_self_ordering source = " " d = REXML::Document.new( source ) cs = XPath.match( d, "/descendant-or-self::c" ) assert_equal( 4, cs.length ) 1.upto(4) {|x| assert_equal( x.to_s, cs[x-1].attributes['id'] ) } end def test_and d = Document.new %q{} assert_equal( nil, d.root.elements["route[@run='0']"] ) assert_equal( nil, d.root.elements["route[@run='0' and @title='HNO']"] ) end def test_numbers d = Document.new %q{} xp1 = "/a[ @x = 0 ]" xp2 = "/a[ @x = '0' ]" xp3 = "/a[ (@x + 1) = 1 ]" xp4 = "/a[ @y = 0 ]" xp5 = "/a[ (@z + 1) = 5 ]" xp6 = "/a[ (@w + 1) = 5 ]" xp7 = "/a[ (@v + 1) = 1 ]" xp8 = "/a[ @n = 0 ]" assert_equal( 1, XPath.match( d, xp1 ).length ) assert_equal( 1, XPath.match( d, xp2 ).length ) assert_equal( 1, XPath.match( d, xp3 ).length ) assert_equal( 0, XPath.match( d, xp4 ).length ) assert_equal( 0, XPath.match( d, xp5 ).length ) assert_equal( 0, XPath.match( d, xp6 ).length ) assert_equal( 0, XPath.match( d, xp7 ).length ) assert_equal( 0, XPath.match( d, xp8 ).length ) end def test_tobis_preceding doc_string = ' ' doc = Document.new(doc_string) # e = REXML::XPath.first(doc,'/a/c/e') e = doc.root.get_elements('/a/c/e')[0] assert_equal( 1, e.get_elements('preceding-sibling::*').length ) assert_equal( 2, XPath.match(e, 'preceding::*').length ) end def test_filtering #doc=Document.new("") #assert_equal( 3, XPath.match( doc, '/a/b/*[1]' ).length ) #assert_equal( 2, XPath.match( doc, '/a/b/following-sibling::*[1]' ).length ) end # Submitted by Alex def test_union data = %Q{

a b
} rd = REXML::Document.new( data ) #union = rd.get_elements("/div/span | /div/em") #assert_equal(2, union.length, "/div/span | /div/em" ) union = rd.get_elements('//*[name()="em" or name()="strong"]') assert_equal(2, union.length, 'name() and "or" failed') union = rd.get_elements('//em|//strong') assert_equal(2, union.length, 'Both tag types are returned by XPath union operator') end def test_union2 src = <<-EOL
a b
EOL rd = REXML::Document.new( src ) union = rd.get_elements('//em|//strong') assert_equal(2, union.length, 'Both tag types are returned by XPath union operator') end def test_a_star_star_one string = <<-EOL EOL d = REXML::Document.new( string ) c1 = XPath.match( d, '/a/*/*[1]' ) assert_equal( 1, c1.length ) assert_equal( 'c1', c1[0].name ) end def test_sum d = Document.new(""+ "123"+ "12"+ ""+ "") for v,p in [[6, "sum(/a/b)"], [9, "sum(//b | //d)"], [3, "sum(/a/e/@*)"] ] assert_equal( v, XPath::match( d, p ).first ) end end def test_xpath_namespace d = REXML::Document.new("xa") x = d.root num = 0 x.each_element('tada') { num += 1 } assert_equal(1, num) end def test_ticket_39 doc = REXML::Document.new( <<-EOL ) true Item 1 Item 2 Thu, 13 Oct 2005 19:59:00 +0000 Item 3 EOL root_node = XPath.first(doc, "rss") assert_not_nil root_node channel_node = XPath.first(root_node, "channel") assert_not_nil channel_node items = XPath.match(channel_node, "*") assert_equal 4, items.size items = XPath.match(channel_node, "item") assert_equal 3, items.size # fails end def test_ticket_42 source = "" doc = Document.new(source) bElem = Element.new('b') doc.root.add_element(bElem) doc.elements.each('//b[name(..) = "a"]') { |x| assert_equal x,bElem } end def test_ticket_56 namespaces = {'h' => 'http://www.w3.org/1999/xhtml'} finaldoc = REXML::Document.new(File.read(fixture_path('google.2.xml'))) column_headers = [] REXML::XPath.each(finaldoc, '//h:form[@action="ModifyCampaign"]//h:th', namespaces) do |el| node = REXML::XPath.first(el, 'h:a/text()', namespaces) column_headers << (node ? node.value : nil) end column_headers.map! { |h| h.to_s.strip.chomp } expected = ["", "", "Current Status", "Current Budget", "Clicks", "Impr.", "CTR", "Avg. CPC", "Cost", "Conv. Rate", "Cost/Conv."] assert_equal( expected, column_headers ) end def test_ticket_70 string = < Text1, text, text Text2, text, text EOF doc = Document.new string assert_equal( 1, XPath.match( doc, "//someelement[contains(@attribute,'1.10')]" ).length ) end def test_ticket_43 #url = http://news.search.yahoo.com/news/rss?p=market&ei=UTF-8&fl=0&x=wrt sum = Document.new(File.new(fixture_path("yahoo.xml"))).elements.to_a("//item").size assert_equal( 10, sum ) text = Document.new(File.new(fixture_path("yahoo.xml"))).elements.to_a(%Q{//title[contains(text(), "'")]}).collect{|e| e.text}.join assert_equal( "Broward labor market's a solid performer (Miami Herald)", text ) end def test_ticket_57 data = "zzz" r = Document.new(data) assert_equal(Text, REXML::XPath.first(r,"a:x/a:y[@p='p' and @q='q']/a:z/text()").class) assert_equal("zzz", REXML::XPath.first(r,"a:x/a:y[@p='p' and @q='q']/a:z/text()").to_s) end def test_ticket_59 data = " " d = Document.new(data) res = d.elements.to_a( "//c" ).collect {|e| e.attributes['id'].to_i} assert_equal( res, res.sort ) end def ticket_61_fixture(doc, xpath) matches = [] doc.elements.each(xpath) do |element| matches << element assert_equal('Add', element.text) assert_equal('ButtonText', element.attributes['class']) end assert_equal(1, matches.length) end def test_ticket_61_text file = File.open(fixture_path("ticket_61.xml")) doc = REXML::Document.new file ticket_61_fixture( doc, "//div[text()='Add' and @class='ButtonText']" ) end def test_ticket_61_contains file = File.open(fixture_path("ticket_61.xml")) doc = REXML::Document.new file ticket_61_fixture( doc, "//div[contains(.,'Add') and @class='ButtonText']" ) end def test_namespaces_0 d = Document.new(%q{}) assert_equal( 1, XPath.match( d, "//x:a" ).size ) assert_equal( 1, XPath.match( d, "//x:*" ).size ) end def test_ticket_71 doc = Document.new(%Q{}) el = doc.root.elements[1] assert_equal( "element", el.name ) el2 = XPath.first( doc.root, "element[@ns:attrname='foo']", { 'ns' => "xyz" } ) assert_equal( el, el2 ) end def test_ticket_78 doc = <<-EOT 123 123a EOT seq = %w{BEGIN 123 END BEGIN 123a END} xmlDoc = Document.new(doc) ["//element[tag='123']/tag", "//element[tag='123a']/tag"].each do |query| assert_equal( "BEGIN", seq.shift ) XPath.each(xmlDoc, query) { |element| assert_equal( seq.shift, element.text ) } assert_equal( "END", seq.shift ) end end def test_ticket_79 source = "test3" d = REXML::Document.new( source ) r = REXML::XPath.match( d, %q{/a/b[c='test']} ) assert_equal(1, r.size()) r = REXML::XPath.match( d, %q{/a/b[c='3']} ) assert_equal(1, r.size()) end def test_or_and doc = " test

A link.

" xmldoc = REXML::Document.new(doc) xpath = "descendant::node()[(local-name()='link' or local-name()='a') and @rel='sub']" hrefs = [] xmldoc.elements.each(xpath) do |element| hrefs << element.attributes["href"] end assert_equal(["/"], hrefs, "Bug #3842 [ruby-core:32447]") end end