Toggle navigation

What’s new in Tornado 3.2

Jan 14, 2014


  • Tornado now depends on the backports.ssl_match_hostname when running on Python 2. This will be installed automatically when using pip or easy_install
  • Tornado now includes an optional C extension module, which greatly improves performance of websockets. This extension will be built automatically if a C compiler is found at install time.

New modules

  • The tornado.platform.asyncio module provides integration with the asyncio module introduced in Python 3.4 (also available for Python 3.3 with pip install asyncio).


  • Added GoogleOAuth2Mixin support authentication to Google services with OAuth 2 instead of OpenID and OAuth 1.
  • FacebookGraphMixin has been updated to use the current Facebook login URL, which saves a redirect.


  • TracebackFuture now accepts a timeout keyword argument (although it is still incorrect to use a non-zero timeout in non-blocking code).


  • tornado.curl_httpclient now works on Python 3 with the soon-to-be-released pycurl 7.19.3, which will officially support Python 3 for the first time. Note that there are some unofficial Python 3 ports of pycurl (Ubuntu has included one for its past several releases); these are not supported for use with Tornado.



  • Coroutines may now yield dicts in addition to lists to wait for multiple tasks in parallel.
  • Improved performance of tornado.gen when yielding a Future that is already done.


  • tornado.httpclient.HTTPRequest now uses property setters so that setting attributes after construction applies the same conversions as __init__ (e.g. converting the body attribute to bytes).


  • Malformed x-www-form-urlencoded request bodies will now log a warning and continue instead of causing the request to fail (similar to the existing handling of malformed multipart/form-data bodies. This is done mainly because some libraries send this content type by default even when the data is not form-encoded.
  • Fix some error messages for unix sockets (and other non-IP sockets)


  • IOLoop now uses handle_callback_exception consistently for error logging.
  • IOLoop now frees callback objects earlier, reducing memory usage while idle.
  • IOLoop will no longer call logging.basicConfig if there is a handler defined for the root logger or for the tornado or tornado.application loggers (previously it only looked at the root logger).


  • IOStream now recognizes ECONNABORTED error codes in more places (which was mainly an issue on Windows).
  • IOStream now frees memory earlier if a connection is closed while there is data in the write buffer.
  • PipeIOStream now handles EAGAIN error codes correctly.
  • SSLIOStream now initiates the SSL handshake automatically without waiting for the application to try and read or write to the connection.
  • Swallow a spurious exception from set_nodelay when a connection has been reset.




  • is_valid_ip (and therefore HTTPRequest.remote_ip) now rejects empty strings.
  • Synchronously using ThreadedResolver at import time to resolve a unicode hostname no longer deadlocks.




  • simple_httpclient now applies the connect_timeout to requests that are queued and have not yet started.
  • On Python 2.6, simple_httpclient now uses TLSv1 instead of SSLv3.
  • simple_httpclient now enforces the connect timeout during DNS resolution.
  • The embedded ca-certificates.crt file has been updated with the current Mozilla CA list.


  • StaticFileHandler no longer fails if the client requests a Range that is larger than the entire file (Facebook has a crawler that does this).
  • RequestHandler.on_connection_close now works correctly on subsequent requests of a keep-alive connection.
  • New application setting default_handler_class can be used to easily set up custom 404 pages.
  • New application settings autoreload, compiled_template_cache, static_hash_cache, and serve_traceback can be used to control individual aspects of debug mode.
  • New methods RequestHandler.get_query_argument and RequestHandler.get_body_argument and new attributes HTTPRequest.query_arguments and HTTPRequest.body_arguments allow access to arguments without intermingling those from the query string with those from the request body.
  • RequestHandler.decode_argument and related methods now raise an HTTPError(400) instead of UnicodeDecodeError when the argument could not be decoded.
  • RequestHandler.clear_all_cookies now accepts domain and path arguments, just like clear_cookie.
  • It is now possible to specify handlers by name when using the tornado.web.URLSpec class.
  • Application now accepts 4-tuples to specify the name parameter (which previously required constructing a tornado.web.URLSpec object instead of a tuple).
  • Fixed an incorrect error message when handler methods return a value other than None or a Future.
  • Exceptions will no longer be logged twice when using both @asynchronous and @gen.coroutine



  • WSGIContainer now calls the iterable’s close() method even if an error is raised, in compliance with the spec.