git clone ''

(ql:quickload :smee.binary)


This library is a high performance binary parser combinator. It enables reading and writing arbitrary binary data from Java's io streams. The focus is on enabling parsing of externally defined binary structures. If you have a format specification for any binary structure, this library is for you!

It is inspired by Gloss but focuses on java's stream classes. The individual codecs do not require explicit knowledge about the length of data that needs to be read.

Build Status


Binary artifacts are released to Clojars. If you are using Maven, add the following repository definition to your pom.xml:


The Most Recent Release

With Leiningen:

[smee/binary "0.5.4"]

With Maven:



All functions given in this document refer to the namespace org.clojars.smee.binary.core (needs to be required or use in your namespace declaration).

Examples / Demo

Please refer to the tests for now. There are several demos:


To read binary data we need two things: A codec that knows how to read and write it's binary representation and convert it to a clojure data structure and an instance of The codec needs to satisfy the protocol BinaryIO (see here).

Codecs are composable, you may combine them as you like.

Each codec can have two attached functions:

Example: Let's represent an instance of java.util.Date as a unix epoch and write it as a little-endian long:

(compile-codec :long-le (fn [^java.util.Date date] (.getTime date)) (fn [^long number] (java.util.Date. number))

The compiler hints are not necessary. They are just a clarification in this example.


Features/Available codecs


Encodes primitive data types, either big-endian or little-endian:

; signed
; unsigned

Please be aware that since Java doesn't support unsigned data types the codecs will consume/produce a bigger data type than for the unsigned case: Unsinged bytes are shorts, unsigned shorts are integers, unsigned integers are longs, unsigned longs are Bigints!


If you want several codecs in a specific order, use a vector:

[:int-le :float-le :float-le]


To name elements in a binary data source maps are ideal. Unfortunately the order of the keys is unspecified. We need to use a map constructor that respects the order of the keys:

(require '[org.clojars.smee.binary.core :as b])
(b/ordered-map :foo :int-le :bar [:float-le :double-le])

As you can see arbitrary nesting of codecs is possible. You can define maps of maps of … If you use clojure's map literals, the order of the binary values is unspecified (it is determined by the sequence of keys and values within the map's implementation).


repeated uses another codec repeatedly until the stream is exhausted. To restrict, how often the codec should be used, you can explicitely give one of trhee parameters:

Caution: When writing the data there WILL be a final separator. This means, the written data may have more bytes than initially read!


blob is essentially an optimized version of (repeated :byte ...) that produces and consumes Java byte arrays. It takes the same options as repeated, except for :separator.


Reads and writes bytes and converts them from/to strings with a specific string encoding. This codec uses repeated, that means it takes either :length or :prefix as parameter to determine the length of the string.

(string "ISO-8859-1" :length 3) ; read three bytes, interpret them as a string with encoding "ISO-8859-1"

C strings

Similar to string, but reads bytes until it finds a null byte:

(c-string "UTF8") ; 


If you have a byte where each bit has a specific meaning you can use a set of keywords as an input. For example, the following definition says, that the lowest bit in a byte gets the value :a, the next one :b, then :c. The bits 4-7 are ignored, the highest bit has the value :last:

(decode (bits [:a :b :c nil nil nil nil :last]) instream); let's assume the next byte in instream is 2r11011010
=> #{:b :last}

If you now read a byte with the value 2r11011001 using this codec you will get the clojure set #{:a :b :last} as a value.


Decodes a header using header-codec. Passes this datastructure to header->body which returns the codec to use to parse the body. For writing this codec calls body->header with the data as parameter and expects a value to use for writing the header information.


Make sure there is always a minimum byte length when reading/writing values. Works by reading length bytes into a byte array, then reading from that array using inner-codec. Currently there are three options:


(padding (repeated :int-le :length 100) :length 1024 :padding-byte (byte \x))
=> [...] ; sequence of 100 integers, the stream will have 1024 bytes read, though

(encode (padding (repeated (string "UTF8" :separator 0)) :length 11 :truncate? true) outstream ["abc" "def" "ghi"])
=> ; writes bytes [97 98 99 0 100 101 102 0 103 104 105]
   ; observe: the last separator byte was truncated!


This codec is related to padding in that it makes sure that the number of bytes written/read to/from a stream always is aligned to a specified byte boundary. For example, if a format requires aligning all data to 8 byte boundaries this codec will pad the written data with padding-byte to make sure that the count of bytes written is divisable by 8.



(encode (align (repeated :short-be :length 3) :modulo 9 :padding-byte 55) [1 2 3] output-stream)
;==> writes these bytes: [0 1 0 2 0 3 55 55 55]


If a binary format uses fixed elements (like the three bytes ‘ID3’ in mp3), you can use this codec. It needs a codec and a fixed value. If the value read using this codec does not match the given fixed value, an exception will be thrown.

(constant (string "ISO-8859-1" :length 3) "ID3")

Alternatively, you may treat strings and byte arrays as constant encoders.


Union is a C-style union. A fixed number of bytes may represent different values depending on the interpretation of the bytes. The value returned by read-data is a map of all valid interpretations according to the specified unioned codecs. Parameter is the number of bytes needed for the longest codec in this union and a map of value names to codecs. This codec will read the specified number of bytes from the input streams and then successively try to read from this byte array using each individual codec.

Example: Four bytes may represent an integer, two shorts, four bytes, a list of bytes with prefix or a string.

(union 4 {:integer :int-be 
          :shorts (repeated :short-be :length 2)
          :bytes (repeated :byte :length 4)
          :prefixed (repeated :byte :prefix :byte)
          :str (string \"UTF8\" :prefix :byte)})


Copyright © 2014 Steffen Dienst

Distributed under the Eclipse Public License, the same as Clojure.