A Clojure wrapper around java.lang.ProcessBuilder
.
This library is included in babashka since 0.2.3 but is also intended as a JVM library:
$ clojure -Sdeps '{:deps {babashka/process {:mvn/version "0.0.1"}}}'
user=> (require '[clojure.string :as str])
nil
user=> (require '[babashka.process :refer [process check]])
nil
user=> (-> (process ["ls" "-la"] {:out :string}) check :out str/split-lines first)
"total 136776"
If clojure.java.shell
works for your purposes, keep using it. But there are
contexts in which you need more flexibility. The major differences compared with
this library:
sh
is blocking,process
makes blocking explicit viaderef
sh
focuses on convenience but limits what you can do with the underlying process,process
exposes as much as possible while still offering an ergonomic APIprocess
supports piping processes via->
orpipeline
sh
offers integration withclojure.java.io/copy
for:in
,process
extends this to:out
and:err
You will probably mostly need process
and check
so it would be good
to start reading the docs for these. Skim over the rest and come back when you
need it.
-
process
: takes a command (vector of strings or objects that will be turned into strings) and optionally a map of options.Returns: a record (called "the process" in this README) with
:proc
: an instance ofjava.lang.Process
:in
,:err
,:out
: the process's streams. To obtain a string from:out
or:err
you will typically useslurp
or use the:string
option (see below). Slurping those streams will block the current thread until the process is finished.:cmd
: the command that was passed to create the process.:prev
: previous process record in case of a pipeline.
The returned record can be passed to
deref
. Doing so will cause the current thread to block until the process is finished and will populate:exit
with the exit code.Supported options:
-
:in
,:out
,:err
: objects compatible withclojure.java.io/copy
that will be copied to or from the process's corresponding stream. May be set to:inherit
for redirecting to the parent process's corresponding stream. Optional:in-enc
,:out-enc
and:err-enc
values will be passed along toclojure.java.io/copy
.The
:out
and:err
options support:string
for writing to a string output. You will need toderef
the process before accessing the string via the process's:out
.For writing output to a file, you can set
:out
and:err
to ajava.io.File
object, or a keyword::write
+ an additional:out-file
/:err-file
+ file to write to the file.:append
+ an additional:out-file
/:err-file
+ file to append to the file.
-
:inherit
: if true, sets:in
,:out
and:err
to:inherit
. -
:dir
: working directory. -
:env
,:extra-env
: a map of environment variables. See Add environment. -
:escape
: function that will applied to each stringified argument. On Windows this defaults to prepending a backslash before a double quote. On other operating systems it defaults toidentity
. -
:shutdown
: shutdown hook, defaults tonil
. Takes process map. Typically used withdestroy
ordestroy-tree
to ensure long running processes are cleaned up on shutdown.
Piping can be achieved with the
->
macro:(-> (process '[echo hello]) (process '[cat]) :out slurp) ;;=> "hello\n"
or using the
pipeline
function (see below) -
check
: takes a process, waits until is finished and throws if exit code is non-zero. -
$
: convenience macro aroundprocess
. Takes command as varargs. Options can be passed via metadata on the form or as a first map arg. Supports interpolation via~
. -
sh
: convenience function similar toclojure.java.shell/sh
that sets:out
and:err
to:string
by default and blocks. Similar tocjs/sh
it does not check the exit code (this can be done withcheck
). -
*defaults*
: dynamic var containing overridable default options. Usealter-var-root
to change permanently orbinding
to change temporarily. -
destroy
: function of process or map with:proc
(java.lang.ProcessBuilder
). Destroys the process and returns the input arg. -
destroy-tree
: same asdestroy
but also destroys all descendants. JDK9+ only. -
pb
: returns a process builder (as record). -
start
: takes a process builder, calls start and returns a process (as record). -
pipeline
:- When passing a process, returns a vector of processes of a pipeline created with
->
orpipeline
. - When passing two or more process builders created with
pb
: creates a pipeline as a vector of processes (JDK9+ only).
Also see Pipelines.
- When passing a process, returns a vector of processes of a pipeline created with
user=> (require '[babashka.process :refer [process check sh pipeline pb]])
Invoke ls
:
user=> (-> (process '[ls]) :out slurp)
"LICENSE\nREADME.md\nsrc\n"
Change working directory:
user=> (-> (process '[ls] {:dir "test/babashka"}) :out slurp)
"process_test.clj\n"
Set the process environment.
user=> (-> (process '[sh -c "echo $FOO"] {:env {:FOO "BAR"}}) :out slurp)
"BAR\n"
The return value of process
implements clojure.lang.IDeref
. When
dereferenced, it will wait for the process to finish and will add the :exit
value:
user=> (-> @(process '[ls foo]) :exit)
1
The function check
takes a process, waits for it to finish and returns it. When
the exit code is non-zero, it will throw.
user=> (-> (process '[ls foo]) check :out slurp)
Execution error (ExceptionInfo) at babashka.process/check (process.clj:74).
ls: foo: No such file or directory
Redirect output to stdout:
user=> (do @(process '[ls] {:out :inherit}) nil)
LICENSE README.md deps.edn src test
nil
Both :in
, :out
may contain objects that are compatible with clojure.java.io/copy
:
user=> (with-out-str (check (process '[cat] {:in "foo" :out *out*})))
"foo"
user=> (with-out-str (check (process '[ls] {:out *out*})))
"LICENSE\nREADME.md\ndeps.edn\nsrc\ntest\n"
The :out
option also supports :string
. You will need to deref
the process
in order for the string to be there:
user=> (-> @(process '[ls] {:out :string}) :out)
"LICENSE\nREADME.md\ndeps.edn\nsrc\ntest\n"
Redirect output stream from one process to input stream of the next process:
user=> (let [is (-> (process '[ls]) :out)]
@(process ["cat"] {:in is
:out :inherit})
nil)
LICENSE
README.md
deps.edn
src
test
nil
Forwarding the output of a process as the input of another process can also be done with thread-first:
user=> (-> (process '[ls])
(process '[grep "README"]) :out slurp)
"README.md\n"
To write to a file use :out :write
and set :out-file
to a file:
user=> (require '[clojure.java.io :as io])
nil
user=> (do @(p/process ["ls"] {:out :write :out-file (io/file "/tmp/out.txt")}) nil)
nil
user=> (slurp "/tmp/out.txt")
"CHANGELOG.md\nLICENSE\nREADME.md\ndeps.edn\nproject.clj\nscript\nsrc\ntest\n"
To append to a file, use :out :append
:
user=> (do @(p/process ["ls"] {:out :append :out-file (io/file "/tmp/out.txt")}) nil)
nil
user=> (slurp "/tmp/out.txt")
"CHANGELOG.md\nLICENSE\nREADME.md\ndeps.edn\nproject.clj\nscript\nsrc\ntest\nCHANGELOG.md\nLICENSE\nREADME.md\ndeps.edn\nproject.clj\nscript\nsrc\ntest\n"
Here is an example of a cat
process to which we send input while the process
is running, then close stdin and read the output of cat afterwards:
(ns cat-demo
(:require [babashka.process :refer [process]]
[clojure.java.io :as io]))
(def catp (process '[cat]))
(.isAlive (:proc catp)) ;; true
(def stdin (io/writer (:in catp)))
(binding [*out* stdin]
(println "hello"))
(.close stdin)
(slurp (:out catp)) ;; "hello\n"
(def exit (:exit @catp)) ;; 0
(.isAlive (:proc catp)) ;; false
Here is an example where we read the output of yes
line by line and print it ourselves:
(require '[babashka.process :as p :refer [process]]
'[clojure.java.io :as io])
(def yes (process ["yes"] {:err :inherit
:shutdown p/destroy}))
(with-open [rdr (io/reader (:out yes))]
(binding [*in* rdr]
(loop []
(let [line (read-line)]
(println :line line))
(recur))))
$
is a convenience macro around process
:
user=> (def config {:output {:format :edn}})
#'user/config
user=> (-> ($ clj-kondo --config ~config --lint "src") deref :out slurp edn/read-string)
{:findings [], :summary {:error 0, :warning 0, :info 0, :type :summary, :duration 34}}
sh
is a convenience function around process
which sets :out
and :err
to
:string
and blocks automatically, similar to clojure.java.shell/sh
:
user=> (def config {:output {:format :edn}})
#'user/config
user=> (-> (sh ["clj-kondo" "--lint" "src"]) :out slurp edn/read-string)
{:findings [], :summary {:error 0, :warning 0, :info 0, :type :summary, :duration 34}}
Both process
and sh
support tokenization when passed a single string argument:
user=> (-> (sh "echo hello there") :out)
"hello there\n"
user=> (-> (sh "clj-kondo --lint -" {:in "(inc)"}) :out print)
<stdin>:1:1: error: clojure.core/inc is called with 0 args but expects 1
linting took 11ms, errors: 1, warnings: 0
Note that check
will wait for the process to end in order to check the exit
code. When the process has lots of data to write to stdout, it is recommended to
add an explicit :out
option to prevent deadlock due to buffering. This example
will deadlock because the process is buffering the output stream but it's not
being consumed, so the process won't be able to finish:
user=> (-> (process ["cat"] {:in (slurp "https://datahub.io/datahq/1mb-test/r/1mb-test.csv")}) check :out slurp count)
The way to deal with this is providing an explicit :out
option so the process
can finish writing its output:
user=> (-> (process ["cat"] {:in (slurp "https://datahub.io/datahq/1mb-test/r/1mb-test.csv") :out :string}) check :out count)
1043005
The :env
option replaces your entire environment with the provided map. To add environment variables you can use :extra-env
instead:
:extra-env {"FOO" "BAR"}
The pipeline
function returns a
sequential
of processes from a process that was created with ->
or by passing multiple
objects created with pb
:
user=> (mapv :cmd (pipeline (-> (process '[ls]) (process '[cat]))))
[["ls"] ["cat"]]
user=> (mapv :cmd (pipeline (pb '[ls]) (pb '[cat])))
[["ls"] ["cat"]]
To obtain the right-most process from the pipeline, use last
(or peek
):
user=> (-> (pipeline (pb ["ls"]) (pb ["cat"])) last :out slurp)
"LICENSE\nREADME.md\ndeps.edn\nsrc\ntest\n"
Calling pipeline
on the right-most process returns the pipeline:
user=> (def p (pipeline (pb ["ls"]) (pb ["cat"])))
#'user/p
user=> (= p (pipeline (last p)))
true
To check an entire pipeline for non-zero exit codes, you can use:
user=> (run! check (pipeline (-> (process '[ls "foo"]) (process '[cat]))))
Execution error (ExceptionInfo) at babashka.process/check (process.clj:37).
ls: foo: No such file or directory
Although you can create pipelines with ->
, for some applications it may be
preferable to create a pipeline with pipeline
which defers to
ProcessBuilder/startPipeline
. In the following case it takes a long time
before you would see any output due to buffering.
(future
(loop []
(spit "log.txt" (str (rand-int 10) "\n") :append true)
(Thread/sleep 10)
(recur)))
(-> (process '[tail -f "log.txt"])
(process '[cat])
(process '[grep "5"] {:out :inherit}))
The solution then it to use pipeline
+ pb
:
(pipeline (pb '[tail -f "log.txt"])
(pb '[cat])
(pb '[grep "5"] {:out :inherit}))
The varargs arity of pipeline
is only available in JDK9 or higher due to the
availability of ProcessBuilder/startPipeline
. If you are on JDK8 or lower, the
following solution that reads the output of tail
line by line may work for
you:
(def tail (process '[tail -f "log.txt"] {:err :inherit}))
(def cat-and-grep
(-> (process '[cat] {:err :inherit})
(process '[grep "5"] {:out :inherit
:err :inherit})))
(binding [*in* (io/reader (:out tail))
*out* (io/writer (:in cat-and-grep))]
(loop []
(when-let [x (read-line)]
(println x)
(recur))))
Another solution is to let bash handle the pipes by shelling out with bash -c
.
Because process
spawns threads for non-blocking I/O, you might have to run
(shutdown-agents)
at the end of your Clojure JVM scripts to force
termination. Babashka does this automatically.
When pretty-printing a process, you will get an exception:
(require '[clojure.pprint :as pprint])
(pprint/pprint (process ["ls"]))
Execution error (IllegalArgumentException) at user/eval257 (REPL:1).
Multiple methods in multimethod 'simple-dispatch' match dispatch value: class babashka.process.Process -> interface clojure.lang.IDeref and interface clojure.lang.IPersistentMap, and neither is preferred
The reason is that a process is both a record and a clojure.lang.IDeref
and
pprint does not have a preference for how to print this. This can be resolved
using:
(prefer-method pprint/simple-dispatch clojure.lang.IPersistentMap clojure.lang.IDeref)
Copyright © 2020-2021 Michiel Borkent
Distributed under the EPL License. See LICENSE.