Clojure implementation of the Asynchronous Messaging Protocol
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.


A Clojure implementation of the Asynchronous Messaging Protocol, or AMP.


Build Status Dependency Status

Clojars Project



An example server and two example clients are provided:

These will interoperate with each other, as well as the and examples from the Twisted documentation.

A note about "byte strings"

The AMP protocol uses sequences of arbitrary bytes in many places that a human-readable identifier is expected. While some languages have a convenient way to express such sequences of bytes with a constant (for example, b'foo' in Python), Clojure lacks such a thing. In their place, clj-amp makes use of regular strings (which are just the Java String type, normally used to represent sequences of Unicode codepoints); these strings are translated to bytes by way of ISO-8859-1 encoding, which has the convenient property that all encodings are 1-byte encodings, and furthermore the first 256 Unicode codepoints map precisely to the 256 possible values for a byte.

This means that you can write "foo" in the places in clj-amp where you would expect to be able to; you can also express arbitrary sequences of bytes by using a string with the appropriate codepoints. For example, if you needed to express the sequence 01 FF 1A, this would be written as the string "\u0001\u00FF\u001A".

It is expected that these values will be written as constants in almost all cases; at the moment, there are no helpers for converting from other representations, but these may be added if demand for them presents itself.

When reference is made to a "byte string" elsewhere in the documentation, this is the construct to which it refers.

Defining commands


A defcommand macro is provided in clj-amp.command for command definitions:

(defcommand name
 {argument-name {:type argument-type
                 :name amp-argument-name?
 {response-name {:type argument-type
                 :name amp-response-name?
 :command-name command-name?
 :errors {::error-type "ERROR-CODE"

The arguments and response items are required, but you can pass an empty map if necessary. Responders are passed a map of arguments, and are expected to return a map in response; the keys are the same as specified in the command definition.

By default, the protocol-level name will be derived from the Clojure-level name, but you can pass :command-name with a byte string to override the command name itself, or :name with a byte string to override an argument or response item name.

An errors mapping can also be provided; the default is an empty mapping. The key in the errors mapping will be used for the :type item in an ex-info exception; the value is the protocol-level identifier for the error. If you need to handle non-ex-info exceptions, you will need to catch these in your responder and rethrow as an ex-info exception.

Argument types

Argument types are specified by qualified keywords. All of the built-in argument types are defined in clj-amp.argument, so you can either address them as ::clj-amp.argument/some-type or import the namespace under an alias: for example, :require [clj-amp.argument] :as a and then ::a/some-type. This also makes it easier to define your own argument types without worrying about clashing with built-in types or custom types from elsewhere.

Argument encoding/decoding is handled by two sets of multimethods in clj-amp.argument: to-bytes / from-bytes (for "simple" types), and to-box / from-box (for "complex" types). Argument types for which the -bytes methods are defined can only be encoded as a single box item, which allows them to be used in composite structures such as ::list.

The "simple" types are as follows:

  • ::integer

An arbitrary-size integer; equivalent to Integer in the reference implementation.

WARNING: Currently represented solely as a Long, thus values outside of [Long.MIN_VALUE, Long.MAX_VALUE] cannot currently be handled.

  • ::string

    A sequence of Unicode codepoints; equivalent to Unicode in the reference implementation. Represented as a Clojure (Java) String.

  • ::byte-string

    A sequence of bytes; equivalent to String in the reference implementation. Represented as a Clojure (Java) String; see the earlier section for more information.

  • ::bytes

    A sequence of bytes; also equivalent to String in the reference implementation. Represented as the native Gloss representation (a sequence of byte buffers); this allows further use of the bytes (eg. writing them to a file) without incurring the cost of additional copying. You may also find and byte-streams/convert of use when working with these values. For encoding, any type that Gloss / byte-streams can convert to bytes is accepted.

  • ::float

    An IEEE 754 binary floating-point number; equivalent to Float in the reference implementation. Represented as a Clojure (Java) Double.

  • ::decimal

    A decimal floating-point number; equivalent to Decimal in the reference implementation. Represented as a Clojure (Java) BigDecimal.

  • ::date-time

    A timestamp in the form of date and time; equivalent to DateTime in the reference implementation. Represented as a clj-time date-time value.

  • ::list

    A list of AMP values of a particular type; equivalent to ListOf in the reference implementation. Represented as a Clojure vector. The element type is specified by a :of item in the argument map; for example, {:type ::a/list :of ::a/integer} for a list of integer values.

The "complex" types (which cannot be used in composites like ::list) are as follows:

  • ::clj-amp.bigbytes/bigbytes

    A sequence of bytes that is too large to fit into a single AMP item; the bytes will be split over multiple AMP items. Also known as BigString.

    WARNING: Use of this argument type may have adverse effects on the latency of concurrent requests on the same AMP connection. For more information, see the AMP website.

Servers, clients, and responders

Connection basics

The AMP protocol is symmetric in that both client and server can send and receive commands. Once you have an AMP box stream, call (clj-amp.core/amp-connection responder stream). This will return a pair of [call-remote close!]. You may then use (call-remote command arguments) to invoke a command (returns a Manifold deferred), and (close!) in order to close the stream. Calling the call-remote function after the stream is closed is an error.

responder is a function that will be called to handle an incoming command box; usually you will construct such a function with clj-amp.core/make-responder, passing a map from commands to responder functions. The function for a particular command will be called with a map of the (decoded) arguments, and should either return a response map, or a Manifold deferred (that will realize with the response).

If you do not need to handle any commands (for example, for a client that only intends to issue commands, not receive them), you should use (make-responder {}) in order to ensure that any commands that may be sent in error receive an appropriate error response.

Higher-level API

In order to make an AMP client connection over TCP, call (clj-amp.core/client host port responder) where responder is a responder function (as returned by clj-amp.core/make-responder).

In order to run an AMP TCP server, call (clj-amp.core/simple-server responder port).

Alternatively, if you need access to the client connection information, you can use (clj-amp.core/start-ampbox-server handler port); handler will be called with the AMP box stream and the client info as arguments, you may then call (clj-amp.core/amp-connection) to set up the connection.

Defining new argument types

The use of multimethods for handling argument encoding and decoding allows for extending the built-in set of argument types with your own, by defining your own methods.

There are three ways to define a new argument type:

  1. Using clj-amp.argument/defargument; this is the preferred way in most cases.

    defargument takes a type keyword and a Gloss codec used to encode/decode the argument value. For example, here is how the built-in ::boolean type is defined:

    (defargument ::boolean
       (g/string :ascii)
       #(if %1 "True" "False")
       (partial = "True")))

    See the Gloss documentation for more information about Gloss codecs. A qualified keyword should be used for the type, in order to avoid clashes.

  2. Defining from-bytes and to-bytes methods. This way is preferrable when encoding and decoding using a Gloss codec is not possible for some reason, but your argument can be represented as a single AMP box value.

    from-bytes is called with the argument map and the bytes to decode (in native Gloss format, ie. a sequence of byte buffers); it must return the decoded value. to-bytes is called with the argument map and the value to encode; it can return the encoded value in any form that Gloss can use as bytes.

  3. Defining from-box and to-box methods. This way is the most flexible, but does not allow the argument type to be used in composite types such as ::list, and thus should be avoided where possible.

    from-box is called with the argument map and the entire command box; it should return the decoded value. to-box is called with the argument map and the value to encode; it should return a map of box keys and byte values, which will be merged into the final command box.