Go has excellent support for executing external programs. Let’s start at the beginning.
In our examples we’ll be running ls -lah
command as it produces an output. There is no ls
on Windows so you can change that to e.g. tasklist
.
Running a command
cmd := exec.Command("ls", "-lah")
if runtime.GOOS == "windows" {
cmd = exec.Command("tasklist")
}
err := cmd.Run()
if err != nil {
log.Fatalf("cmd.Run() failed with %s\n", err)
}
If you run it, nothing seems to happen. Fear not, the command has actually been executed.
If we were running ls -lah
in the shell, the shell would copy programs’ stdout and stderr to console, so that we can see it.
We’re executing the program via Go standard library function and by default stdout and stderr are discarded.
Running a command and showing output
To let the human see the output, we can connect the output (cmd.Stdout
and cmd.Stderr
) of the program we’re executing to os.Stdout
and os.Stderr
, which is the output of our program:
cmd := exec.Command("ls", "-lah")
cmd.Stdout = os.Stdout
cmd.Stderr = os.Stderr
err := cmd.Run()
if err != nil {
log.Fatalf("cmd.Run() failed with %s\n", err)
}
cmd.Stdout
and cmd.Stderr
are declared as io.Writer
interface so we can set them to any type that implements Write()
method, like os.File
or an in-memory buffer bytes.Buffer
.
io.Reader
and io.Writer
are very simple, yet very powerful, abstractions.
Running a command and capturing the output
The above examples allows human to see the output but sometimes we want to capture the output and analyze it:
func main() {
cmd := exec.Command("ls", "-lah")
out, err := cmd.CombinedOutput()
if err != nil {
log.Fatalf("cmd.Run() failed with %s\n", err)
}
fmt.Printf("combined out:\n%s\n", string(out))
}
CombinedOutput
runs a command and returns combined stdout and stderr.
Behind the scenes of CombinedOutput
The good thing about Go is that it’s open source so we can peek at how a given functionality is implemented.
Another good thing is that most of the code in the standard library is simple. Here’s how CombinedOutput
is implemented:
func (c *Cmd) CombinedOutput() ([]byte, error) {
if c.Stdout != nil {
return nil, errors.New("exec: Stdout already set")
}
if c.Stderr != nil {
return nil, errors.New("exec: Stderr already set")
}
var b bytes.Buffer
c.Stdout = &b
c.Stderr = &b
err := c.Run()
return b.Bytes(), err
}
Notice that it’s almost as simple as our second example. Instead of setting cmd.Stdout
and cmd.Stderr
to standard output, we set them to a single in-memory buffer.
When program finishes, we returned everything written to that buffer.
Don’t be afraid to peruse the code of standard library.
Capture stdout and stderr separately
What if you want to do the same but capture stdout and stderr separately?
func main() {
cmd := exec.Command("ls", "-lah")
var stdout, stderr bytes.Buffer
cmd.Stdout = &stdout
cmd.Stderr = &stderr
err := cmd.Run()
if err != nil {
log.Fatalf("cmd.Run() failed with %s\n", err)
}
outStr, errStr := string(stdout.Bytes()), string(stderr.Bytes())
fmt.Printf("out:\n%s\nerr:\n%s\n", outStr, errStr)
}
Capture output but also show progress
What if the command takes a long time to finish?
It would be nice to see its progress on the console as it happens in addition to capturing stdout/stderr.
It’s a little bit more involved, but not terribly so.
First, a helper function that copies from reader to a writer and also captures copied data:
func copyAndCapture(w io.Writer, r io.Reader) ([]byte, error) {
var out []byte
buf := make([]byte, 1024, 1024)
for {
n, err := r.Read(buf[:])
if n > 0 {
d := buf[:n]
out = append(out, d...)
_, err := w.Write(d)
if err != nil {
return out, err
}
}
if err != nil {
// Read returns io.EOF at the end of file, which is not an error for us
if err == io.EOF {
err = nil
}
return out, err
}
}
}
Handling errors from Read
is subtle. An error io.EOF
means that we’ve read everything. It’s not an actual error so we turn io.EOF
into nil
.
The meat of the code:
func main() {
cmd := exec.Command("ls", "-lah")
var stdout, stderr []byte
var errStdout, errStderr error
stdoutIn, _ := cmd.StdoutPipe()
stderrIn, _ := cmd.StderrPipe()
err := cmd.Start()
if err != nil {
log.Fatalf("cmd.Start() failed with '%s'\n", err)
}
// cmd.Wait() should be called only after we finish reading
// from stdoutIn and stderrIn.
// wg ensures that we finish
var wg sync.WaitGroup
wg.Add(1)
go func() {
stdout, errStdout = copyAndCapture(os.Stdout, stdoutIn)
wg.Done()
}()
stderr, errStderr = copyAndCapture(os.Stderr, stderrIn)
wg.Wait()
err = cmd.Wait()
if err != nil {
log.Fatalf("cmd.Run() failed with %s\n", err)
}
if errStdout != nil || errStderr != nil {
log.Fatal("failed to capture stdout or stderr\n")
}
outStr, errStr := string(stdout), string(stderr)
fmt.Printf("\nout:\n%s\nerr:\n%s\n", outStr, errStr)
}
We have two outputs to copy. To avoid serializing them, we’ll read one in a goroutine.
As documentation of StdoutPipe
warns, Wait
will close the pipes when the process finishes. This might lead to losing some output if we haven’t finished reading it.
To prevent that we use sync.WaitGroup
to ensure that the gorutine handling os.Stdout
finishes reading before we call cmd.Wait
.
I encourage you to read the implementation of cmd.StdoutPipe
. You’ll be surprised by how short it is.
Capture output but also show progress #2
Previous solution works but copyAndCapture
looks like we’re re-implementing io.Copy
. Thanks to Go’s use of interfaces we can re-use io.Copy
.
We’ll write CapturingPassThroughWriter
struct implementing io.Writer
interface. It’ll capture everything that’s written to it and also write it to underlying io.Writer
.
// CapturingPassThroughWriter is a writer that remembers
// data written to it and passes it to w
type CapturingPassThroughWriter struct {
buf bytes.Buffer
w io.Writer
}
// NewCapturingPassThroughWriter creates new CapturingPassThroughWriter
func NewCapturingPassThroughWriter(w io.Writer) *CapturingPassThroughWriter {
return &CapturingPassThroughWriter{
w: w,
}
}
func (w *CapturingPassThroughWriter) Write(d []byte) (int, error) {
w.buf.Write(d)
return w.w.Write(d)
}
// Bytes returns bytes written to the writer
func (w *CapturingPassThroughWriter) Bytes() []byte {
return w.buf.Bytes()
}
func main() {
cmd := exec.Command("ls", "-lah")
var errStdout, errStderr error
stdoutIn, _ := cmd.StdoutPipe()
stderrIn, _ := cmd.StderrPipe()
stdout := NewCapturingPassThroughWriter(os.Stdout)
stderr := NewCapturingPassThroughWriter(os.Stderr)
err := cmd.Start()
if err != nil {
log.Fatalf("cmd.Start() failed with '%s'\n", err)
}
var wg sync.WaitGroup
wg.Add(1)
go func() {
_, errStdout = io.Copy(stdout, stdoutIn)
wg.Done()
}()
_, errStderr = io.Copy(stderr, stderrIn)
wg.Wait()
err = cmd.Wait()
if err != nil {
log.Fatalf("cmd.Run() failed with %s\n", err)
}
if errStdout != nil || errStderr != nil {
log.Fatalf("failed to capture stdout or stderr\n")
}
outStr, errStr := string(stdout.Bytes()), string(stderr.Bytes())
fmt.Printf("\nout:\n%s\nerr:\n%s\n", outStr, errStr)
}
Capture output but also show progress #3
Turns out Go’s standard library implements
io.MultiWriter, which is a more generic version of
CapturingPassThroughWriter
. Let’s use that instead:
func main() {
cmd := exec.Command("ls", "-lah")
if runtime.GOOS == "windows" {
cmd = exec.Command("tasklist")
}
var stdoutBuf, stderrBuf bytes.Buffer
cmd.Stdout = io.MultiWriter(os.Stdout, &stdoutBuf)
cmd.Stderr = io.MultiWriter(os.Stderr, &stderrBuf)
err := cmd.Run()
if err != nil {
log.Fatalf("cmd.Run() failed with %s\n", err)
}
outStr, errStr := string(stdoutBuf.Bytes()), string(stderrBuf.Bytes())
fmt.Printf("\nout:\n%s\nerr:\n%s\n", outStr, errStr)
}
It’s good to be able to write the code ourselves, but it’s even better to know standard library well!
Writing to program’s stdin
We know how to read program’s stdout but we can also write to its stdin.
There is no Go library to do bzip2 compression (only decompression is available in standard library).
We can use bzip2
to do the compression by: * writing the data to a temporary file * call bzip2 -c ${file_in}
and capture its stdout
It would be even better if we didn’t have to create a temporary file.
Most compression programs accept data to compress/decompress on stdin.
To do that on command-line we would use the following command: bzip2 -c <${file_in} >${file_out}
.
Here’s the same thing in Go:
// compress data using bzip2 without creating temporary files
func bzipCompress(d []byte) ([]byte, error) {
var out bytes.Buffer
// -c : compress
// -9 : select the highest level of compresion
cmd := exec.Command("bzip2", "-c", "-9")
cmd.Stdin = bytes.NewBuffer(d)
cmd.Stdout = &out
err := cmd.Run()
if err != nil {
return nil, err
}
return out.Bytes(), nil
}
We can also call cmd.StdinPipe()
, which returns io.WriteCloser
. It’s more complicated but gives more control over writing.
Changing environment of executed program
Things to know about environment variables in Go:
os.Environ()
returns []string
where each string is in form of FOO=bar
, where FOO
is the name of environment variable and bar
is the value
os.Getenv("FOO")
returns the value of environment variable FOO
.
Sometimes you need to modify the environment of the executed program.
Go supports that by setting Env
member of exec.Cmd
. cmd.Env
has the same format as os.Environ()
.
If Env
is not set, the process inherits environment of the calling process.
Usually, you don’t want to construct a completely new environment from scratch but pass a modified version of an environment of the current process. Here’s how to add a new variable:
cmd := exec.Command("programToExecute")
additionalEnv := "FOO=bar"
newEnv := append(os.Environ(), additionalEnv))
cmd.Env = newEnv
out, err := cmd.CombinedOutput()
if err != nil {
log.Fatalf("cmd.Run() failed with %s\n", err)
}
fmt.Printf("%s", out)
Full example
Things get more complicated if you want to delete an environment or to ensure you’re not setting the same variable twice. Package
shurcooL/go/osutil offers an easier way of manipulating environment variables.
Check early that a program is installed
Imagine you wrote a program that takes a long time to run. At the end, you call executable foo
to perform an essential task.
If foo
executable is not present, the call will fail.
It’s a good idea to detect lack of executable foo
at the beginning and fail early with descriptive error message.
You can do it using exec.LookPath
.
func checkLsExists() {
path, err := exec.LookPath("ls")
if err != nil {
fmt.Printf("didn't find 'ls' executable\n")
} else {
fmt.Printf("'ls' executable is in '%s'\n", path)
}
}
Another way to check if a program exists is to try to execute it in a no-op mode (e.g. many programs support --help
option).