What’s new in Tornado 4.4¶

Jul 15, 2016¶

General¶

Tornado now requires Python 2.7 or 3.3+; versions 2.6 and 3.2 are no longer supported. Pypy3 is still supported even though its latest release is mainly based on Python 3.2.
The monotonic package is now supported as an alternative to Monotime for monotonic clock support on Python 2.

`tornado.curl_httpclient`¶

Failures in _curl_setup_request no longer cause the max_clients pool to be exhausted.
Non-ascii header values are now handled correctly.

`tornado.gen`¶

with_timeout now accepts any yieldable object (except YieldPoint), not just tornado.concurrent.Future.

`tornado.httpclient`¶

The errors raised by timeouts now indicate what state the request was in; the error message is no longer simply “599 Timeout”.
Calling repr on a tornado.httpclient.HTTPError no longer raises an error.

`tornado.httpserver`¶

Int-like enums (including http.HTTPStatus) can now be used as status codes.
Responses with status code 204 No Content no longer emit a Content-Length: 0 header.

`tornado.ioloop`¶

Improved performance when there are large numbers of active timeouts.

`tornado.netutil`¶

All included Resolver implementations raise IOError (or a subclass) for any resolution failure.

`tornado.options`¶

Options can now be modified with subscript syntax in addition to attribute syntax.
The special variable __file__ is now available inside config files.

`tornado.simple_httpclient`¶

HTTP/1.0 (not 1.1) responses without a Content-Length header now work correctly.

`tornado.tcpserver`¶

TCPServer.bind now accepts a reuse_port argument.

`tornado.testing`¶

Test sockets now always use 127.0.0.1 instead of localhost. This avoids conflicts when the automatically-assigned port is available on IPv4 but not IPv6, or in unusual network configurations when localhost has multiple IP addresses.

`tornado.web`¶

image/svg+xml is now on the list of compressible mime types.
Fixed an error on Python 3 when compression is used with multiple Vary headers.

`tornado.websocket`¶

WebSocketHandler.__init__ now uses super, which improves support for multiple inheritance.