Allow non-ascii (but still latin1) in our fake byte literals

tornado/test/httpserver_test.py

@@ -110,7 +110,7 @@ class HTTPConnectionTest(AsyncHTTPTestCase, LogTrapTestCase):
         response = self.raw_fetch([
                 b("POST /multipart HTTP/1.0"),
                 b("Content-Type: multipart/form-data; boundary=1234567890"),
-                u"X-Header-encoding-test: \u00e9".encode("latin1"),
+                b("X-Header-encoding-test: \xe9"),
                 ],
                                   b("\r\n").join([
                     b("Content-Disposition: form-data; name=argument"),

tornado/util.py

@@ -19,11 +19,11 @@ def import_object(name):
 # a byte literal (str in 2.x, bytes in 3.x).  There's no way to do this
 # in a way that supports 2.5, though, so we need a function wrapper
 # to convert our string literals.  b() should only be applied to literal
-# ascii strings.  Once we drop support for 2.5, we can remove this function
+# latin1 strings.  Once we drop support for 2.5, we can remove this function
 # and just use byte literals.
 if str is unicode:
     def b(s):
-        return s.encode('ascii')
+        return s.encode('latin1')
     bytes_type = bytes
 else:
     def b(s):