A monadic stream library in Clojure (port of Haskell's enumerator).
Switch branches/tags
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.


river Build Status

river tries to provide an implementation of Oleg's Iteratee library. The terminology was changed in order to offer a more intuitive API for the end user, however if you are familiar with Haskell's Iteratees, this might help you get started with this library:

  • Iteratee will be called Consumer
  • Enumerator will be called Producer
  • Enumeratee will be called Filter

The library is initally divided in 3 namespaces

  • river.core provides:

    • Basic constructs to build your own consumers/producers/filters.
    • A monad implemenation so that you can build monadic consumers.
    • Basic consumers for debugging streams.
    • A macro to easily execute the stream generation, composition and execution.
  • river.seq provides:

    • All the expected combinators that are available for lazy seqs.
    • Handy filters for the streams like map\*, filter\*, etc.
  • river.io provides:

    • Ways to produce binary (byte) streams out of Java's InputStreams.
    • Produce line streams out of Java's InputStreams.
    • Produce input out of files (binary or lines).
    • Produce input out of commands (binary or lines).

The code provided in this library is still beta, there is a lot of room for improvement and new features, it serves as a starting point to exploit the Iteratee concepts in Clojure.


[org.van-clj/river "0.1.0"]


To execute a consumer you will use the run function:

(river.core/run (river.seq/produce-seq (range 1 100))
                (river.core/*c (river.seq/filter* #(= 0 (mod % 2)))
                               (river.seq/take 5)))

; => #river.core/ConsumerDome {:result (2 4 6 8 10), :remainder (12 14)}

The code above streams a seq from 1 to 99, then calls \*c to bind a filter to a consumer that takes 5 items from the feed. The result of this execution will be a ConsumerDone record that has the yielded result and the remainder of the given chunks.


The stream is what the consumer receive as an input, the stream could either be a seq of items (called chunks), or an EOF signal, represented by river.core/eof.


The consumer is the one that process the stream, and it will either yield a result (using river.core/yield) or a continuation (using river.core/continue).

river.core/yield will be used when the consumer is done consuming from the stream, two values are returned with the yield, one being the result value, and the second being the remainder of the stream that wasn't consumed, this is kept in order to compose several consumers together using the monadic interface.

river.core/continue will be used when the consumer hasn't received enough chunks to yield a result, some consumers might consume part of the stream, some others would be greedy and consume all the available stream.


The producer generates the stream that the consumer will use, they normally consume a resource like a lazy-seq, file, socket, etc; and then transmits it over to the consumer. The advantage of using producers is that the stream generation is kept independent from the consumption, so the consumer doesn't need to know where the data is coming from as long as it is valid.

It stops consuming from the given resource as soon as the consumer returns a yield value.


The filter transforms the stream into something different, it either changes the type of the stream, or modifies the way the input is given.

One of the powerful characteristics of this stream architecture is that both Producers and Filters are also consumers, they receive a consumer as a parameter and they enhance the consumer behavior, they could be perceived as some sort of decorator in the OO world.


Building filters on the fly

Say for example you want to be able to sum a list of numbers, but this numbers may come from different resources, some from stdin, others from a list, etc. To implement this using the river library, you would do something like the following:

(ns river.examples.sum

  (require [clojure.string :as string])

  (use [river.core])
  (require [river.seq :as rs]
           [river.io  :as rio]))

(def words* (rs/mapcat* #(string/split % #"\s+")))
; ^ a filter that applies a split to whatever it recieves
; in the stream, this is assuming the stream is of strings

(def numbers* (rs/map* #(Integer/parseInt %)))
; ^ a fitler that transforms each received item into an
; Integer using the parseInt function, this is assuming
; that the stream is of strings

(defn produce-numbers-from-file
  ([] (produce-numbers-from-file "input.in"))
    ; ^ whe bind a producer with some filters
    ; using the p* function.
        (rio/produce-file-lines file-path)
        ; ^ produces a stream of lines from a file path
        ; ^ applies the words filter
        ; ^ applies the number filter and transform
        ; the stream from strings to numbers

(defn -main []
  (println (run 
           ; ^ executes a group of producers/consumers
                ; ^ produce a stream of numbers from a file
                (rs/produce-seq (range 1 10))
                ; ^ produce a stream of numbers from a seq
                (rs/reduce + 0))))
                ; ^ sums up the numbers ignoring where they come from

Building consumers using monadic interface

Sometimes we want to create a new consumer composing simple consumers together, we can do that using the monadic API:

(ns river.examples.monadic
  (:use [river.core])
  (:require [river.seq :as rs]
            [river.io  :as rio]))

(def drop-and-head
  ; ^ allows you to create a new consumer using
  ; a monadic notation as in clojure.algo.monads
    [_ (rs/drop-while #(< % 5))
       ; ^ drops from the stream until the condition is met
     b rs/first]
       ; ^ gets the first element after the dropping is done

(defn -main []
  (println (run
           ; ^ function to execute producers/consumers
                (rs/produce-seq (range -20 20))
                ; ^ produce a stream of numbers from a seq
                ; drops until < 5 and then gives the first element found
                ; (in this case 6)


Copyright (C) 2012 Roman Gonzalez

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