Switch branches/tags
Nothing to show
Find file
Fetching contributors…
Cannot retrieve contributors at this time
140 lines (98 sloc) 4.81 KB


This document explains in as much details as possible the list of planned changes and work to be done on the Cowboy server. It is non-exhaustive and subject to change. Items are not ordered.

  • Write more, better examples.

    The first step would be to port misultin's examples to Cowboy. Then these examples could be completed with examples for Cowboy specific features.

    The extend/cowboy_examples is to be used for this. As it is a separate repository, we can organize the file structure as appropriate. Ideally we would have one complete example per folder.

    Examples should be commented. They may or may not be used for writing the user guides.

  • Write user guides.

    We currently have good API documentation, but no step by step user guides.

  • Write more, better tests.

    Amongst the areas less tested there is protocol upgrades and the REST handler.

    Current tests should be completed with unit tests where applicable. We should probably also test the dependencies used, like erlang:decode_packet/3.

    While eunit and ct tests are fine, some parts of the code could benefit from PropEr tests.

  • Continuous performance testing.

    Initially dubbed the Horse project, Cowboy could benefit from a continuous performance testing tool that would allow us to easily compare the impact of the changes we are introducing, similar to what the Phoronix test suite allows.

    Depending on the test it may be interesting to compare Cowboy to other servers and eventually take ideas from the servers that outperform Cowboy for the task being tested.

  • Improve HTTP/1.0 support.

    Most of the work on Cowboy has been done with HTTP/1.1 in mind. But there is still a need for HTTP/1.0 code in Cowboy. The server code should be reviewed and tested to ensure compatibility with remaining HTTP/1.0 products.

  • HTTP 100 Continue support.

    Tools like curl expect a 100 Continue before sending a request body by default.

  • Content-Encoding support.

    Cowboy should be able to send encoded content automatically. The default should be to encode, but the developer must be able to override this default either for the whole listener or just for a single reply.

  • Improve body reading API.

    We want to have various different things for reading the body. First, there should be raw functions for the different ways to read the body: basic, transfer encoded, multipart. Each should allow us to limit the size of what is read.

    On top of these functions there should be two more advanced functions: one would return the result of parsing a x-www-form-urlencoded body; the other would parse a multipart request, save files from the multipart data to a temporary location and return a proplist of values if any along with the files details. This behavior is similar to what is done automatically by PHP with its $_FILES array.

    The advanced functions are of course here for convenience only and it should be trivial to reimplement them directly in a Cowboy application if needed.

  • Complete the work on Websockets.

    Now that the Autobahn test suite is available (make inttests), we have a definite way to know whether Cowboy's implementation of Websockets is right. The work can thus be completed. The remaining tasks are proper UTF8 handling and fragmentation.

  • SPDY support.

    While SPDY probably won't be added directly to Cowboy, work has been started on making Cowboy use SPDY.

  • Hooks.

    Customizable hooks would allow the developer to extend Cowboy easily. Two kinds of hooks are needed: before dispatching the request, and before sending a reply.

    The first would allow us to apply site-wide functions like authentication or request logging and modify the Req if needed.

    The second is more interesting for response logging or to filter the replies, for example to send custom error pages.

  • Transport upgrades.

    Some protocols allow an upgrade from TCP to SSL without closing the connection. This is currently not possible through the Cowboy API.

  • Resizing the acceptor pool.

    We should be able to add more acceptors to a pool but also to remove some of them as needed.

  • Simplified dispatch list.

    For convenience purposes, the dispatch list should allow lists instead of binaries. The lists can be converted to binary by Cowboy at listener initialization.

    There has also been discussion on allowing the dispatch list to be hierarchical.

  • Add Transport:secure/0.

    Currently Cowboy checks if a connection is secure by checking if its name is 'ssl'. This isn't a very modular solution, adding an API function that returns whether a connection is secure would fix that issue.