A flexible library for shelling out in Clojure
Clojure Shell
Switch branches/tags
Nothing to show
Latest commit 3acbfe5 Oct 26, 2016 @Raynes Whitespace cleanup



Build Status

Conch is actually two libraries. The first, me.raynes.conch.low-level is a simple low-level interface to the Java process APIs. The second and more interesting library is an interface to low-level inspired by the Python sh library.

The general idea is to be able to call programs just like you would call Clojure functions. See Usage for examples.


In Leiningen:



First of all, let's require a few things.

user> (require '[me.raynes.conch :refer [programs with-programs let-programs] :as sh])

Let's call a program. What should we call? Let's call echo, because I obviously like to hear myself talk.

user> (programs echo)
user> (echo "Hi!")

Cool! programs is a simple macro that takes a number of symbols and creates functions that calls programs on the PATH with those names. echo is now just a normal function defined with defn just like any other.

with-programs and let-programs are lexical and specialized versions of programs. with-programs is exactly the same as programs, only it defines functions lexically:

user> (with-programs [ls] (ls))
user> ls
CompilerException java.lang.RuntimeException: Unable to resolve symbol: ls in
this context, compiling:(NO_SOURCE_PATH:1) 

let-programs is similar, but is useful for when you want to specify a path to a program that is not on the PATH:

user> (let-programs [echo "/bin/echo"] (echo "hi!"))

Bad example since echo is on the path, but if it wasn't there already it still would have worked. I promise.


You can pass input to a program easily:

user> (programs cat)
user> (cat {:in "hi"})
user> (cat {:in ["hi" "there"]})
user> (cat {:in (java.io.StringReader. "hi")})

:in is handled by a protocol and can thus be extended to support other data.


So, going back to our ls example. Of course, ls gives us a bunch of lines. In a lot of cases, we're going to want to process lines individually. We can do that by telling conch that we want a lazy seq of lines instead of a monolithic string:

user> (ls {:seq true})
("#README.md#" "README.md" "classes" "docs" "foo.py" "lib" "lol" "pom.xml"
"pom.xml.asc" "project.clj" "src" "target" "test" "testfile")

We can also redirect output to other places.

user> (let [writer (java.io.StringWriter.)] (echo "foo" {:out writer}) (str writer))
user> (echo "foo" {:out (java.io.File. "conch")})
user> (slurp "conch")

And if that wasn't cool enough for you, :out is handled by a protocol and thus can be extended.


Need the exit code and stuff? Sure:

user> (echo "foo" {:verbose true})
{:proc {:out ("foo\n"), :in #<ProcessPipeOutputStream
java.lang.UNIXProcess$ProcessPipeOutputStream@19c12ee7>, :err (), :process
#<UNIXProcess java.lang.UNIXProcess@2bfabe2a>}, :exit-code
#<core$future_call$reify__6110@5adacdf4: 0>, :stdout "foo\n", :stderr ""}


user> (sleep "5")
... yawn ...

Tired of waiting for that pesky process to exit? Make it go away!

user> (sleep "5" {:timeout 2000})
... two seconds later ...
ExceptionInfo Program returned non-zero exit code :timeout  clojure.core/ex-info (core.clj:4227)

Much better.


Conch can handle exit codes pretty well. You can make it do pretty much whatever you want in failure scenarios.

By default, conch throws ExceptionInfo exceptions for non-zero exit codes, as demonstrated here:

user> (ls "-2")
ExceptionInfo Program returned non-zero exit code 1  clojure.core/ex-info

This exception's data is the same result you'd get by passing the :verbose option:

user> (ex-data *e)
{:proc {:out (), :in #<ProcessPipeOutputStream java.lang.UNIXProcess$ProcessPipeOutputStream@79bb65ee>, :err ("ls: illegal
option -- 2\nusage: ls [-ABCFGHLOPRSTUWabcdefghiklmnopqrstuwx1] [file ...]\n"), :process #<UNIXProcess java.lang.UNIXProcess@1172897f>}, :exit-code #<core$future_call$reify__6110@76a0f9cb: 1>, :stdout "", :stderr "ls: illegal option -- 2\nusage: ls [-ABCFGHLOPRSTUWabcdefghiklmnopqrstuwx1] [file ...]\n"}

You can control this behavior in two ways. The first way is to set me.raynes.conch/*throw* to false:

user> (binding [sh/*throw* false] (ls "-2"))

You can also just override whatever *throw* is with the :throw argument to the functions themselves:

user> (ls "-2" {:throw false})


You can pipe the output of one program as the input to another about how you'd expect:

user> (programs grep ps)
user> (grep "ssh" {:in (ps "-e" {:seq true})})
" 4554 ??         0:00.77 /usr/bin/ssh-agent -l\n"

These functions also look for a lazy seq arg, so you can get rid of the explicit :in part.

user> (programs grep ps)
user> (grep "ssh" (ps "-e" {:seq true}))
" 4554 ??         0:00.77 /usr/bin/ssh-agent -l\n"


Conch gets rid of some ugly edge-cases by always reading process output immediately when it becomes available. It buffers this data into a queue that you consume however you want. This is how returning lazy seqs work. Keep in mind that if you don't consume data, it is being held in memory.

You can change how conch buffers data using the :buffer key.

user> (ls {:seq true :buffer 5})
("#READ" "ME.md" "#\nREA" "DME.m" "d\ncla" "sses\n" "conch" "\ndocs" "\nfoo." "py\nli" "b\nlol" "\npom." "xml\np" "om.xm" "l.asc" "\nproj" "ect.c" "lj\nsr" "c\ntar" "get\nt" "est\nt" "estfi" "le\n")
user> (ls {:seq true :buffer :none})
(\# \R \E \A \D \M \E \. \m \d \# \newline \R \E \A \D \M \E \. \m \d \newline
\c \l \a \s \s \e \s \newline \c \o \n \c \h \newline \d \o \c \s \newline \f \o
\o \. \p \y \newline \l \i \b \newline \l \o \l \newline \p \o \m \. \x \m \l
\newline \p \o \m \. \x \m \l \. \a \s \c \newline \p \r \o \j \e \c \t \. \c \l
\j \newline \s \r \c \newline \t \a \r \g \e \t \newline \t \e \s \t \newline \t
\e \s \t \f \i \l \e \newline)

Another nice thing gained by the way conch consumes data is that it is able to kill a process after a timeout and keep whatever data it has already consumed.

PTY stuff

PTY stuff seems like it'd be a lot of work and would involve non-Clojure stuff. If you need a PTY for output, I suggest wrapping your programs in 'unbuffer' from the expect package. It usually does the trick for unbuffering program output by making it think a terminal is talking to it.


You might run into an issue where your program finishes after using conch but does not exit. Conch uses futures under the hood which spin off threads that stick around for a minute or so after everything else is done. This is an unfortunate side effect, but futures are necessary to conch's functionality so I'm not sure there is much I can do about it.

You can work around this by adding a (System/exit 0) call to the end of your program.

Low Level Usage

The low-level API is available in a separate package:

(use '[me.raynes.conch.low-level :as sh])

It is pretty simple. You spin off a process with proc.

user=> (def p (sh/proc "cat"))
user=> p
{:out #<BufferedInputStream java.io.BufferedInputStream@5809fdee>, :in #<BufferedOutputStream java.io.BufferedOutputStream@5ebbfa5c>, :err #<DeferredCloseInputStream java.lang.UNIXProcess$DeferredCloseInputStream@58e5f46e>, :process #<UNIXProcess java.lang.UNIXProcess@18f42160>}

When you create a process with proc, you get back a map containing the keys :out, :err, :in, and :process.

  • :out is the process's stdout.
  • :err is the process's stderr.
  • :in is the process's stdin.
  • :process is the process object itself.

Conch is more flexible than clojure.java.shell because you have direct access to all of the streams and the process object itself.

So, now we have a cat process. This is a unix tool. If you run cat with no arguments, it echos whatever you type in. This makes it perfect for testing input and output.

Conch defines a few utility functions for streaming output and feeding input. Since we want to make sure that our input is going to the right place, let's set up a way to see the output of our process in realtime:

user=> (future (sh/stream-to-out p :out))
#<core$future_call$reify__5684@77b5c22f: :pending>

The stream-to-out function takes a process and either :out or :err and streams that to System/out. In this case, it has the effect of printing everything we pipe into our cat process, since our cat process just outputs whatever we input.

user=> (sh/feed-from-string p "foo\n")

The feed-from-string function just feeds a string to the process. It automatically flushes (which is why this prints immediately) but you can stop it from doing that by passing :flush false.

I think our cat process has lived long enough. Let's kill it and get its exit code. We can use the exit-code function to get the exit code. However, since exit-code stops the thread and waits for the process to terminate, we should run it in a future until we actually destroy the process.

user=> (def exit (future (sh/exit-code p)))

Now let's kill. R.I.P process.

user=> (sh/destroy p)

And the exit code, which we should be able to obtain now that the process has been terminated:

user=> @exit

Awesome! Let's go back to proc and see what else we can do with it. We can pass multiple strings to proc. The first string will be considered the executable and the rest of them the arguments to that executable.

user=> (sh/proc "ls" "-l")
{:out #<BufferedInputStream java.io.BufferedInputStream@7fb6634c>, :in #<BufferedOutputStream java.io.BufferedOutputStream@1f315415>, :err #<DeferredCloseInputStream java.lang.UNIXProcess$DeferredCloseInputStream@5f873eb2>, :process #<UNIXProcess java.lang.UNIXProcess@2825491d>}


(require '[me.raynes.conch.low-level :as sh])

Here is an easy way to get the output of a one-off process like this as a string:

user=> (sh/stream-to-string (sh/proc "ls" "-l") :out)
"total 16\n-rw-r--r--  1 anthony  staff  2545 Jan 24 16:37 README.md\ndrwxr-xr-x  2 anthony  staff    68 Jan 19 19:23 classes\ndrwxr-xr-x  3 anthony  staff   102 Jan 19 19:23 lib\n-rw-r--r--  1 anthony  staff   120 Jan 20 14:45 project.clj\ndrwxr-xr-x  3 anthony  staff   102 Jan 20 14:45 src\ndrwxr-xr-x  3 anthony  staff   102 Jan 19 16:36 test\n"

Let's print that for readability:

user=> (print (sh/stream-to-string (sh/proc "ls" "-l") :out))
total 16
-rw-r--r--  1 anthony  staff  2545 Jan 24 16:37 README.md
drwxr-xr-x  2 anthony  staff    68 Jan 19 19:23 classes
drwxr-xr-x  3 anthony  staff   102 Jan 19 19:23 lib
-rw-r--r--  1 anthony  staff   120 Jan 20 14:45 project.clj
drwxr-xr-x  3 anthony  staff   102 Jan 20 14:45 src
drwxr-xr-x  3 anthony  staff   102 Jan 19 16:36 test

So, that's the ls of the current directory. I ran this REPL in the conch project directory. We can, of course, pass a directory to ls to get it to list the files in that directory, but that isn't any fun. We can pass a directory to proc itself and it'll run in the context of that directory.

user=> (print (sh/stream-to-string (sh/proc "ls" "-l" :dir "lib/") :out))
total 6624
-rw-r--r--  1 anthony  staff  3390414 Jan 19 19:23 clojure-1.3.0.jar

You can also pass a java.io.File or anything that can be passed to clojure.java.io/file.

We can also set environment variables:

user=> (print (sh/stream-to-string (sh/proc "env" :env {"FOO" "bar"}) :out))

The map passed to :env completely replaces any other environment variables that were in place.

Finally, there a couple of low-level functions for streaming from and feeding to a process. They are stream-to and feed-from. These functions are what the utility functions are built off of, and you can probably use them to stream to and feed from your own special places.

You might want to fire off a program that listens for input until EOF. In these cases, you can feed it data for as long as you want and just tell it when you are done. Let's use pygmentize as an example:

user=> (def proc (sh/proc "pygmentize" "-fhtml" "-lclojure"))
user=> (sh/feed-from-string proc "(+ 3 3)")
user=> (sh/done proc)
user=> (sh/stream-to-string proc :out)
"<div class=\"highlight\"><pre><span class=\"p\">(</span><span class=\"nb\">+ </span><span class=\"mi\">3</span> <span class=\"mi\">3</span><span class=\"p\">)</span>\n</pre></div>\n"

When we call done, it closes the process's output stream which is like sending EOF. The process processes its input and then puts it on its input stream where we read it with stream-to-string.

Other options

All of conch's streaming and feeding functions (including the lower level ones) pass all of their keyword options to clojure.java.io/copy. It can take an :encoding and :buffer-size option. Guess what they do.

Key names

You might notice that the map that proc returns is mapped like so:

  • :in -> output stream
  • :out -> input stream

I did this because swapping them feels counterintuitive. The output stream is what you put :in to and the input stream is what you pull :out from.


Copyright (C) 2012 Anthony Grimes

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