Fix ReDoS

lib/rdoc/comment.rb

@@ -97,42 +97,26 @@ def extract_call_seq method
     # comment.  The difficulty is to make sure not to match lines starting
     # with ARGF at the same indent, but that are after the first description
     # paragraph.
-    if @text =~ /^\s*:?call-seq:(.*?(?:\S).*?)^\s*$/m then
+    if /^(?<S> ((?!\n)\s)*+        (?# whitespaces except newline))
+         :?call-seq:
+           (?<B> \g<S>(?<N>\n|\z)  (?# trailing spaces))?
+         (?<seq>
+           (\g<S>(?!\w)\S.*\g<N>)*
+           (?>
+             (?<H> \g<S>\w+        (?# ' #   ARGF' in the example above))
+             .*\g<N>)?
+           (\g<S>\S.*\g<N>         (?# other non-blank line))*+
+           (\g<B>+(\k<H>.*\g<N>    (?# ARGF.to_a lines))++)*+
+         )
+         (?m:^\s*$|\z)
+        /x =~ @text
+      seq = $~[:seq]
+
       all_start, all_stop = $~.offset(0)
-      seq_start, seq_stop = $~.offset(1)
-
-      # we get the following lines that start with the leading word at the
-      # same indent, even if they have blank lines before
-      if $1 =~ /(^\s*\n)+^(\s*\w+)/m then
-        leading = $2 # ' *    ARGF' in the example above
-        re = %r%
-          \A(
-             (^\s*\n)+
-             (^#{Regexp.escape leading}.*?\n)+
-            )+
-          ^\s*$
-        %xm
-
-        if @text[seq_stop..-1] =~ re then
-          all_stop = seq_stop + $~.offset(0).last
-          seq_stop = seq_stop + $~.offset(1).last
-        end
-      end
-
-      seq = @text[seq_start..seq_stop]
-      seq.gsub!(/^\s*(\S|\n)/m, '\1')
       @text.slice! all_start...all_stop
 
-      method.call_seq = seq.chomp
-
-    else
-      regexp = /^\s*:?call-seq:(.*?)(^\s*$|\z)/m
-      if regexp =~ @text then
-        @text = @text.sub(regexp, '')
-        seq = $1
-        seq.gsub!(/^\s*/, '')
-        method.call_seq = seq
-      end
+      seq.gsub!(/^\s*/, '')
+      method.call_seq = seq
     end
 
     method

lib/rdoc/parser/ruby.rb

@@ -2134,7 +2134,7 @@ def read_directive allowed
       if :on_nl == tk[:kind] or (:on_kw == tk[:kind] && 'def' == tk[:text]) then
         return
       elsif :on_comment == tk[:kind] or :on_embdoc == tk[:kind] then
-        return unless tk[:text] =~ /\s*:?([\w-]+):\s*(.*)/
+        return unless tk[:text] =~ /:?\b([\w-]+):\s*(.*)/
 
         directive = $1.downcase
 

test/rdoc/test_rdoc_comment.rb

@@ -206,6 +206,15 @@ def test_extract_call_seq_c_separator
     assert_equal expected, comment.text
   end
 
+  def test_extract_call_linear_performance
+    pre = ->(n) {[n, RDoc::Comment.new("\n"*n + 'call-seq:' + 'a'*n)]}
+    method_obj = RDoc::AnyMethod.new nil, 'blah'
+    assert_linear_performance((2..5).map {|i| 10**i}, pre: pre) do |n, comment|
+      comment.extract_call_seq method_obj
+      assert_equal n, method_obj.call_seq.size
+    end
+  end
+
   def test_force_encoding
     @comment = RDoc::Encoding.change_encoding @comment, Encoding::UTF_8
 

test/rdoc/test_rdoc_parser_ruby.rb

@@ -3350,6 +3350,13 @@ def test_read_directive_one_liner
     assert_equal :on_const, parser.get_tk[:kind]
   end
 
+  def test_read_directive_linear_performance
+    pre = ->(i) {util_parser '# ' + '0'*i + '=000:'}
+    assert_linear_performance((1..5).map{|i|10**i}, pre: pre) do |parser|
+      assert_nil parser.read_directive []
+    end
+  end
+
   def test_read_documentation_modifiers
     c = RDoc::Context.new