You should rectify the line delimeters immediately after reading a file, if you continue using binmode.

[unak]

this patch is for MRI, not for rdoc. sorry.

Index: lib/rdoc/encoding.rb
===================================================================
--- lib/rdoc/encoding.rb    (revision 33648)
+++ lib/rdoc/encoding.rb    (working copy)
@@ -20,6 +20,7 @@ module RDoc::Encoding

   def self.read_file filename, encoding, force_transcode = false
     content = open filename, "rb" do |f| f.read end
+    content.gsub!("\r\n", "\n") if RUBY_PLATFORM =~ /mswin|mingw/

     utf8 = content.sub!(/\A\xef\xbb\xbf/, '')

Index: test/rdoc/test_rdoc_markup_pre_process.rb
===================================================================
--- test/rdoc/test_rdoc_markup_pre_process.rb   (revision 33648)
+++ test/rdoc/test_rdoc_markup_pre_process.rb   (working copy)
@@ -43,11 +43,6 @@ Regular expressions (<i>regexp</i>s) are
 contents of a string.
     EXPECTED

-    # FIXME 1.9 fix on windoze
-    # preprocessor uses binread, so line endings are \r\n
-    expected.gsub!("\n", "\r\n") if
-      RUBY_VERSION < "1.9.3" && RUBY_PLATFORM =~ /mswin|mingw/
-
     assert_equal expected, content
   end

@@ -67,11 +62,6 @@ contents of a string.

     expected = "?\n"

-    # FIXME 1.9 fix on windoze
-    # preprocessor uses binread, so line endings are \r\n
-    expected.gsub!("\n", "\r\n") if
-      RUBY_VERSION < "1.9.3" && RUBY_PLATFORM =~ /mswin|mingw/
-
     assert_equal expected, content
   end

[unak]

r33902 of MRI

[drbrain]

Thank you for the patch, it was easy to apply.