hpack

HPACK Implementation for Erlang

This implementation was designed for use by Chatterbox, but could be used by any HTTP/2 implementation (or, if something other than HTTP/2 has need for HPACK, this would work as well)

Why Separate?

Use by other projects
A separate RFC seemed like a really clear abstraction

What's Covered

Indexed Header Field Representation
Literal Header Field Representation
Dynamic Table Size Update

What's not covered

HTTP/2 Frames

An HTTP/2 implementation should be responsible for anything you need to do in order to read a compressed header block from various PUSH_PROMISE, HEADERS, and/or CONTINUATION frames. Once you have a complete block, you can use this library to turn it into something you can use for fulfilling HTTP requests

API

Creating Contexts

Encoding Contexts and Decoding Contexts are now the same type. You can create new one with hpack:new_context() or pass an argument which is an integer of byte size provided by HTTP/2's HEADER_TABLE_SIZE setting.

Changing table size

If HTTP/2 settings get renegotiated, you can pass that information along by calling hpack:new_max_table_size/2, like this:

NewContext = hpack:new_max_table_size(NewSize, OldContext),

Decoding Headers

Decode a headers binary with hpack:decode/2. It's your job to assemble the binary if it's coming from multiple HTTP/2 frames.

{Headers, NewContext} = hpack:decode(Binary, OldContext),

Headers is of type [{binary(), binary()}]

Encoding Headers

Encoding headers works the same way, only a [{binary(), binary()}] goes in, and a binary() comes out.

{Bin, NewContext} = hpack:encode(Headers, OldContext),

Soup to nuts

Here's how to do the whole thing!

DecodeContext1 = hpack:new_context(), %% Server context
EncodeContext1 = hpack:new_context(), %% client context

ClientRequestHeaders = [
        {<<":method">>, <<"GET">>},
        {<<":path">>, <<"/">>},
        {<<"X-Whatev">>, <<"commands">>}
    ],

%% Client operation
{RequestHeadersBin, EncodeContext2} = hpack:encode(
    ClientRequestHeaders,
    EncodeContext1),

%% Server operation, after receiving RequestHeadersBin
{ServerRequestHeaders, DecodeContext2} = hpack:decode(
    RequestHeadersBin,
    DecodeContext1),

%% Note the following truths:
ClientRequestHeaders = ServerRequestHeaders,
EncodeContext1 = DecodeContext1,
EncodeContext2 = DecodeContext2.

The whole reason hpack works is that the client and server both keep their contexts in sync with each other.

** Note: I used the terms client and server here, but it could as easily be sender and receiver if you're a proxy

lexus300h/hpack