aboutsummaryrefslogtreecommitdiffstats
path: root/doc/src/guide/migrating_from_2.4.asciidoc
blob: 3cdeaa549e0e393808d4b811256f3573c88da629 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
[appendix]
== Migrating from Cowboy 2.4 to 2.5

Cowboy 2.5 focused on making the test suites pass. A
variety of new features, fixes and improvements have
also been worked on.

=== Features added

* Add option `linger_timeout` to control how long
  Cowboy will wait before closing the socket when
  shutting down the connection. This helps avoid
  the TCP reset problem HTTP/1.1 suffers from. The
  default is now 1000 ms.

* It is now possible to stream a response body
  without using chunked transfer-encoding when the
  protocol is HTTP/1.1. To enable this behavior,
  simply pass the content-length header with the
  expected size when initiating the streamed response.

* Update Ranch to 1.6.2

* Update Cowlib to 2.6.0

=== Experimental features added

* Websocket handlers now feature a commands-based interface.
  The return value from the callbacks can now take the form
  `{Commands, State}` where `Commands` can be frames to be
  sent or commands yet to be introduced. New commands will
  be available only through this new interface.

* Add the `{active, boolean()}` Websocket handler command.
  It allows disabling reading from the socket when `false`
  is returned. `true` reenables reading from the socket.

* Add the protocol option `logger` that allows configuring
  which logger module will be used. The logger module must
  follow the interface of the new `logger` module in Erlang/OTP 21,
  or be set to `error_logger` to keep the old behavior. A
  similar transport option exists in Ranch 1.6; both options
  are necessary to override Cowboy's default behavior completely.

* Add the `{log, Level, Format, Args}` stream handler command.
  Making it a command rather than a direct call will simplify
  silencing particular log messages.

=== New functions

* The function `cowboy_req:stream_events/3` streams one or more
  text/event-stream events, encoding them automatically.

* The functions `cowboy_req:read_and_match_urlencoded_body/2,3`
  can be used to read, parse and match application/x-www-form-urlencoded
  request bodies, in a similar way to `cowboy_req:match_qs/2`.

=== Bugs fixed

* Fix Erlang/OTP 21 warnings.

* Ensure that the port number is always defined in the
  Req object. When it is not provided in the request,
  the default port number for the protocol being used
  will be set.

* Ensure stream handlers can run after `cowboy_stream_h`.

* Honor the SETTINGS_ENABLE_PUSH HTTP/2 setting: don't
  send PUSH frames to clients that disabled it.

* Fix HTTP/2 `settings_timeout` option when the value
  is set to `infinity`.

* HTTP/1.1 responses will no longer include a trailer header
  when the request had no te header.

* HTTP/1.1 204 responses no longer send the transfer-encoding
  header when `cowboy_req:stream_reply/2,3` is used to send
  a response.

* Improve HTTP/1.1 keepalive handling to avoid processing
  requests that follow the final request that will receive
  a response.

* Improve the validation of HTTP/1.1 absolute-form requests.

* When the `switch_protocol` is used after a response was
  sent, Cowboy will no longer attempt to send the 101 informational
  response for the protocol upgrade. This caused a crash of the
  connection previously.

* Errors that occur when a callback returned by
  `content_types_provided` does not exist have been improved.

* Prevent annoying error logs when using sendfile in
  Erlang/OTP 20 and lower.

* Add missing frame types to `websocket_handle`.

* A test suite has been added for RFC8297 to ensure that
  103 informational responses can be sent.

* Numerous test cases have been fixed, improved or removed in order
  to make the test suites pass. Most of the failures were caused
  by broken tests.

* Some misguiding or incorrect statements in the documentation
  have been removed or clarified.