Skip to content

alexflint/go-arg

Repository files navigation

go-arg
go-arg

Struct-based argument parsing for Go

Sourcegraph Documentation Build Status Coverage Status Go Report Card


Declare command line arguments for your program by defining a struct.

var args struct {
	Foo string
	Bar bool
}
arg.MustParse(&args)
fmt.Println(args.Foo, args.Bar)
$ ./example --foo=hello --bar
hello true

Installation

go get github.com/alexflint/go-arg

Required arguments

var args struct {
	ID      int `arg:"required"`
	Timeout time.Duration
}
arg.MustParse(&args)
$ ./example
Usage: example --id ID [--timeout TIMEOUT]
error: --id is required

Positional arguments

var args struct {
	Input   string   `arg:"positional"`
	Output  []string `arg:"positional"`
}
arg.MustParse(&args)
fmt.Println("Input:", args.Input)
fmt.Println("Output:", args.Output)
$ ./example src.txt x.out y.out z.out
Input: src.txt
Output: [x.out y.out z.out]

Environment variables

var args struct {
	Workers int `arg:"env"`
}
arg.MustParse(&args)
fmt.Println("Workers:", args.Workers)
$ WORKERS=4 ./example
Workers: 4
$ WORKERS=4 ./example --workers=6
Workers: 6

You can also override the name of the environment variable:

var args struct {
	Workers int `arg:"env:NUM_WORKERS"`
}
arg.MustParse(&args)
fmt.Println("Workers:", args.Workers)
$ NUM_WORKERS=4 ./example
Workers: 4

You can provide multiple values in environment variables using commas:

var args struct {
    Workers []int `arg:"env"`
}
arg.MustParse(&args)
fmt.Println("Workers:", args.Workers)
$ WORKERS='1,99' ./example
Workers: [1 99]

Command line arguments take precedence over environment variables:

var args struct {
	Workers int `arg:"--count,env:NUM_WORKERS"`
}
arg.MustParse(&args)
fmt.Println("Workers:", args.Workers)
$ NUM_WORKERS=6 ./example
Workers: 6
$ NUM_WORKERS=6 ./example --count 4
Workers: 4

Configuring a global environment variable name prefix is also possible:

var args struct {
	Workers int `arg:"--count,env:NUM_WORKERS"`
}

p, err := arg.NewParser(arg.Config{
    EnvPrefix: "MYAPP_",
}, &args)

p.MustParse(os.Args[1:])
fmt.Println("Workers:", args.Workers)
$ MYAPP_NUM_WORKERS=6 ./example
Workers: 6

Usage strings

var args struct {
	Input    string   `arg:"positional"`
	Output   []string `arg:"positional"`
	Verbose  bool     `arg:"-v,--verbose" help:"verbosity level"`
	Dataset  string   `help:"dataset to use"`
	Optimize int      `arg:"-O" help:"optimization level"`
}
arg.MustParse(&args)
$ ./example -h
Usage: [--verbose] [--dataset DATASET] [--optimize OPTIMIZE] [--help] INPUT [OUTPUT [OUTPUT ...]]

Positional arguments:
  INPUT
  OUTPUT

Options:
  --verbose, -v            verbosity level
  --dataset DATASET        dataset to use
  --optimize OPTIMIZE, -O OPTIMIZE
                           optimization level
  --help, -h               print this help message

Default values

var args struct {
	Foo string `default:"abc"`
	Bar bool
}
arg.MustParse(&args)

Command line arguments take precedence over environment variables, which take precedence over default values. This means that we check whether a certain option was provided on the command line, then if not, we check for an environment variable (only if an env tag was provided), then if none is found, we check for a default tag containing a default value.

var args struct {
    Test  string `arg:"-t,env:TEST" default:"something"`
}
arg.MustParse(&args)

Ignoring environment variables and/or default values

var args struct {
    Test  string `arg:"-t,env:TEST" default:"something"`
}

p, err := arg.NewParser(arg.Config{
    IgnoreEnv: true,
    IgnoreDefault: true,
}, &args)

err = p.Parse(os.Args[1:])

Arguments with multiple values

var args struct {
	Database string
	IDs      []int64
}
arg.MustParse(&args)
fmt.Printf("Fetching the following IDs from %s: %q", args.Database, args.IDs)
./example -database foo -ids 1 2 3
Fetching the following IDs from foo: [1 2 3]

Arguments that can be specified multiple times, mixed with positionals

var args struct {
    Commands  []string `arg:"-c,separate"`
    Files     []string `arg:"-f,separate"`
    Databases []string `arg:"positional"`
}
arg.MustParse(&args)
./example -c cmd1 db1 -f file1 db2 -c cmd2 -f file2 -f file3 db3 -c cmd3
Commands: [cmd1 cmd2 cmd3]
Files [file1 file2 file3]
Databases [db1 db2 db3]

Arguments with keys and values

var args struct {
	UserIDs map[string]int
}
arg.MustParse(&args)
fmt.Println(args.UserIDs)
./example --userids john=123 mary=456
map[john:123 mary:456]

Version strings

type args struct {
	...
}

func (args) Version() string {
	return "someprogram 4.3.0"
}

func main() {
	var args args
	arg.MustParse(&args)
}
$ ./example --version
someprogram 4.3.0

Note If a --version flag is defined in args or any subcommand, it overrides the built-in versioning.

Custom validation

var args struct {
	Foo string
	Bar string
}
p := arg.MustParse(&args)
if args.Foo == "" && args.Bar == "" {
	p.Fail("you must provide either --foo or --bar")
}
./example
Usage: samples [--foo FOO] [--bar BAR]
error: you must provide either --foo or --bar

Overriding option names

var args struct {
	Short        string `arg:"-s"`
	Long         string `arg:"--custom-long-option"`
	ShortAndLong string `arg:"-x,--my-option"`
	OnlyShort    string `arg:"-o,--"`
}
arg.MustParse(&args)
$ ./example --help
Usage: example [-o ONLYSHORT] [--short SHORT] [--custom-long-option CUSTOM-LONG-OPTION] [--my-option MY-OPTION]

Options:
  --short SHORT, -s SHORT
  --custom-long-option CUSTOM-LONG-OPTION
  --my-option MY-OPTION, -x MY-OPTION
  -o ONLYSHORT
  --help, -h             display this help and exit

Embedded structs

The fields of embedded structs are treated just like regular fields:

type DatabaseOptions struct {
	Host     string
	Username string
	Password string
}

type LogOptions struct {
	LogFile string
	Verbose bool
}

func main() {
	var args struct {
		DatabaseOptions
		LogOptions
	}
	arg.MustParse(&args)
}

As usual, any field tagged with arg:"-" is ignored.

Supported types

The following types may be used as arguments:

  • built-in integer types: int, int8, int16, int32, int64, byte, rune
  • built-in floating point types: float32, float64
  • strings
  • booleans
  • URLs represented as url.URL
  • time durations represented as time.Duration
  • email addresses represented as mail.Address
  • MAC addresses represented as net.HardwareAddr
  • pointers to any of the above
  • slices of any of the above
  • maps using any of the above as keys and values
  • any type that implements encoding.TextUnmarshaler

Custom parsing

Implement encoding.TextUnmarshaler to define your own parsing logic.

// Accepts command line arguments of the form "head.tail"
type NameDotName struct {
	Head, Tail string
}

func (n *NameDotName) UnmarshalText(b []byte) error {
	s := string(b)
	pos := strings.Index(s, ".")
	if pos == -1 {
		return fmt.Errorf("missing period in %s", s)
	}
	n.Head = s[:pos]
	n.Tail = s[pos+1:]
	return nil
}

func main() {
	var args struct {
		Name NameDotName
	}
	arg.MustParse(&args)
	fmt.Printf("%#v\n", args.Name)
}
$ ./example --name=foo.bar
main.NameDotName{Head:"foo", Tail:"bar"}

$ ./example --name=oops
Usage: example [--name NAME]
error: error processing --name: missing period in "oops"

Custom parsing with default values

Implement encoding.TextMarshaler to define your own default value strings:

// Accepts command line arguments of the form "head.tail"
type NameDotName struct {
	Head, Tail string
}

func (n *NameDotName) UnmarshalText(b []byte) error {
	// same as previous example
}

// this is only needed if you want to display a default value in the usage string
func (n *NameDotName) MarshalText() ([]byte, error) {
	return []byte(fmt.Sprintf("%s.%s", n.Head, n.Tail)), nil
}

func main() {
	var args struct {
		Name NameDotName `default:"file.txt"`
	}
	arg.MustParse(&args)
	fmt.Printf("%#v\n", args.Name)
}
$ ./example --help
Usage: test [--name NAME]

Options:
  --name NAME [default: file.txt]
  --help, -h             display this help and exit

$ ./example
main.NameDotName{Head:"file", Tail:"txt"}

Custom placeholders

Use the placeholder tag to control which placeholder text is used in the usage text.

var args struct {
	Input    string   `arg:"positional" placeholder:"SRC"`
	Output   []string `arg:"positional" placeholder:"DST"`
	Optimize int      `arg:"-O" help:"optimization level" placeholder:"LEVEL"`
	MaxJobs  int      `arg:"-j" help:"maximum number of simultaneous jobs" placeholder:"N"`
}
arg.MustParse(&args)
$ ./example -h
Usage: example [--optimize LEVEL] [--maxjobs N] SRC [DST [DST ...]]

Positional arguments:
  SRC
  DST

Options:
  --optimize LEVEL, -O LEVEL
                         optimization level
  --maxjobs N, -j N      maximum number of simultaneous jobs
  --help, -h             display this help and exit

Description strings

A descriptive message can be added at the top of the help text by implementing a Description function that returns a string.

type args struct {
	Foo string
}

func (args) Description() string {
	return "this program does this and that"
}

func main() {
	var args args
	arg.MustParse(&args)
}
$ ./example -h
this program does this and that
Usage: example [--foo FOO]

Options:
  --foo FOO
  --help, -h             display this help and exit

Similarly an epilogue can be added at the end of the help text by implementing the Epilogue function.

type args struct {
	Foo string
}

func (args) Epilogue() string {
	return "For more information visit github.com/alexflint/go-arg"
}

func main() {
	var args args
	arg.MustParse(&args)
}
$ ./example -h
Usage: example [--foo FOO]

Options:
  --foo FOO
  --help, -h             display this help and exit

For more information visit github.com/alexflint/go-arg

Subcommands

Subcommands are commonly used in tools that wish to group multiple functions into a single program. An example is the git tool:

$ git checkout [arguments specific to checking out code]
$ git commit [arguments specific to committing]
$ git push [arguments specific to pushing]

The strings "checkout", "commit", and "push" are different from simple positional arguments because the options available to the user change depending on which subcommand they choose.

This can be implemented with go-arg as follows:

type CheckoutCmd struct {
	Branch string `arg:"positional"`
	Track  bool   `arg:"-t"`
}
type CommitCmd struct {
	All     bool   `arg:"-a"`
	Message string `arg:"-m"`
}
type PushCmd struct {
	Remote      string `arg:"positional"`
	Branch      string `arg:"positional"`
	SetUpstream bool   `arg:"-u"`
}
var args struct {
	Checkout *CheckoutCmd `arg:"subcommand:checkout"`
	Commit   *CommitCmd   `arg:"subcommand:commit"`
	Push     *PushCmd     `arg:"subcommand:push"`
	Quiet    bool         `arg:"-q"` // this flag is global to all subcommands
}

arg.MustParse(&args)

switch {
case args.Checkout != nil:
	fmt.Printf("checkout requested for branch %s\n", args.Checkout.Branch)
case args.Commit != nil:
	fmt.Printf("commit requested with message \"%s\"\n", args.Commit.Message)
case args.Push != nil:
	fmt.Printf("push requested from %s to %s\n", args.Push.Branch, args.Push.Remote)
}

Some additional rules apply when working with subcommands:

  • The subcommand tag can only be used with fields that are pointers to structs
  • Any struct that contains a subcommand must not contain any positionals

This package allows to have a program that accepts subcommands, but also does something else when no subcommands are specified. If on the other hand you want the program to terminate when no subcommands are specified, the recommended way is:

p := arg.MustParse(&args)
if p.Subcommand() == nil {
    p.Fail("missing subcommand")
}

Custom handling of --help and --version

The following reproduces the internal logic of MustParse for the simple case where you are not using subcommands or --version. This allows you to respond programatically to --help, and to any errors that come up.

var args struct {
	Something string
}

p, err := arg.NewParser(arg.Config{}, &args)
if err != nil {
	log.Fatalf("there was an error in the definition of the Go struct: %v", err)
}

err = p.Parse(os.Args[1:])
switch {
case err == arg.ErrHelp:  // indicates that user wrote "--help" on command line
	p.WriteHelp(os.Stdout)
	os.Exit(0)
case err != nil:
	fmt.Printf("error: %v\n", err)
	p.WriteUsage(os.Stdout)
	os.Exit(1)
}
$ go run ./example --help
Usage: ./example --something SOMETHING

Options:
  --something SOMETHING
  --help, -h             display this help and exit

$ ./example --wrong
error: unknown argument --wrong
Usage: ./example --something SOMETHING

$ ./example
error: --something is required
Usage: ./example --something SOMETHING

To also handle --version programatically, use the following:

type args struct {
	Something string
}

func (args) Version() string {
	return "1.2.3"
}

func main() {
	var args args
	p, err := arg.NewParser(arg.Config{}, &args)
	if err != nil {
		log.Fatalf("there was an error in the definition of the Go struct: %v", err)
	}

	err = p.Parse(os.Args[1:])
	switch {
	case err == arg.ErrHelp: // found "--help" on command line
		p.WriteHelp(os.Stdout)
		os.Exit(0)
	case err == arg.ErrVersion: // found "--version" on command line
		fmt.Println(args.Version())
		os.Exit(0)
	case err != nil:
		fmt.Printf("error: %v\n", err)
		p.WriteUsage(os.Stdout)
		os.Exit(1)
	}

	fmt.Printf("got %q\n", args.Something)
}
$ ./example --version
1.2.3

$ go run ./example --help
1.2.3
Usage: example --something SOMETHING

Options:
  --something SOMETHING
  --help, -h             display this help and exit

$ ./example --wrong
1.2.3
error: unknown argument --wrong
Usage: example --something SOMETHING

$ ./example
error: --something is required
Usage: example --something SOMETHING

To generate subcommand-specific help messages, use the following most general version (this also works in absence of subcommands but is a bit more complex):

type fetchCmd struct {
	Count int
}

type args struct {
	Something string
	Fetch     *fetchCmd `arg:"subcommand"`
}

func (args) Version() string {
	return "1.2.3"
}

func main() {
	var args args
	p, err := arg.NewParser(arg.Config{}, &args)
	if err != nil {
		log.Fatalf("there was an error in the definition of the Go struct: %v", err)
	}

	err = p.Parse(os.Args[1:])
	switch {
	case err == arg.ErrHelp: // found "--help" on command line
		p.WriteHelpForSubcommand(os.Stdout, p.SubcommandNames()...)
		os.Exit(0)
	case err == arg.ErrVersion: // found "--version" on command line
		fmt.Println(args.Version())
		os.Exit(0)
	case err != nil:
		fmt.Printf("error: %v\n", err)
		p.WriteUsageForSubcommand(os.Stdout, p.SubcommandNames()...)
		os.Exit(1)
	}
}
$ ./example --version
1.2.3

$ ./example --help
1.2.3
Usage: example [--something SOMETHING] <command> [<args>]

Options:
  --something SOMETHING
  --help, -h             display this help and exit
  --version              display version and exit

Commands:
  fetch

$ ./example fetch --help
1.2.3
Usage: example fetch [--count COUNT]

Options:
  --count COUNT

Global options:
  --something SOMETHING
  --help, -h             display this help and exit
  --version              display version and exit

API Documentation

https://pkg.go.dev/github.com/alexflint/go-arg

Rationale

There are many command line argument parsing libraries for Go, including one in the standard library, so why build another?

The flag library that ships in the standard library seems awkward to me. Positional arguments must precede options, so ./prog x --foo=1 does what you expect but ./prog --foo=1 x does not. It also does not allow arguments to have both long (--foo) and short (-f) forms.

Many third-party argument parsing libraries are great for writing sophisticated command line interfaces, but feel to me like overkill for a simple script with a few flags.

The idea behind go-arg is that Go already has an excellent way to describe data structures using structs, so there is no need to develop additional levels of abstraction. Instead of one API to specify which arguments your program accepts, and then another API to get the values of those arguments, go-arg replaces both with a single struct.

Backward compatibility notes

Earlier versions of this library required the help text to be part of the arg tag. This is still supported but is now deprecated. Instead, you should use a separate help tag, described above, which makes it possible to include commas inside help text.