bert.ws
A JavaScript library for encoding and decoding BERT data. Primarily targeted for use with web sockets, but it can work with any transport protocol.
What is BERT?
In short, BERT is a special binary format for use with Erlang/Elixir applications.
In a slightly longer explanation: BERT stands for Binary ERlang Term. It's a concise means of representing the native data types present in Erlang/Elixir, specifically designed for efficient communication of these terms, as Erlang calls them, over the wire.
If you want even more detailed explanation, you can refer to the BERT-RPC spec or the official Erlang External Term Format:
Installation
Two methods: manual and npm.
- To install manually, simply copy
bert.js
into your project. Done! Unfortunately, if the file is ever updated it's on you to keep track of the version manually and update the file accordingly. - To install via npm, simply do
npm install bert.ws
. You can also add it as a dependency topackages.json
and let npm manage version locking for you. Fun times!
Why bert.ws
? Both bert
and bert.js
are taken by packages equally as old as this one that will also probably never receive an update. However, mine has the advantage of including a web socket module that simplifies integration for that protocol, thus the .ws
suffix.
Usage
Two core methods exist:
Bert.encode/1
Bert.decode/1
Obviously the first one encodes JavaScript into BERT and the second one decodes BERT into JavaScript. The main constraints are:
- To
encode
data you should pass in aBertObj
. If you don't, a function calledsmartCast/1
will be used to try to "intelligently" determine the appropriate associated Erlang type. But because JavaScript has fewer types than Erlang, chances are that something might be cast incorrectly, so explicitly creating Bert helper objects is recommended. But not required. - To
decode
data you must pass in a string or an array. This assumes that you're dealing with data from ablob
orarraybuffer
, the two type options for web sockets when sending and receiving data.
Helper functions exist to quickly and easily create BertObj
instances. I shall reiterate: it is highly recommended that you use these helpers instead of passing raw JavaScript objects to the encoder. It's the only way to create and encode atoms and tuples, for example. Not using the helper functions will seriously limit what kind of Erlang types you can send upstream.
e.g.
// Encode string as a binaryBertBert // Encode string as a "string" (char list)Bert // Encode a list of valuesBertBert // Encode a map of key-value pairsBertBert
Maps
A note about using the Map
BERT type:
JavaScript does not have a way to iterate over the keys in a Map while retaining the key type. In JavaScript, retrieving keys from a regular object will always return a list of strings. If you need to use typed keys and can't have strings, then you should pass in an array of arrays, where each inside array has two elements: a key and a value, in that order. This will construct a special kind of Map that can be used to preserve key types during encoding.
e.g.
// All keys become binary strings (values are encoded correctly though)Bert // Preserve key types when encoding a "map"// Notice we can still take advantage of smart casting with this approachBert
It's also worth noting that decoding an Erlang Map to JavaScript will result in either Number
or String
"keys" for the resulting object. For string keys, the dot notation (e.g. obj.some_key
) will work but for number keys the index operator must be used (e.g. obj[42]
). The index operator may still be used with string keys (e.g. obj["some_key"]
).
WebSockets
I spoke about how I was targeting WebSockets earlier. I've included an additional JavaScript file, called bertws.js
, which shows a basic implementation of using bert.js
to read and write BERT data over a WebSocket connection.
The basic concept, though, is that you set the WebSocket to use "arraybuffer"
as your binary type. The default is "blob"
, but that's more like working with a file. It's doable, but I like the idea of working with a Uint8Array
because it is more akin to a "byte array." So we use an array buffer to generate a typed array which we then pass to Bert.decode
. And because Bert.encode
returns a typed array we can send it directly over the wire; no need to transform the array further because the WebSocket implementation handles that for us.
TODO
- Cleanup code. The current implementation works great and looks good thanks to ES6, but it could be even better.