Fix performance issue caused by using repeated > characters inside CDATA [ PAYLOAD ] (#172)

A `<` is treated as a string delimiter. 
In certain cases, if `<` is used in succession, read and match are
repeated, which slows down the process. Therefore, the following is used
to read ahead to a specific part of the string in advance.

Author: Watson1978

lib/rexml/parsers/baseparser.rb

@@ -127,6 +127,7 @@ module Private
         INSTRUCTION_END = /#{NAME}(\s+.*?)?\?>/um
         INSTRUCTION_TERM = "?>"
         COMMENT_TERM = "-->"
+        CDATA_TERM = "]]>"
         TAG_PATTERN = /((?>#{QNAME_STR}))\s*/um
         CLOSE_PATTERN = /(#{QNAME_STR})\s*>/um
         ATTLISTDECL_END = /\s+#{NAME}(?:#{ATTDEF})*\s*>/um
@@ -431,7 +432,7 @@ def pull_event
 
                 return [ :comment, md[1] ]
               else
-                md = @source.match(/\[CDATA\[(.*?)\]\]>/um, true)
+                md = @source.match(/\[CDATA\[(.*?)\]\]>/um, true, term: Private::CDATA_TERM)
                 return [ :cdata, md[1] ] if md
               end
               raise REXML::ParseException.new( "Declarations can only occur "+

test/parse/test_cdata.rb

@@ -0,0 +1,17 @@
+require "test/unit"
+require "core_assertions"
+
+require "rexml/document"
+
+module REXMLTests
+  class TestParseCData < Test::Unit::TestCase
+    include Test::Unit::CoreAssertions
+
+    def test_gt_linear_performance
+      seq = [10000, 50000, 100000, 150000, 200000]
+      assert_linear_performance(seq, rehearsal: 10) do |n|
+        REXML::Document.new('<description><![CDATA[ ' + ">" * n + ' ]]></description>')
+      end
+    end
+  end
+end