Skip to content

A Go library for writing CLI programs. It includes flag parsing, colour escape codes, and various helpful utility functions, and makes testing fairly easy.



Switch branches/tags

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time

zli is a Go library for writing CLI programs. It includes flag parsing, color escape codes, various helpful utility functions, and makes testing fairly easy. There's a little example at cmd/grep, which should give a decent overview of how actual programs look like.

Import as; API docs:

Other packages:

Readme index: Utility functions · Flag parsing · Colors · Testing

Utility functions

zli.Errorf() and zli.Fatalf() work like fmt.Printf(), except that they print to stderr, prepend the program name, and always append a newline:

zli.Errorf("oh noes: %s", "u brok it")   // "progname: oh noes: u brok it" to stderr
zli.Fatalf("I swear it was %s", "Dave")  // "progname: I swear it was Dave" to stderr and exit 1

zli.F() is a small wrapper/shortcut around zli.Fatalf() which accepts an error and checks if it's nil first:

err := f()

For many programs it's useful to be able to read from stdin or from a file, depending on what arguments the user gave. With zli.InputOrFile() this is pretty easy:

arg := "/a-file"
fp, err := zli.InputOrFile(arg, false)        // Open stdin if arg is "-" or "-", or a file otherwise.
defer fp.Close()                              // No-op close on stdin.

The second argument controls if a reading from stdin... message should be printed to stderr, which is a bit better UX IMHO (how often have you typed grep foo and waited, only to realize it's waiting for stdin?) See Better UX when reading from stdin.

zli.Pager() pipes the contents of a reader $PAGER. It will copy the contents to stdout if $PAGER isn't set or on other errors:

fp, _ := os.Open("/file")        // Display file in $PAGER.

If you want to page output your program generates you can use zli.PagerStdout() to swap zli.Stdout to a buffer:

defer zli.PagerStdout()()               // Double ()()!
fmt.Fprintln(zli.Stdout, "page me!")    // Displayed in the $PAGER.

This does require that your program writes to zli.Stdout instead of os.Stdout, which is probably a good idea for testing anyway. See the Testing section.

You need to be a bit careful when calling Exit() explicitly, since that will exit immediately without running any defered functions. You have to either use a wrapper or call the returned function explicitly:

func main() { zli.Exit(run()) }

func run() int {
    defer zli.PagerStdout()()
    fmt.Fprintln(zli.Stdout, "XXX")
    return 1
func main() {
    runPager := zli.PagerStdout()
    fmt.Fprintln(zli.Stdout, "XXX")


zli helpfully includes the GetSize() and IsTerminal() from x/term as they're so commonly used:

interactive := zli.IsTerminal(os.Stdout.Fd())  // Check if stdout is a terminal.
w, h, err := zli.TerminalSize(os.Stdout.Fd())  // Get terminal size.

Flag parsing

zli comes with a flag parser which, IMHO, gives a better experience than Go's flag package. See flag.markdown for some rationale on "why this and not stdlib flags?"

// Create new flags; normally you'd pass in os.Args here.
f := zli.NewFlags([]string{"example", "-vv", "-f=csv", "-a", "xx", "yy"})

// The first argument is the default and everything after that is the flag name
// with aliases.
var (
    verbose = f.IntCounter(0, "v", "verbose")        // Count the number of -v flags.
    exclude = f.StringList(nil, "e", "exclude")      // Can appear more than once.
    all     = f.Bool(false, "a", "all")              // Regular bool.
    format  = f.String("", "f", "format")            // Regular string.
    asJSON  = f.Optional().String("", "-j", "json")  // Optional value

// Shift the first argument (i.e. os.Args[1]). Useful to get the "subcommand"
// name. This works before and after Parse().
// Can also use "cmd := f.ShiftCommand("help", "install")", in which case it
// will use the first non-ambiguous match.
switch f.Shift() {
case "help":
    // Run help
case "install":
    // Run install
case "": // os.Args wasn't long enough.
    // Error: need a command (or just print the usage)
    // Error: Unknown command

// Parse the shebang!
err := f.Parse()
if err != nil {
    // Print error, usage.

// You can check if the flag was present on the CLI with Set(). This way you can
// distinguish between "was an empty value passed" (-format '') and "this flag
// wasn't on the CLI".
if format.Set() {
    fmt.Println("Format was set to", format.String())

// The IntCounter adds 1 for every time the -v flag is on the CLI.
if verbose.Int() > 1 {
    // ...Print very verbose info.
} else if verbose.Int() > 0 {
    // ...Print less verbose info.

// Just a bool!
fmt.Println("All:", all.Bool())

// Allow a flag to appear more than once.
fmt.Println("%s exclude patterns: %v", len(all.Strings()), all.Strings())

// -json flag has an optional value.
if asJSON.Set() {
    printas := "json"
    if asJSON.String() == "indent" {
        printas = "json-indent"

// f.Args is set to everything that's not a flag or argument.
fmt.Println("Remaining:", f.Args)

The flag format is as follows:

  • Flags can have a single - or two --, they're treated identical.

  • Arguments are after a space or =: -f v or -f=v. Arguments that start with a - must use the = variant (-f=-val).

  • Booleans can be grouped; -ab is the same as -a -b; this only works with a single - (--ab would be an error).

  • Positional arguments may appear anywhere; these are all identical: -a -b arg, arg -a -b, -a arg -b.

There is no automatic generation of a usage message; I find that much of the time you get a much higher quality by writing one manually. It does provide zli.Usage() you can apply some generic substitutions giving a format somewhat reminiscent of manpages:

UsageTrim      Trim leading/trailing whitespace, and ensure it ends with \n
UsageHeaders   Format headers in the form "^Name:" as bold and underline.
UsageFlags     Format flags (-v, --flag, --flag=foo) as underlined.

See the grep example.


You can add colors and some other text attributes to a string with zli.Colorize(), which returns a modified string with the terminal escape codes, ending with reset.

It won't do anything if zli.WantColor is false; this is disabled by default if the output isn't a terminal or NO_COLOR is set, but you can override it if the user sets --color=force or something.

zli.Colorln() and zli.Colorf() are convenience wrappers for fmt.Println() and fmt.Printf() with colors.

There are constants for the basic terminal attributes and 16-color palette which may be combined freely by adding them together:

zli.Colorln("You're looking rather red", zli.Red)     // Apply a color.
zli.Colorln("A bold move", zli.Bold)                  // Or an attribute.
zli.Colorln("A bold move", zli.Red | zli.Bold)        // Or both.

To set a background color transform the color with the Bg() method:

zli.Colorln("Tomato", zli.Red.Bg())                   // Transform to background color.
zli.Colorln("Wow, such beautiful text",               // Can be combined.
    zli.Bold | zli.Red | zli.Green.Bg())

There are no pre-defined constants for the 256-color palette or true colors, you need to use Color256() and ColorHex() to create them; you can use the Bg() to transform them to a background color as well:

zli.Colorln("Contrast ratios is for suckers",         // 256 color.
    zli.Color256(56) | zli.Color256(99).Bg())

zli.Colorln("REAL men use TRUE color!",               // True color.
    zli.ColorHex("#fff") | zli.ColorHex("#00f").Bg())

With Brighten() you can change the brightness of a color:

zli.Colorln("Brighter! BRIGHTER!", zli.Color256(99).Brighten(1))
zli.Colorln("Dim the lights.              // Negative values darken.

See cmd/colortest/main.go for a little program to display and test colors.

For some more advanced cases you can use Color.String() directly, but this won't look at zli.WantColor and you'll need to manually apply the reset code:

fmt.Println(zli.Red|zli.Bold, "red!")                 // Print escape codes.
fmt.Println("and bold!", zli.Reset)

fmt.Printf("%sc%so%sl%so%sr%s\n", zli.Red, zli.Magenta, zli.Cyan, zli.Blue, zli.Yellow, zli.Reset)

Because the color is stored in an uint64 you can assign them to a constant:

const colorMatch = zli.Bold | zli.Red

This won't work if you use Color256() or ColorHex(); although you can get around this by constructing it all yourself:

// zli.Color256(99)
const color = zli.Bold | (zli.Color(99) << zli.ColorOffsetFg) | zli.ColorMode256

// zli.ColorHex("#ff7711").Bg(); can also use 1144831 directly instead of the
// bit shifts.
const color2 = zli.Bold | zli.Red | zli.ColorModeTrueBg |
               (zli.Color(0xff|0x77<<8|0x11<<16) << zli.ColorOffsetBg)

This creates a color stored as an int, shifts it to the correct location, and sets the flag to signal how to interpret it.

Do you really want to do this just to create a const instead of a var? Probably not 😅


zli uses to zli.Stdin, zli.Stdout, zli.Stderr, and zli.Exit instead of the os.* variants for everything. You can swap this out with test variants with the zli.Test() function.

You can use these in your own program as well, if you want to test the output of a program.

func TestX(t *testing.T) {
    exit, in, out := Test(t) // Resets everything back to os.* with t.Cleanup()

    // Write something to stderr (a bytes.Buffer) and read the output.
    Error("oh noes!")
    fmt.Println(out.String()) // zli.test: oh noes!

    // Read from stdin.
    fp, _ := InputOrFile("-", true)
    got, _ := ioutil.ReadAll(fp)
    fmt.Println(string(got)) // Hello


    et := func() {
        fmt.Fprintln(Stdout, "one")
        fmt.Fprintln(Stdout, "two")

    // exit panics to ensure the regular control flow of the program is aborted;
    // to capture this run the function to be tested in a closure with
    // exit.Recover(), which will recover() from the panic and set the exit
    // code.
    func() {
        defer exit.Recover()
    // Helper to check the statis code, so you don't have to derefrence and cast
    // the value to int.
    exit.Want(t, 1)

    fmt.Println("Exit %d: %s\n", *exit, out.String()) // Exit 1: one

You don't need to use the zli.Test() function if you won't want to, you can just swap out stuff yourself as well:

buf := new(bytes.Buffer)
zli.Stderr = buf
defer func() { Stderr = os.Stderr }()

Error("oh noes!")
out := buf.String()
fmt.Printf("buffer has: %q\n", out) // buffer has: "zli.test: oh noes!\n"

zli.IsTerminal() and zli.TerminalSize() are variables, and can be swapped out as well:

save := zli.IsTerminal
zli.IsTerminal = func(uintptr) bool { return true }
defer func() { IsTerminal = save }()


A few notes on replacing zli.Exit() in tests: the difficulty with this is that os.Exit() will terminate the entire program, including the test, which is rarely what you want and difficult to test. You can replace zli.Exit with something like (zli.TestExit() takes care of all of this):

var code int
zli.Exit = func(c int) { code = c }
fmt.Println("exit code", code)

This works well enough for simple cases, but there's a big caveat with this; for example consider:

func mayExit() {
    err := f()
    if err != nil {


With the above the program will continue after zli.Exit(); which is a different program flow from normal execution. A simple way to fix it so to modify the function to explicitly call return:

func mayExit() {
    err := f()
    if err != nil {


This still isn't quite the same, as callers of mayExit() in your program will still continue happily. It's also rather ugly and clunky.

To solve this you can replace zli.Exit with a function that panics and then recover that:

func TestFoo(t *testing.T) {
    var code int
    zli.Exit = func(c int) {
        code = c

    func() {
        defer func() {
            r := recover()
            if r == nil {


    fmt.Println("Exited with", code)

This will abort the program flow similar to os.Exit(), and the call to mayExit is wrapped in a function the test function itself will continue after the recover.


A Go library for writing CLI programs. It includes flag parsing, colour escape codes, and various helpful utility functions, and makes testing fairly easy.








No releases published

Sponsor this project