Use String#scrub when available to tidy bytes

activesupport/lib/active_support/multibyte/unicode.rb

@@ -212,6 +212,8 @@ def compose(codepoints)
         codepoints
       end
 
+      # Ruby >= 2.1 has String#scrub, which is faster than the workaround used for < 2.1.
+      if RUBY_VERSION >= '2.1'
         # Replaces all ISO-8859-1 or CP1252 characters by their UTF-8 equivalent
         # resulting in a valid UTF-8 string.
         #
@@ -219,10 +221,13 @@ def compose(codepoints)
         # encoding is entirely CP1252 or ISO-8859-1.
         def tidy_bytes(string, force = false)
           return string if string.empty?
-
-        if force
-          return string.encode(Encoding::UTF_8, Encoding::Windows_1252, invalid: :replace, undef: :replace)
+          return recode_windows1252_chars(string) if force
+          string.scrub { |bad| recode_windows1252_chars(bad) }
         end
+      else
+        def tidy_bytes(string, force = false)
+          return string if string.empty?
+          return recode_windows1252_chars(string) if force
 
           # We can't transcode to the same format, so we choose a nearly-identical encoding.
           # We're going to 'transcode' bytes from UTF-8 when possible, then fall back to
@@ -244,6 +249,7 @@ def tidy_bytes(string, force = false)
 
           out.encode!(Encoding::UTF_8)
         end
+      end
 
       # Returns the KC normalization of the string by default. NFKC is
       # considered the best normalization form for passing strings to databases
@@ -371,14 +377,8 @@ def apply_mapping(string, mapping) #:nodoc:
         end.pack('U*')
       end
 
-      def tidy_byte(byte)
-        if byte < 160
-          [database.cp1252[byte] || byte].pack("U").unpack("C*")
-        elsif byte < 192
-          [194, byte]
-        else
-          [195, byte - 64]
-        end
+      def recode_windows1252_chars(string)
+        string.encode(Encoding::UTF_8, Encoding::Windows_1252, invalid: :replace, undef: :replace)
       end
 
       def database