Sam Newman's site, a Consultant at ThoughtWorks

Archive for ‘June, 2010’

Updated to reflect some feedback and one example of using commons-exec as an alternative to the plain old Runtime.exec

Second Update to reflect use of shell-out – thanks Scott!


Making use of

(ns utils

(defn execute [command]
  (let [process (.exec (Runtime/getRuntime) command)]
    (if (= 0 (.waitFor  process))
        (read-lines (.getInputStream process))
        (read-lines (.getErrorStream process)))))

user=> (execute "ls")
("MyProject.iml" "lib" "out" "src" "test")

It could be improved obviously – for example catching some of the potential IOExceptions that can result to rethrow additional information, such as the command being executed, or the ability to take a seq of program arguments.

Error & Argument Handling

This version adds some basic (and ugly) exception handling, and also handles spacing out arguments passed in (so passing "ls" "-la" gets processed into "ls -la"):

(defn execute
  "Executes a command-line program, returning stdout if a zero return code, else the
  error out. Takes a list of strings which represent the command & arguments"
  [& args]
    (let [process (.exec (Runtime/getRuntime) (reduce str (interleave args (iterate str " "))))]
      (if (= 0 (.waitFor  process))
          (read-lines (.getInputStream process))
          (read-lines (.getErrorStream process))))
    (catch IOException ioe
      (throw (new RuntimeException (str "Cannot run" args) ioe)))))

Using commons-exec

I had some problems with hanging processes, so knocked up a version using Apache’s commons-exec. This version has the added advantage of killing long-running processes, and I folded in Steve’s suggestion for a better way of splicing in the spaces in the command line args (see his comment). commons-exec is part of the special sauce inside Ant, so is a rock solid way of launching command-line processes (well, as rock solid as Java gets).

The use of the ByteArrayOutputStream is probably inefficient, and again, decent error handling is left as an exercise to the reader.

(defn alternative-execute
  "Executes a command-line program, returning stdout if a zero return code, else the
  error out. Takes a list of strings which represent the command & arguments"
  [& args]
  (let [output-stream (new ByteArrayOutputStream)
        error-stream (new ByteArrayOutputStream)
        stream-handler (new PumpStreamHandler output-stream error-stream)
        executor (doto
                  (new DefaultExecutor)
                  (.setExitValue 0)
                  (.setStreamHandler stream-handler)
                  (.setWatchdog (new ExecuteWatchdog 20000)))]
     (if (= 0 (.execute executor (CommandLine/parse (apply str (interpose " " args)))))
       (.toString output-stream)
       (.toString error-stream))))


Many thanks to Scott for this. clojure.contrib supplies the very neat shell-out:

user=> (use '
user=> (sh "ls" "-la")

I haven’t probed further to see if this deals with my hanging process problem, but it certainly doesn’t seem to have any support for killing timeout processes. If you’re worried about runaway tasks, the commons-exec version above might be the right choice for you.

I’ve been playing around with partially applied functions in Clojure, and have hit an interesting snag when dealing with Java interop. First, lets examine what partial does in Clojure, by cribbing off an example from Stuart Halloway’s Programming Clojure:

user=> (defn add [one two] (+ one two))
user=> (add 1 2)
user=> (def increment-by-two (partial add 2))
user=> (increment-by-two 5)

What partial is doing is partially applying the function – in our case we have applied one of the two arguments our add implementation requires, and got back another function we can call later to pass in the second argument. This example is obviously rather trivial, but partially applied functions can be very handy in a number of situations.

Anyway, this wasn’t supposed to be a discussion of partial in general, but one problem I’ve hit with it when trying to partially apply a call to a Java static method. So, let’s implement our trivial add method in plain old Java:

public class Functions {
    public static int add(int first, int second) {
        return first + second;

Then try using partial as before:

user=> (import
user=> (Functions/add 1 2)
user=> (def increment-by-two (partial Functions/add 1))
java.lang.Exception: Unable to find static field: add in class (NO_SOURCE_FILE:3)

So it seems like the partial call can’t handle static calls in this situation. But what if I wrap the call in another function?

user=> (defn java-add [arg1 arg2] (Functions/add arg1 arg2))
user=> (def increment-by-two (partial java-add 2))
user=> (increment-by-two 10)

Which works. There is probably a reason why, but I can’t quite work it out right now.

Posted in the “I hope no-one else has to go through this” category in the hope that Google surfaces this for some other poor soul.

Picture a rather trivial split function:

(defn split [str delimiter]
  ((seq (.split str delimiter))))

Which helpfully spits out:

java.lang.ClassCastException: clojure.lang.ArraySeq cannot be cast to clojure.lang.IFn

The issue here is the additional set of parentheses – a hang-over from a previous edit. Removing this fixed the trouble. These parentheses were causing Clojure to expect a function call…