Thin, declarative API for creating console applications with features like text decoration, user input handling, and seamless terminal text repainting. Supports rerendering sections, reactive state management, and input completions.
session {
var wantsToLearn by liveVarOf(false)
section {
text("Would you like to learn "); cyan { text("Kotter") }; textLine("? (Y/n)")
text("> "); input(Completions("yes", "no"))
if (wantsToLearn) {
yellow(isBright = true) { p { textLine("""\(^o^)/""") } }
}
}.runUntilInputEntered {
onInputEntered { wantsToLearn = "yes".startsWith(input.lowercase()) }
}
}
See also: the game of life, snake, sliding tiles, doom fire, and Wordle implemented in Kotter!
Kotter (a KOTlin TERminal library) aims to be a relatively thin, declarative, Kotlin-idiomatic API that provides
useful functionality for writing delightful console applications. It strives to keep things simple, providing a solution
a bit more opinionated than making raw println calls but way less featured than something like Java Curses.
Specifically, this library helps with:
Kotter is multiplatform, supporting JVM and native targets.
The next sections deal with setting Kotter up, but you may wish to jump straight to the usage section ▼ to immediately start learning about this library.
Kotter supports JVM and native targets.
[!TIP] If you're not sure what you want, start with a JVM project. That target is far easier to distribute. It also means your project will have access to a very broad ecosystem of Kotlin and Java libraries.
In case it affects your decision, you can read more about distributing Kotter applications ▼ later in this document.
Multiplatform can be useful if you want to distribute binaries to users without requiring they have Java installed on their machine.
// build.gradle.kts (kotlin script)
plugins {
kotlin("multiplatform")
}
repositories {
mavenCentral()
}
kotlin {
// Choose the targets you care about.
// Note: You will need the right machine to build each one; otherwise, the target is disabled automatically
listOf(
linuxX64(),
mingwX64(),
macosArm64(),
macosX64(),
).forEach { nativeTarget ->
nativeTarget.apply {
binaries {
executable {
entryPoint =
}
}
}
}
sourceSets {
commonMain getting {
dependencies {
implementation()
}
}
commonTest getting {
dependencies {
implementation()
}
}
}
}
[!NOTE] Building native binaries is a little tricky, as you may need different host machines in order to build the various binaries. For example, here is Kotter's CI workflow which runs on both Linux and Mac targets to build platform-specific Kotter artifacts.
Most users won't ever need to run a Kotter snapshot, so feel free to skip over this section! However, occasionally, bug fixes and new features will be available for testing for a short period before they are released.
If you ever file a bug with Kotter and are asked to test a fix using a snapshot, you must add an entry for the sonatype
snapshots repository to your repositories block in order to allow Gradle to find it:
// build.gradle.kts
repositories {
mavenCentral()
+ maven("https://central.sonatype.com/repository/maven-snapshots/") {
+ mavenContent {
+ includeGroup("com.varabyte.kotter")
+ includeGroup("com.varabyte.kotterx")
+ snapshotsOnly()
+ }
+ }
}
If you've cloned this repository, examples are located under the examples folder.
Most of the examples (except examples/native) target the JVM. To try one of them, you can navigate into it on the
command line and run it via Gradle.
$ cd examples/life
$ ../../gradlew run
However, because Gradle itself has taken over the terminal to do its own fancy command line magic, the example will actually open up and run inside a virtual terminal.
If you want to run the program directly inside your system terminal, which is hopefully the way most users will see your
application, you should use the installDist task to accomplish this:
$ cd examples/life
$ ../../gradlew installDist
$ cd build/install/life/bin
$ ./life
[!WARNING] If your terminal does not support features needed by Kotter, which could happen on legacy machines for example, then this still may end up running inside a virtual terminal.
Unlike the JVM target, native targets do not have a virtual terminal fallback. So be sure you do not use any of the
Gradle run tasks (e.g. runDebugExecutabule...). This will also fail if you try to run your program through the IDE via
the green "play" arrow.
Instead, you should link your executable and then run it directly.
For example, on Linux:
$ cd examples/native
$ ../../gradlew linkDebugExecutableLinuxX64
$ ./build/bin/linuxX64/debugExecutable/native.kexe
The following is equivalent to println("Hello, World"). In this simple case, it's definitely overkill!
session {
section { textLine("Hello, World") }.run()
}
section { ... } defines a Section which, on its own, is inert. It needs to be run to output text to the
console. Above, we use the run method to trigger this. The method blocks until the render (i.e. text printing to the
console) is finished (which, in the above case, will be almost instant).
session { ... } sets the outer scope for your whole program. While we're just calling it with default arguments here,
you can also pass in parameters that apply to the entire application.
While the above simple case is a bit verbose for what it's doing, Kotter starts to show its strength when doing background work (or other async tasks like waiting for user input) during which time the section block may render several times. We'll see many examples throughout this document later.
A Kotter session can contain one or more sections. Your own app may only ever contain a single section and that's
fine! But if you have multiple sections, it will feel to the user like your app has a current, active area, following
a history of text paragraphs from previous interactions that no longer change.
You can call color methods directly, which remain in effect until the next color method is called:
section {
green(layer = BG)
red() // defaults to FG layer if no layer specified
textLine("Red on green")
blue()
textLine("Blue on green")
}.run()
If you only want the color effect to live for a limited time, you can use scoped helper versions that handle clearing colors for you automatically at the end of their block:
section {
green(layer = BG) {
red {
textLine("Red on green")
}
textLine("Default on green")
blue {
textLine("Blue on green")
}
}
}.run()
If the user's terminal supports truecolor mode, you can specify rgb (or hsv) values directly:
section {
rgb(0xFFFF00) { textLine("Yellow!") }
hsv(35, 1.0f, 1.0f) { textLine("Orange!") }
}.run()
[!NOTE] If truecolor is not supported, terminals may attempt to emulate it by falling back to a nearby color, which may look decent! However, to be safe, you may want to avoid rendering smooth gradient color changes, as they may come out clumped for some users.
Various text effects (like bold) are also available:
section {
bold {
textLine("Title")
}
p {
textLine("A paragraph is content auto-surrounded by newlines")
}
p {
text("This paragraph has an ")
underline { text("underlined") }
textLine(" word in it")
}
}.run()
[!NOTE] Italics functionality is not currently exposed, as it is not a standard feature and is inconsistently supported across terminals.
You can also define links:
section {
text("Would you like to ")
link("https://github.com/varabyte/kotter", "learn Kotter")
textLine("?")
}
although keep in mind that this feature is not guaranteed to work on every terminal. In that case, it will simply render as normal text.
To reduce the chance of introducing unexpected bugs later, state changes (like colors) will be localized to the current
section block only:
section {
blue(BG)
red()
text("This text is red on blue")
}.run()
section {
text("This text is rendered using default colors")
}.run()
Within a section, you can also use the scopedState method. This creates a new scope within which any state will be
automatically discarded after it ends.
section {
scopedState {
red()
blue(BG)
underline()
textLine("Underlined red on blue")
}
text("Text without color or decorations")
}.run()
[!NOTE] Scoped text effect methods (like
red { ... }) work by callingscopedStatefor you under the hood.
The section block is designed to be run one or more times. That is, you can write logic inside it which may not get
executed on the first run but will be on a followup run.
Here, instead of just calling run(), we create a run block, having it update a variable that is also referenced by
the section block. This example will render the section twice - once when run is first called and again when it
calls rerender:
var result: Int? = null
section {
text("Calculating... ")
if (result != null) {
text("Done! Result = $result")
}
}.run {
result = doNetworkFetchAndExpensiveCalculation()
rerender()
}
The run block runs as a suspend function, so you can call other suspend methods from within it.
Your program will be blocked until the run block has finished (or, if it has triggered a rerender, until the last rerender finishes).
In our example above, the run block calls a rerender method, which you can call to request another render pass:
/* ... */
run {
result = doNetworkFetchAndExpensiveCalculation()
rerender()
}
However, remembering to call rerender yourself is potentially fragile and could be a source of bugs in the future when
trying to figure out why your console isn't updating.
For this purpose, Kotter provides the LiveVar class, which, when modified, will automatically request a rerender.
An example will demonstrate this in action shortly.
To create a LiveVar, simply change a normal variable declaration line like:
session {
var result: Int? = null
/* ... */
}
to:
session {
var result by liveVarOf<Int?>(null)
/* ... */
}
[!IMPORTANT] The
liveVarOfmethod is provided by thesessionblock. For many remaining examples, we'll elide thesessionboilerplate, but that doesn't mean you can omit it in your own program!
Let's apply liveVarOf to our earlier example in order to remove the rerender call:
var result by liveVarOf<Int?>(null)
section {
/* ... no changes ... */
}.run {
result = doNetworkFetchAndExpensiveCalculation()
}
And done! Fewer lines and less error pone.
Here's another example, showing how you can use run and a LiveVar to render a progress bar:
// Prints something like: [****------]
val BAR_LENGTH = 10
numFilledSegments liveVarOf()
section {
text()
(i until BAR_LENGTH) {
text( (i < numFilledSegments) )
}
text()
}.run {
percent =
(percent < ) {
delay(Random.nextLong(, ))
percent += Random.nextInt(,)
numFilledSegments = ((percent / ) * BAR_LENGTH).roundToInt()
}
}
Similar to LiveVar, a LiveList is a reactive primitive which, when modified by having elements added to or
removed from it, causes a rerender to happen automatically.
You don't need to use the by keyword with LiveList. Instead, within a session, just assign a variable to the
result of the liveListOf method:
val fileWalker = FileWalker(".") // This class doesn't exist but just pretend for this example...
val fileMatches = liveListOf<String>()
section {
textLine("Matches found so far:")
if (fileMatches.isNotEmpty()) {
for (match in fileMatches) {
textLine(" - $match")
}
}
else {
textLine("No matches so far...")
}
}.run {
fileWalker.findFiles("*.txt") { file ->
fileMatches += file.name
}
}
The LiveList class is thread safe, but you can still run into trouble if you access multiple values on the list one
after the other, as a lock is released between each check. It's always possible that modifying the first property will
kick off a new render which will start before the additional values are set, in other words.
To handle this, you can use the LiveList#withWriteLock method:
val fileWalker = FileWalker(".")
val last10Matches = liveListOf<String>()
section {
/* ... */
}.run {
fileWalker.findFiles("*.txt") { file ->
last10Matches.withWriteLock {
add(file.name)
if (size > 10) { removeAt(0) }
}
}
}
The general rule of thumb is: use withWriteLock if you want to modify more than one property from the list at the same
time within your run block.
[!NOTE] You don't have to worry about locking within a
section { ... }block. Data access is already locked for you in that context.
In addition to LiveList, Kotter also provides LiveMap and LiveSet. There's no need to extensively document these
classes here as much of the earlier LiveList section applies to them as well. It's just the data structure that is
different.
You can create these classes using liveMapOf(...) and liveSetOf(...), respectfully.
A common pattern is for the run block to wait for some sort of signal before finishing, e.g. in response to some
event. You could always use a general threading trick for this, such as a CountDownLatch or a
CompletableDeffered<Unit> to stop the block from finishing until you're ready:
val fileDownloader = FileDownloader("...")
section {
/* ... */
}.run {
val finished = CompletableDeffered<Unit>()
fileDownloader.onFinished += { finished.complete(Unit) }
fileDownloader.start()
finished.await()
}
but, for convenience, Kotter provides the signal and waitForSignal methods, which do this for you.
val fileDownloader = FileDownloader("...")
section {
/* ... */
}.run {
fileDownloader.onFinished += { signal() }
fileDownloader.start()
waitForSignal()
}
These methods are enough in most cases.
[!NOTE] If you call
signalbefore you reachwaitForSignal, thenwaitForSignalwill just pass through without stopping.
There's also a convenience runUntilSignal method you can use, within which you don't need to call waitForSignal
yourself, since this case is so common:
val fileDownloader = FileDownloader("...")
section {
/* ... */
}.runUntilSignal {
fileDownloader.onFinished += { signal() }
fileDownloader.start()
}
Kotter consumes keypresses, so as the user types into the console, nothing will show up unless you intentionally print
it. You can easily do this using the input method, which handles listening to kepresses and adding text into your
section at that location:
section {
// `input` is a method that inserts any user input typed so far in place where it is called.
// Your section block will automatically rerender when its value changes.
text("Please enter your name: "); input()
}.run { /* ... */ }
The input method automatically adds a cursor for you. It also handles keys like LEFT/RIGHT and HOME/END, moving the cursor back and forth between the bounds of the input string.
You can intercept input as it is typed using the onInputChanged event:
section {
text("Please enter your name: "); input()
}.run {
onInputChanged {
input = input.toUpperCase()
}
/* ... */
}
You can also use the rejectInput method to return your input to the previous (presumably valid) state.
section {
text("Please enter your name: "); input()
}.run {
onInputChanged {
if (input.any { !it.isLetter() }) { rejectInput() }
// Would also work: input = input.filter { it.isLetter() }
// although often `rejectInput()` specifies your intention more clearly
}
/* ... */
}
To handle when the user presses the ENTER key, use the onInputEntered event. You can use it in conjunction with the
onInputChanged event we just discussed:
var name = ""
section {
text("Please enter your name: "); input()
}.runUntilSignal {
onInputChanged { if (input.any { !it.isLetter() }) { rejectInput() } }
onInputEntered { name = input; signal() }
}
Above, we've indicated that we want to close the section when the user presses ENTER. Since this is actually a fairly
common case, Kotter provides runUntilInputEntered for your convenience. Using it, we can simplify the above example a
bit, typing fewer characters for identical behavior and expressing clearer intention:
var name = ""
section {
text("Please enter your name: "); input()
}.runUntilInputEntered {
onInputChanged { if (input.any { !it.isLetter() }) { rejectInput() } }
onInputEntered { name = input }
}
You can pass in an InputCompleter implementation to input that can generate suggestions based on the current input.
The user can press RIGHT at any time to autocomplete any suggestions shown to them.
Here's the interface (with some parts elided for simplicity):
interface InputCompleter {
fun complete(input: String): String?
}
input(object : InputCompleter {
override fun complete(input: String): String? { /* ... */ }
})
Perhaps you have a database of names in your program? You can use it to provide suggestions. If your implementation returns null, that means no suggestion was found:
object : InputCompleter {
override fun complete(input: String): String? {
return names
.firstOrNull { it.startsWith(input) }
?.let { it.drop(input.length) }
// NOTE: ^^^^^^^^^^^^^^^^^^^^
// Don't return the whole word; just the part that comes after the user's input so far.
}
}
Kotter provides a very useful implementation out of the box, called Completions, which lets you specify a list of
values that will be autocompleted as long as the user's input matches one of them.
section {
text("Continue? "); input(Completions("yes", "no"))
}.run()
Order matters! If nothing is typed, the first completion will be suggested. If multiple values match, the one earliest in the list will be suggested.
There are two callbacks you can pass into input to affect how it looks, viewMap (for altering the character that
appears) and customFormat (for applying rendering effects). Both callbacks operate character by character, exposed in
the callback as ch. You can also query the whole text string in (text) along with the current index (index) in
case those help you with the current logic.
viewMapviewMap intercepts an incoming character and outputs a new character which will get rendered in its place. This is a
visual change only! The onInputEntered callback will still be triggered with the original input without the view
mapping applied.
It is commonly used to mask something like a password field:
var password = ""
section {
text("Password: "); input(viewMap = { '*' })
}.runUntilInputEntered {
onInputEntered { password = input }
}
// "password" will be set to the actual password; user will only ever see "*"s
customFormatcustomFormat lets you change the color or text effects of the input. Within the callback, you can call methods
color(...), bold(), underline(), and strikethrough() to apply each effect, respectively.
If you want to highlight valid characters and/or emphasize invalid characters in your input, customFormat is the way
to go:
section {
text("PIN: "); input(customFormat = { if (ch.isDigit()) green() else red() })
}.runUntilInputEntered {
onInputEntered { if (input.any { !it.isDigit() }) rejectInput() }
}
Occasionally, you may want to allow users to type long-form text, with newlines. input is quite useful but it
intentionally doesn't allow newlines. That's because pressing ENTER fires the onInputEntered event.
If you need longer form input, you can reach to multilineInput. Users must terminate their input by sending an EOF
signal (generated by pressing CTRL-D), since ENTER is now used for newlines.
[!NOTE] Unfortunately, SHIFT+ENTER, although it is commonly used for handling newlines in most modern editors, is unavailable to us as consoles don't reveal meta-key (e.g. CTRL, SHIFT) states. As console application developers, we're essentially blind to them. CTRL-D is the traditional way to close input streams in many CLIs, because the system translates those keystrokes into an EOF signal, which is all the applications see.
Multiline inputs allow the user to navigate text typed in by pressing the arrow, home, end, page up, and page down keys. In order to support this feature, multiline inputs always start on a new line and any text following it will also appear on a new line after the input block.
section {
black(isBright = true) { text("Send a text message (press CTRL-D when finished)") }
multilineInput()
}.runUntilInputEntered {
onInputEntered { sendMessage(input.trim()) }
}
If you're interested in specific keypresses and not simply input that's been typed in, you can register a listener to
the onKeyPressed event:
section {
textLine("Press Q to quit")
/* ... */
}.run {
var quit = false
onKeyPressed {
when(key) {
Keys.Q -> quit = true
}
}
while (!quit) {
delay(16)
/* ... */
}
}
For convenience, there's also a runUntilKeyPressed method you can use to help with patterns like the above. It can be
nice, for example, to let the user press Q to quit your application:
section {
textLine("Press Q to quit")
/* ... */
}.runUntilKeyPressed(Keys.Q) {
while (true) {
delay(16)
/* ... */
}
}
Kotter can manage a set of timers for you. Use the addTimer method in your run block to add some:
section {
/* ... */
}.runUntilSignal {
addTimer(500.milliseconds) {
println("500ms passed!")
signal()
}
}
You can create a repeating timer by passing in repeat = true to the method. And if you want to stop it from repeating
at some point, set repeat = false inside the timer block when it is triggered:
val BLINK_TOTAL_LEN = 5.seconds
val BLINK_LEN = 250.milliseconds
var blinkOn by liveVarOf(false)
section {
scopedState {
if (blinkOn) invert()
textLine("This line will blink for ${BLINK_TOTAL_LEN.toSeconds()} seconds")
}
}.run {
var blinkCount = BLINK_TOTAL_LEN.toMillis() / BLINK_LEN.toMillis()
addTimer(BLINK_LEN, repeat = true) {
blinkOn = !blinkOn
blinkCount--
(blinkCount == ) {
repeat =
}
}
}
With timers running, it's possible your run block will exit while things are in a state you didn't intend (e.g. in the
above example with the blink effect still on). You should use the onFinishing event to handle this case:
var blinkOn by liveVarOf(false)
section {
/* ... */
}.onFinishing {
blinkOn = false // Because user might press Q while the blinking state was on
}.runUntilKeyPressed(Keys.Q) {
addTimer(250.milliseconds, repeat = true) { blinkOn = !blinkOn }
/* ... */
}
[!IMPORTANT] Unlike all the other events we discussed earlier,
onFinishingis registered directly against the underlyingsectionand not inside therunblock, because it is actually triggered AFTER the run pass is finished but before the block is torn down.
onFinishing will only run after all timers are stopped, so you don't have to worry about setting a value that an
errant timer will clobber later.
Animations make a huge difference for how the user experiences your application, so Kotter strives to make it trivial to add them into your program.
You can easily create quick animations by calling textAnimOf:
var finished = false
val spinnerAnim = textAnimOf(listOf("\\", "|", "/", "-"), milliseconds)
thinkingAnim = textAnimOf(listOf(, , , ), milliseconds)
section {
(!finished) { text(spinnerAnim) } { text() }
text()
(!finished) { text(thinkingAnim) } { text() }
}.run {
doExpensiveFileSearching()
finished =
}
When you reference an animation in a render for the first time, it kickstarts a timer automatically for you. In other words, all you have to do is treat your animation instance as if it were a string, and Kotter takes care of the rest!
If you have an animation that you want to share in a bunch of places, you can create a template for it and instantiate
instances from the template. TextAnim.Template takes exactly the same arguments as the textAnimOf method:
val SPINNER_TEMPATE = TextAnim.Template(listOf("\\", "|", "/", "-"), 250.milliseconds)
val spinners = (1..10).map { textAnimOf(SPINNER_TEMPLATE) }
/* ... */
If you need a bit more power than text animations, you can use a render animation instead. You create one with a callback that is given a frame index and access to the current render scope. You can interpret the frame index however you want and use the render scope to call any of Kotter's text rendering methods that you need.
Declare a render animation using the renderAnimOf method and then invoke the result inside your render block:
val exampleAnim = renderAnimOf(numFrames = 5, 250.milliseconds) { i -> /* ... */ }
section {
// Call your render animation passing in the section block (i.e. `this`) as a parameter
exampleAnim(this)
/* ... */
}
For example, let's say we want to rotate through a list of colors and apply those to some text. Text animations only
deal with raw text and don't have access to text effects like colors and styles. Therefore, we need to use a render
animation instead, giving us access to the color(Color) method:
// Note: `Color` is a Kotter enum that enumerates all the standard colors it supports
val colorAnim = renderAnimOf(Color.entries.size, 250.milliseconds) { i ->
color(Color.entries[i])
}
section {
colorAnim(this) // Side-effect: sets the color for this section
text("RAINBOW")
}.runUntilSignal { /* ... */ }
Both text and render animations can be created with a looping parameter set to false, if you only want them to run
once and stop:
val arrow = "=============>"
val wipeRightAnim = renderAnimOf(
arrow.length + 1, // `length + 1` because empty string is also a frame
40.milliseconds,
looping = false
) { frameIndex ->
textLine(arrow.take(frameIndex))
}
section {
text("Go this way: "); wipeRightAnim(this)
}.runUntilSignal {
// Give the animation time to complete:
addTimer(wipeRightAnim.totalDuration) { signal() }
}
You can restart a one-shot animation by setting its currFrame property back to 0.
Occasionally, when you want to render some marked up text, you'll wish you could measure it first, for example allowing you to pad both sides of each line with spaces to center everything, or putting the right count of "=" characters above and below a block of text to give it a sort of header effect. But by the time you've rendered something out, then it's too late to measure it!
offscreen to the rescue. You can think of offscreen as a temporary buffer to render to, after which you can both
query it and control when it actually renders to the screen.
offscreen returns a buffer, which is a read-only view of the content. You can query its raw text or line widths,
for example. To render it, you need to call offscreen.createRenderer and then use renderer.renderNextRow to render
out each line at a time.
[!CAUTION] Note that a line's width may be different from its string length, as many symbols, such as emojis and Asian characters, may be built out of many combined characters or take up double the space of a regular Latin letter.
Here, we use offscreen to render the header effect described above:
section {
val buffer = offscreen {
textLine("Multi-line"); textLine("Header"); textLine("Example")
}
val headerWidth = buffer.lineWidths.maxOrNull() ?: 0
val renderer = buffer.createRenderer()
repeat(headerWidth) { text('=') }; textLine()
while (renderer.hasNextRow()) {
renderer.renderNextRow()
textLine()
}
repeat(headerWidth) { text('=') }; textLine()
}.run()
[!NOTE] Although you usually won't need to, you can create multiple renderers per offscreen buffer, each which manages its own state for what row to render out next.
One nice thing about the offscreen buffer is it manages its own local state, and while it originally inherits its parent scope's state, any changes you make within the offscreen buffer will be remembered to its end.
This is easier seen than described. The following example:
section {
val buffer = offscreen {
textLine("Inherited color (red)")
cyan()
textLine("Local color (cyan)")
textLine("Still blue")
}
val renderer = buffer.createRenderer()
red()
while (renderer.hasNextRow()) {
text("red -- "); renderer.renderNextRow(); textLine(" -- red")
}
}.run()
will render:
The driving motivation for adding offscreen buffers was to be able to easily add borders around any block of text, where
the border might be a different color than its contents. So when this functionality went in, we also added the
bordered method (link to example).
If you want to implement your own utility method that uses offscreen under the hood, you can check
bordered's implementation
yourself to see how it delegates to offscreen, padding each row with the right number of spaces so that the border sides all line up.
You can actually make one-off render requests directly inside a run block:
section {
/* ... */
}.run {
aside {
textLine("Hello from an aside block")
}
}
which will output text directly before the active section.
In order to understand aside blocks, you should start to think of Kotter output as two parts -- some static history, and a dynamic, active area at the bottom. The static history will never change, while the active area will be written and cleared and rewritten over and over and over again as needed.
In general, a section is active until it is finished running, at which point it becomes static history, and the next section becomes active. You can almost think about consuming an active section, which freezes it after one final render, at which point it becomes static.
In fact, it's a common pattern to get static instructions out of the way first, in its own section, so we don't waste time rerendering them over and over in the main block:
session {
// The following instructions are static, just render them immediately
section {
textLine("Press arrow keys to move")
textLine("Press R to restart")
textLine("Press Q to quit")
textLine()
}.run()
section {
/* ... constantly rerendered lines ... */
}.runUntilKeyPressed(Keys.Q) { /* ... */ }
}
Occasionally, however, you want to generate static history while a block is still active.
Let's revisit an example from above, our FileWalker demo which searched a list of files and added every matching
result to a list. We can, instead, put a spinner in the active section and use the aside block to output matches:
val fileWalker = FileWalker(".")
var isFinished by liveVarOf(false)
val searchingAnim = textAnimOf(listOf(, , , ), milliseconds)
section {
textLine()
(!isFinished) {
textLine()
}
{
textLine()
}
}.run {
aside {
textLine()
textLine()
}
fileWalker.findFiles() { file ->
aside { textLine() }
}
isFinished =
}
Asides are very useful if you have some long-running process that generates text as a side effect. You could imagine a compiler spitting out warnings and errors as it continues to process more code, or a test runner reporting failures as it continues to run more tests. In fact, Kotter provides a fake compiler example that you can reference.
Kotter provides support for creating arbitrarily sized grids with multiple rows and columns.
With Kotter's approach to grids, you specify the number of columns explicitly; rows are auto added as you declare new grid cells.
section {
// A grid with two columns, each with space for 6 characters
grid(Cols(6, 6), characters = GridCharacters.CURVED) {
cell { // Auto set to row=0, col=0
textLine("Cell1")
}
cell {
textLine()
}
cell {
textLine()
}
cell {
textLine()
}
cell(row = , col = ) {
textLine()
}
}
}.run()
[!TIP] There is also a
Cols.uniformmethod for when you want to create multiple columns of the same width. For example, instead ofCols(6, 6)which we used above, you could also callCols.uniform(2, width = 6). It is a bit more verbose but may express intention more clearly.
You can check out the grid example for a more comprehensive example.
Fixed width columns are useful, but Kotter also provides even more functionality via fit- and star-sized columns.
A fit-sized column will check all cells it contains and choose a width that fits all of them.
A star-sized column will be sized dynamically based on how much space is remaining (more on this in a bit). If you have multiple star-sized columns, then space will be divided between them based on their ratio with each other.
For example, if you have one star-sized column set to "2" and another set to "1", then the first column will be twice as wide as the second. If you have two star-sized columns both set to "2" then they will share the remaining space equally.
To determine "remaining space", the grid method accepts a targetWidth parameter. If you don't have any star-sized
columns, the targetWidth value does nothing. If you do, then the grid will subtract all fixed and fit width values
from it and share any remaining space between the star-sized columns.
For a trivial example, say you have a two-column grid with targetWidth set to 10. The first column is fixed to 4, and
the second column is set to star-sized. The star-sized column will then receive 6 characters of space.
If you do not set the targetWidth at all, then all star-sized columns will shrink to size 1.
Earlier, we used Cols(6, 6), a convenience constructor that accepts only integer values indicating fixed column
widths. But for more control, you can construct the Cols class using a builder block:
grid(Cols { fit(); fixed(10); star() }, targetWidth = 80) {
/* ... */
}
In addition to their base value, columns have a few properties you can set: minimum value, maximum value, and justification.
Here's an example of setting all three properties:
grid(
Cols {
fit(maxValue = 10)
fixed(10, justification = Justification.CENTER)
star(minValue = 5)
},
targetWidth = 80
) {
/* ... */
}
The above means that the first column will be fit-sized but will never exceed 10 characters. The second column is fixed to 10 characters, and its contents will be centered. The final column is star-sized, but it will never be less than 5 characters.
You can declare that a cell should span multiple rows and/or columns by setting the rowSpan and colSpan parameters.
If either rowSpan or colSpan are not specified, then they default to 1.
[!CAUTION] If
colSpanis set to a value that would cause the cell to go out of bounds of the number of columns in this grid, an exception is thrown.
A few examples should help illustrate this:
// Spanning multiple columns
grid(Cols(3, 3, 3), characters = GridCharacters.CURVED) {
cell(row = 0, col = 0, colSpan = 3)
cell(row = 2) // Force three rows to be created
}
// Spanning multiple rows
grid(Cols(3, 3, 3), characters = GridCharacters.CURVED) {
cell(row = 0, col = 0, rowSpan = 3)
}
// Spanning both
grid(Cols(3, 3, 3, 3), characters = GridCharacters.CURVED) {
cell(row = 1, col = 1, rowSpan = 2, colSpan = 2)
cell(row = 3) // Force four rows to be created
}
[!CAUTION] A cell spanning columns will inherit its justification from its left-most cell. Additionally, any cell that spans multiple columns will not be included in any fit-size calculations.
When you declare a cell block without specifying a row or column, it will automatically be placed in the next empty
slot after the last cell that was declared. This goes from left-to-right, top-to-bottom.
For example:
grid(Cols(1, 1, 1)) {
cell(row = 1) { text("1") } // declared cell at row=1, col=0
cell(row = 0) { text("2") } // declared cell at row=0, col=0
cell { text("3") } // next empty slot is row=0, col=1
cell { text("4") }
cell { text("5") }
}
While the above example feels forced (it should be pretty rare to intentionally register cells out of order), the way cells flow is intuitive when used in conjunction with row spans:
grid(Cols(1, 1, 1)) {
cell(rowSpan = 2) { text("1") } // declared cell at row=0, col=0
cell { text("2") }
cell { text("3") }
cell { text("4") }
}
// +-+-+-+
// |1|2|3|
// | +-+-+
// | |4| |
// +-+-+-+
Finally, here's an example of how cells flow following a cell spanning multiple columns:
grid(Cols(1, 1, 1)) {
cell(colSpan = 2) { text("1") }
cell { text("2") }
cell { text("3") }
}
// +---+-+
// |1 |2|
// +-+-+-+
// |3| | |
// +-+-+-+
Terminal applications can be forcefully interrupted if the user presses CTRL-C. Some apps may want to handle this.
While you can register hooks for such an event directly with the underlying platform, Kotter offers an additional
API for this: addShutdownHook.
You can register a handler insider your run block, like so:
section { /* ... */ }.run {
addShutdownHook { /* this will get called only if the user presses ctrl-c */ }
}
Unlike a shutdown hook registered with the system, Kotter's managed shutdown hook will additionally try to give the render block one more chance to run before the program exits. You can take advantage of this to send a message to the user:
var emergencyShutdown by liveVarOf(false)
section {
/* ... */
if (emergencyShutdown) {
yellow {
textLine("This program is exiting NOW because the user pressed CTRL-C.")
textLine("We sent a request to shut down the server but could not confirm it was received.")
textLine("Consider running `./stop-server.sh` later to make sure it actually stopped.")
}
}
}.run {
addShutdownHook {
sendServerShutdownRequestAsync()
emergencyShutdown = true
}
}
It's important that you never run any long-running logic inside a shutdown hook. If your program continues to run for too long after an interrupt request, the system may just halt your program anyway.
Finally, you should not rely on shutdown hooks actually getting run. They don't get triggered if the system exits normally, the program crashes, or if the process gets aggressively halted by the OS (perhaps because things were taking too long to shut down, or maybe the user issued a kill command from the terminal).
Unicode is hard! But if you want to use non-Latin characters in your application, Kotter tries to handle it correctly.
The class that supports this functionality is exposed from the Kotter session via the textMetrics property, so you can
use it as well, anytime you need to measure text that may contain complex Unicode characters.
Kotter's grids and bordered text areas use text metrics so that their borders still line up correctly when containing
emojis or double-width, non-Latin characters. And the offscreen buffer uses it internally and exposes the final render
width values via the lineWidths property.
class TextMetrics {
fun :
:
:
:
}
[!NOTE] There is also a
renderWidthOfmethod provided that takes a singleCharvalue, but theStringversion is recommended as sometimes the render width of a character can change based on surrounding context, which is why we elide it here. But you can use it if you know what you're doing!
A grapheme is the final symbol that gets rendered to the screen. A grapheme cluster is the underlying characters
which are associated with the resulting grapheme. For basic text, the grapheme is often one-to-one with the underlying
character, but with emoji, that's often not the case. For example, the emoji 👨👩👧👦 is built up from 11 underlying
characters! Use graphemeCusterLengthAt with an index pointing at the beginning of a grapheme cluster to get this
value.
You can also use graphemeStartIndex to return the start of a cluster given any index inside said cluster. It is not
expected that most users will ever need this, but it can be useful if you want to iterate a Unicode string backwards one
grapheme at a time.
When a grapheme is rendered, it takes up either 1 or 2 spaces in the terminal grid. This is important to know when
measuring out how much text will fit inside a container. Use renderWidthOf to get this value.
Bringing it all together, if you wanted to iterate a text string that might contain some Unicode, the skeleton of your loop would generally look like this:
// Iterating forward
currIndex =
currWidth =
(currIndex < str.length) {
graphemeLen = textMetrics.graphemeClusterLengthAt(str, currIndex)
grapheme = textMetrics.substring(currIndex, currIndex + graphemeLen)
currWidth += textMetrics.renderWidthOf(grapheme)
currIndex += nextGraphemeLen
}
currIndex = string.length -
currWidth =
(currIndex >= ) {
graphemeStart = textMetrics.graphemeStartIndex(str, currIndex)
graphemeLen = textMetrics.graphemeClusterLengthAt(str, graphemeStart)
grapheme = textMetrics.substring(currIndex, graphemeStart + graphemeLen)
currWidth += textMetrics.renderWidthOf(grapheme)
currIndex = graphemeStart -
}
Kotter extends the core TextMetrics class with a useful method that lets you truncate text to ensure it fits within a
target render width:
fun TextMetrics.truncateToWidth(
text: String,
maxWidth: Int,
truncateAt: TruncateAt = TruncateAt.END,
ellipsis: String = ""
): String
By default, the method will just truncate your string if it doesn't fit, but you can also provide an optional ellipsis value. If present, the method will more "gently" cut off the end of the text (after securing any additional space needed for the ellipsis value itself)
// Truncate with 1-width ellipsis character
textMetrics.truncateToWidth("Hello world", maxWidth = 8, ellipsis = "…")
// Returns: "Hello w…"
// Truncate with 3-width ellipsis character
textMetrics.truncateToWidth("Hello world", maxWidth = 8, ellipsis = "...")
// Returns: "Hello..."
You can also specify that the truncation should happen at the start or middle of the string:
textMetrics.truncateToWidth("Hello world", maxWidth = 8, TruncateAt.END, ellipsis = "…")
// Returns: "Hello w…"
textMetrics.truncateToWidth("Hello world", maxWidth = 8, TruncateAt.START, ellipsis = "…")
// Returns: "…o world"
textMetrics.truncateToWidth("Hello world", maxWidth = 8, TruncateAt.MIDDLE, ellipsis = "…")
// Returns: "Hell…rld"
Ellipsis presets are provided for your convenience:
object EllipsisPresets {
const val NONE = ""
const val SYMBOL = "…"
const val PERIODS = "..."
}
textMetrics.truncateToWidth("Hello world", maxWidth = 8, ellipsis = EllipsisPresets.SYMBOL)
Kotter aims to provide all the primitives you need to write dynamic, interactive console applications, such as
textLine, input, offscreen, aside, onKeyPressed, etc.
But we may have missed your use case, or maybe you just want to refactor out some logic to share across sections.
This is totally doable, but it requires writing extension methods against the correct receiving classes. At this point,
we need to discuss the framework in a bit more detail than beginners need to know.
For reference, you should also look at the extend sample project, which was written to demonstrate some of the concepts that will be discussed here.
Before continuing, let's look at the overview of a Kotter application. The following may look a bit complex at first glance, but don't worry as the remaining subsections will break it down:
┌───────── session {
│ ┌─┬───── section {
│ │ │ ...
│ │ 3a┌─── offscreen {
│ │ │ 3b ...
│ │ │ └─── }
│ │ └───── }.onFinished {
1 2 ...
│ │ ┌───── }.run {
│ │ │ ...
│ │ 4 ┌─── aside {
│ │ │ 3c ...
│ │ │ └─── }
│ └─┴───── }
└───────── }
1 - Session
┌─ session {
│ section {
│ ...
│ }.run {
│ ...
│ }
└─ }
The top level of your whole application. Session owns a data: ConcurrentScopedData field which we'll talk more about
a little later. However, it's worth understanding that data lives inside a session, although many scopes additionally
expose it themselves. Any time you see some Kotter code interacting with a data field, it is really pointing back to
this singular one.
Session is the scope you need when you want to call liveVarOf or liveListOf, or even to declare a section:
fun Session.firstSection() {
var name by liveVarOf("")
var age by liveVarOf(18)
section { }.run { }
}
{ }
{ }
session {
firstSection()
secondSection()
thirdSection()
}
2 - Section
┌─ section {
│ ...
│ }.run {
│ ...
└─ }
The section { ... } block receives a RenderScope and the run { ... } block receives a RunScope. These are
discussed next.
3 - RenderScope
3a
┌─ section {
│ ...
└─ }
3b
┌─ offscreen {
│ ...
└─ }
3c
┌─ aside {
│ ...
└─ }
This scope represents a render pass. This is the scope that owns textLine, red, green, bold, underline, and
other text rendering methods.
This can be a useful scope for extracting out a common text rendering pattern. For example, let's say you wanted to display a bunch of terminal commands and their arguments, and you want to highlight the command a particular color, e.g. cyan:
section {
cyan { text("cd") }; textLine(" /path/to/code")
cyan { text("git") }; textLine(" init")
}
That can get a bit repetitive. Plus, if you decide to change the color later, or maybe bold the command part, you'd have to do it all over the place. Also, putting the logic behind a function can help you express your intention more clearly.
Let's refactor it!
fun RenderScope.shellCommand(command: String, arg: String) {
cyan { text(command) }; textLine(" $arg")
}
section {
shellCommand("cd", "/path/to/code")
shellCommand("git", "init")
}
Much better!
On its surface, the concept of a RenderScope seems pretty straightforward, but the gotcha is that Kotter offers a few
separate areas that accept render blocks. In addition to the main rendering block, there are also offscreen and
aside blocks, which allow rendering to different targets (these are discussed earlier in the README in case you aren't
familiar with them here).
Occasionally, you may want to define an extension method that only applies to one of the three blocks (usually the
main one). In order to narrow down the place your helper method will appear in, you can use MainRenderScope (3a),
OffscreenRenderScope (3b), and/or AsideRenderScope (3c) as your method receiver instead of just RenderScope.
For example, the input() method doesn't make sense in an offscreen or aside context (as those aren't interactive).
Therefore, its definition looks like fun MainRenderScope.input(...) { ... }
4 - RunScope
section {
...
┌─ }.run {
│ ...
└─ }
RunScope is used for run blocks. It's useful for extracting logic that deals with handling user input or
long-running tasks. Functions like onKeyPressed, onInputChanged, addTimer etc. are defined on top of this scope.
fun RunScope.exec(vararg command: String) {
val process = Runtime.getRuntime().exec(*command)
process.waitFor()
}
section {
textLine("Please wait, cloning the repo...")
}.run {
exec("git", "clone", "https://github.com/varabyte/kotter.git")
/* ... */
}
As the run method is a suspend function, you may declare your RunScope extending methods suspend as well:
suspend fun RunScope.doLongRunningTask() { /* ... */ }
section {
textLine("Please wait...")
}.run {
doLongRunningTask()
}
SectionScope
To close off all this scope discussion, it's worth mentioning that a SectionScope interface exists. It is the base
interface to both RenderScope AND a RunScope, and using it can allow you to define the occasional helper method that
can be called from both of them.
It's not expected that most users will ever use this, but it can be a way to write a common getter that both the render block and run block can use (perhaps for data that is set by the run block elsewhere).
The one thing that all scopes have in common is they expose access to a session's data field. OK, but what is it?
ConcurrentScopedData is a thread-safe hashmap, where the keys are always of type ConcurrentScopedData.Key<T>, and
such keys are associated with a ConcurrentScopedData.Lifecycle (meaning that any data you register into the map will
always be released when some parent lifecycle ends, unless you remove it yourself manually first).
Kotter itself manages four lifecycles: Session.Lifecycle, Section.Lifecycle, MainRenderScope.Lifecycle, and
RunScope.Lifecycle (each associated with the scopes discussed above).
[!NOTE] No lifecycles are provided for
offscreenorasideblocks at the moment. Feel free to open up an issue with a use-case requiring additional lifecycles if you run into one. So far, theRunScope.Lifecyclehas been effective enough for our needs.
Keep in mind that the MainRenderScope.Lifecycle dies after a single render pass. Almost always you want to tie data
to Section.Lifecycle, as it survives across multiple runs.
Nothing prevents you from defining your own lifecycle - just be sure to call data.start(...) and data.stop(...) with
it.
object MyLifecycle : ConcurrentScopedData.Lifecycle
private val MySetting = MyLifecycle.createKey<Boolean>()
try {
data.start(MyLifecycle)
data[MySetting] = true
/* ... */
} finally {
data.stop(MyLifecycle) // Side effect: Removes MySetting from data
}
Lifecycles can be defined as subordinate to other lifecycles, so if you create a lifecycle that is tied to the
RunScope lifecycle for example, then you don't need to explicitly call stop yourself (but you still need to call
start).
object MyLifecycle : ConcurrentScopedData.Lifecycle {
override val parent = RunScope.Lifecycle
}
section { /* ... */ }.run {
data.start(MyLifecycle) // Will be stopped when the run block finishes
}
You can review the ConcurrentScopedData class for its full list of documented API calls, but the three common ways to
add values to it are:
data[key] = valuedata.tryPut(key, value) // returns true if added, false otherwisedata.putIfAbsent(key, provideInitialValue = { value }) { ... logic using value ... }By having a session own and expose such a data structure, it makes it possible for anyone to write their own extension
methods on top of Kotter, using data as a way to manage long-lived state. For example, input(), which may get called
many times in a row as the section rerenders, can distinguish the first time it is called from later calls based on
whether some value is present in the data store or not.
To close this section, we just wanted to say that it was very tempting at first to create a bunch of hardcoded functions
baked inside Section, MainRenderScope, etc., with access to some private state, but implementing everything through
ConcurrentScopedData plus extension methods ensured that we were using the same tools as users.
So go forth, and extend Kotter!
Kotter includes a separate library that provides useful testing utilities, called kotter-test-support. You can review
the Gradle section▲ from earlier to see how to include it in your project.
The library comes with its own README which goes into more detail about how to write Kotter unit tests.
Sections are rendered sequentially on a single render thread. Anytime you see a section { ... }, no matter which
thread it is called from, a single thread ultimately handles it. However, if you use two threads to attempt to run one
section while another is already running, an exception is thrown.
The run block runs in place on the thread that called it. In this way, progress is prevented until the run logic
finishes running.
I made the decision to lock section rendering down so that:
printlning at the same time - who likes clobbered text?In practice, I expect this decision won't be an issue for most users. Command line apps are expected to have a main flow
anyway -- ask the user a question, do some work, then ask another question, etc. It is expected that a user won't ever
even need to call section from more than one thread. It is hoped that the
session {
section { /* ... render thread ... */ }.run { /* ... current thread ... */ }
section { /* ... render thread ... */ }.run { /* ... current thread ... */ }
section { /* ... render thread ... */ }.run { /* ... current thread ... */ }
}
pattern (just calling sections one after another on a single thread) is powerful enough for most (all?) cases.
[!IMPORTANT] The virtual terminal is only supported for JVM targets. Kotlin/Native targets don't implement this feature.
It's not guaranteed that your program will be called in a terminal that allows interactivity. Kotter, when first run, requests a specific configuration from your system that may be rejected.
For example, pressing the debug button on your project inside IntelliJ as well as calling ./gradlew run are two very
common scenarios! In those cases, IntelliJ/Gradle are already consuming the terminal's interactivity themselves, so when
your program tries to run, it does so inside a more limited environment.
Kotter can (and by default is setup to) detect this case and open up a virtual terminal instead. This fallback gives your application better cross-platform support.
To modify your program to ALWAYS open the virtual terminal, you can set the terminal parameter in session like so:
session(terminal = VirtualTerminal.create()) {
section { /* ... */ }
/* ... */
}
or you can chain multiple factory methods together using the firstSuccess method, which will try to start each
terminal type in turn:
session(
terminal = listOf(
{ SystemTerminal() },
{ VirtualTerminal.create(title = "My App", terminalSize = Dimension(30, 30)) },
).firstSuccess()
) {
/* ... */
}
Our team produces the com.varabyte.kotterx:twemoji artifact, a dependency which provides rendering support for emoji
characters using SVGs provided by the Twemoji (Twitter Emoji) project.
[!IMPORTANT] Above, we mentioned that the virtual terminal is a JVM-only concept. As such, this dependency is only provided for the JVM target.
The benefit of using an emoji library instead of system fonts is a consistent look and feel whether you are using Mac, Windows, or Linux. It also seems to provide more complete emoji support than the fonts do.
All you have to do is add a dependency on the Kotterx twemoji coordinate in you're build script, and it will
automatically be picked up by the virtual terminal if it opens up.
// JVM project
plugins {
kotlin("jvm")
}
dependencies {
implementation("com.varabyte.kotter:kotter:1.3.0")
implementation("com.varabyte.kotterx:twemoji:1.3.0")
}
// Mulitplatform project
plugins {
kotlin("multiplatform")
}
kotlin {
jvm()
// optionally other targets
sourceSets {
commonMain {
dependencies {
implementation("com.varabyte.kotter:kotter:1.3.0")
}
}
jvmMain {
dependencies {
implementation("com.varabyte.kotterx:twemoji:1.3.0")
}
}
}
}
On the left is a screenshot of my virtual terminal running on MacOS using the default system font; on the right, Twemoji. Some reports indicate that the left side looks even more dismal on Windows.
The biggest drawback for including this dependency is it adds about 4MB of data to your application bundle.
We don't include this as part of core Kotter since it's assumed most users will want to write an application that targets a system terminal and hopefully will never even open up the virtual terminal, so why pay the memory cost? Additionally, allowing an external module to control emoji rendering opens us up to supporting alternate emoji sets in the future.
Kotter is designed to only run within a rich, interactive terminal environment, such as a console with ANSI support or inside a virtual terminal.
However, there may be cases that Kotter won't be able to run. For example:
VirtualTerminal.VirtualTerminal
implementation).If you run a console app in such a non-interactive environment, calls like println and readLine still work, but any
attempts to move the cursor, erase previous lines, and many other ANSI commands which Kotter builds on top of are not
supported.
Such cases are increasingly rare on modern machines, so you may just decide to ignore them and crash!
However, if you're very determined, you could consider writing some parts of your program twice, once with Kotter using all its fancy bells and whistles, and a second time providing a much simpler presentation, limited to printing information and asking the occasional question.
To accomplish this, you can use the following code structure:
For a concrete example, imagine you are writing a file downloading app. You can have the Kotter version show animated progress bars, but if you end up in the fallback zone, you can simply print "10%... 20%... 30%..." instead. Both sections could delegate to some downloader class that did all the heavy lifting -- you should absolutely share as much non-UI code as you can!
You finished your Kotter application. Congratulations!! 🎉
Now what? How do you get your amazing program to your users?
Let's explore a few options.
Pros:
Cons:
If your application targets the JVM, you can easily build zip and tar files of your project using the assembleDist
task:
$ cd yourkotterapp
$ ./assembleDist
$ cd build/distributions
$ ls
yourkotterapp-version.tar yourkotterapp-version.zip
You can ask your user to download either file, extract it, and then run the program under the bin folder:
$ unzip yourkotterapp-version.zip
$ ./yourkotterapp-version/bin/yourkotterapp
Pros:
Cons:
For a concrete example, let's consider Homebrew, a very popular package manager. They're one of the easier ones to
support -- you can create a custom repo that declares a manifest (example here).
Once set up, a user can install / update your software simply by running: brew install yourkotterapp
Notice how that manifest declares JDK11 as a dependency. You'll need to figure out how to do that with each package manager you decide to support.
[!NOTE] I have a project that uses JReleaser to publish my program to several package managers with a single Gradle publish task. If you decide to try JReleaser in your own project, you can review my jreleaser block in this build.gradle.kts.
Pros:
Cons:
Configuring a Kotter/Native application is relatively easy but outside the scope of this document. Start with the official docs and review the native example for guidance.
Once your Kotlin/Native project is set up, you can build it using ./gradlew linkDebugExecutable[Host] (or
./gradlew linkReleaseExecutable[Host]) which puts a binary under ./build/bin/[host]/debugExecutable/native.exe
(or native.kexe on Linux).
Unless you already have Mac, Windows, and Linux machines at home, you may want to use a CI to build binaries for you. How to do this is also outside the scope of this document, but I personally use GitHub Actions to handle this, creating a workflow that runs on several machines, publishing artifacts based on which runner is active. You can read more about GitHub CI strategies here.
For example, you could set up a workflow to link various binaries into executables whenever a new commit is pushed to the main branch. You could then use the upload artifact action to push binaries to a location you can download from later.
[!NOTE] For reference, you may wish to refer to Kotter's publishing workflow script. It doesn't use the upload action, but you can see how we run multiple target hosts in order to build all the different flavors of artifacts.
Once built, you can share your binaries with your users, either from cloud storage somewhere or by publishing it via a package manager (as discussed above ▲).
jlink / jpackage to convert your JVM project into a binaryPros:
Cons:
[!WARNING] This section is incomplete as I have not found time to try out these tools yet. However, any readers familiar with them are welcome to contact me with information so that I can update this section.
jlink, introduced in Java 9, allows you to assemble modules into a custom runtime image, which can significantly
reduce the final size of a runtime you'd want to ship (as you'd be excluding a bunch of standard library code you don't
need). Official docs here.
jpackage, introduced in Java 14, allows you to bundle a JVM program plus a runtime (e.g. produced by jlink) into
a final binary + installer, one per platform. Official docs here.
JReleaser, discussed in a previous section, exposes support for jlink and jpackage configurations.
If you don't need access to the JVM library ecosystem, using Kotlin/Native ▲ is probably easier.
Pros:
Cons:
[!WARNING] This section is incomplete as my own experimentation fell a bit short with it. However, GraalVM is a very promising technology, so if anyone reading this knows how to get this solution to work, please contact me and I can update this section.
GraalVM is a high-performance JDK distribution, while GraalVM Native Image is an ahead-of-time compiler for Java programs. What this means to you is that you can build a Java application jar and then target it with native image to convert it into a standalone binary.
You may be able to further minify your GraalVM output with UPX, which may be able to shrink your final binary size dramatically.
[!IMPORTANT] If you decide to try using GraalVM on your project, you should strongly consider excluding the virtual terminal by overriding the default
terminalparameter when creating a session:session(terminal = SystemTerminal.create()) { /* ... */ }This allows GraalVM to strip out all Swing code, which is otherwise very tricky to configure.
Kotter's API is inspired by Compose, which astute readers may have already noticed -- it has a core block which gets rerun for you automatically as necessary without you having to worry about it, and special state variables which, when modified, automatically "recompose" the current console block. Why not just use Compose directly?
In fact, this is exactly what Jake Wharton's Mosaic is doing. Actually, I tried using it first but ultimately decided against it before deciding to write Kotter, for the following reasons:
Mosaic and Kotter programs look very similar, but they are organized slightly differently. Instead of Compose, where you
have a single code block where values, layout, and logic are combined (with judicious use of remember and
LaunchedEffect), in Kotter you tend to have three separate areas for these concepts: before a section, a section
block, and a run block.
// Mosaic
runMosaic {
var count by remember { mutableStateOf(0) }
Text("The count is: $count")
LaunchedEffect() {
(i ) {
delay()
count++
}
}
}
session {
count liveVarOf()
section {
textLine()
}.run {
(i ) {
delay()
count++
}
}
}
Comparisons with Mosaic are included in the examples/mosaic folder.
Mordant is another Kotlin console API. And honestly, it's awesome. It's worth a look!
Mordant provides an API where you instantiate a Terminal object and issue commands to it directly. It's very clean!
The library also provides markdown support and builders for complex tables, which are really nice features that don't
currently exist in Kotter. It has a few other opinionated components, such as an animated progress bar and an input
prompter that requires the answer be one of a few choices.
If you are mostly rendering output text, Mordant may honestly result in more streamlined code. It also handles falling back better if you run it inside a terminal that does not support interactive mode. (Kotter will open a virtual terminal if it can, or crash if it can't).
You may still prefer using Kotter for cases where you plan to have a lot of interactive elements, such as several animations running side by side in parallel, or if you want keypress handling, or if you want to want the ability to manage timers, or if you want to show interactive prompts with rich auto-completion behavior.
Additionally, Kotter's split between the section block and run blocks benefit more and more in increasingly complex
scenarios. It can be nice to have a clear separation of the rendering logic from background computation logic.
Still, for concreteness, let's take a few examples from Mordant's README and show them side-by-side with equivalent Kotter implementations:
Declaring basic styles
// Mordant
val t = Terminal()
t.println("${red("red")} ${white("white")} and ${blue("blue")}")
// Kotter
section {
red { text("red") }; text(' ')
white { text("white")}; text(" and ")
blue { textLine("blue") }
}.run()
Nesting styles
// Mordant
t.println(white("You ${(blue on yellow)("can ${(black + strikethrough)("nest")} styles")} arbitrarily"))
// Kotter
section {
white {
text("You ")
blue { yellow(BG) {
text("can ")
black { strikethrough {
text("nest")
}}
text(" styles")
}}
textLine(" arbitrarily")
}
}.run()
Reusing styles
// Mordant
val style = (bold + white + underline)
t.println(style("You can save styles"))
t.println(style("to reuse"))
// Kotter
fun RenderScope.withStyle(block: () -> Unit) {
bold { white { underline { block() }}}
}
section {
withStyle { textLine("You can refactor styles") }
withStyle { textLine("to reuse") }
}.run()
Animating a horizontal bar
Prompting for input
// Mordant
val t = Terminal()
val response = t.prompt("Choose a size", choices=listOf("small", "large"))
t.println()
choices = listOf(, )
choice: String =
section {
text()
input(Completions(*choices))
}.runUntilInputEntered {
onInputEntered {
(input ! choices) rejectInput() choice = input
}
}
section {
text()
}.run()
The above samples definitely look really nice in Mordant, and if those cases capture the main sort of functionality you were planning to use in your own app, Mordant may be the better API for your project.
Meanwhile, for examples that respond to user input like snake, or which do a lot of clearing / repainting like doomfire, or which query for input in the middle of a bunch of other text like wordle, Kotter may be the better choice in those cases.
And finally, it's possible to use Kotter and Mordant together. For example, referring back to the
fallback section above ▲, you can use Mordant in the fallback block, since
it provides a friendlier API than raw println/readLine calls.
// build.gradle.kts (kotlin script)
plugins {
kotlin("jvm")
application
}
repositories {
mavenCentral()
}
dependencies {
implementation("com.varabyte.kotter:kotter:1.3.0")
testImplementation("com.varabyte.kotterx:kotter-test-support:1.3.0")
}
application {
applicationDefaultJvmArgs = listOf(
// JDK24 started reporting warnings for libraries that use restricted native methods, at least one which Kotter
// uses indirectly (via jline/jansi). It looks like this:
//
// WARNING: A restricted method in java.lang.System has been called
// WARNING: java.lang.System::loadLibrary has been called by ...
// WARNING: Use --enable-native-access=ALL-UNNAMED to avoid a warning for callers in this module
// WARNING: Restricted methods will be blocked in a future release unless native access is enabled
//
// The best solution we have for now is to disable the warning by explicitly enabling access.
// We also suggest the IgnoreUnrecognizedVMOptions flag here to allow kotter applications to be able to compile
// with JDKs older than JDK24. You can remove it if you are intentionally using JDK24+.
// See also: https://docs.oracle.com/en/java/javase/24/docs/api/java.base/java/lang/doc-files/RestrictedMethods.html
// And also: https://github.com/jline/jline3/issues/1067
"-XX:+IgnoreUnrecognizedVMOptions",
"--enable-native-access=ALL-UNNAMED",
)
// The following assumes a top-level `main.kt` file in your project; adjust as needed otherwise
mainClass.set("MainKt")
}
tryPut and then getting the value, but doing so in a lock-safe manner
that ensures no one else grabs the thread from you in between.private fun Session.runKotterLogic() { /* ... */ }
private fun runFallbackLogic() { /* ... */ }
fun main() {
var kotterStarted = false
try {
session {
kotterStarted = true
runKotterLogic()
}
} catch (ex: Exception) {
if (!kotterStarted) {
runFallbackLogic()
} else {
// This exception came from after startup, when the user was
// interacting with Kotter. Crashing with an informative stack
// is probably the best thing we can do at this point.
throw ex
}
}
}
Compose is tightly tied to the current Kotlin compiler version, which means if you are targeting a particular
version of the Kotlin language, you can easily see the dreaded error message: This version (x.y.z) of the Compose Compiler requires Kotlin version a.b.c but you appear to be using Kotlin version d.e.f which is not known to be compatible.
Compose is great for rendering a whole, interactive UI, but console printing is often two parts: the active part that the user is interacting with, and the history, which is static. To support this with Compose, you'd need to manage the history list yourself and keep appending to it, which would be a waste of render cycles and memory when you could just lean on the console to do it. It was while thinking about an API that addressed this limitation that I envisioned Kotter.
Compose encourages using a set of powerful layout primitives, namely Box, Column, and Row, with margins and
shapes and layers. Command line apps don't really need this level of power, however.
Compose has a lot of strengths built around, well, composing methods! But for a simple CLI library, being able to focus on simple render blocks that don't nest at all allowed a more pared down API to emerge.
Compose does a lot of nice tricks due to the fact it is ultimately a compiler plugin, but it is interesting to see what the API could look like with no magic at all (although, admittedly, with some features sacrificed).
Mosaic doesn't support input well yet (at the time of writing this README, but maybe this has changed in the future). For example, compare Mosaic to Kotter.
// Mordant
val t = Terminal()
val a = t.textAnimation<Int> { frame ->
(1..50).joinToString("") {
val hue = (frame + it) * 3 % 360
t.colors.hsv(hue, 1, 1)("━")
}
}
t.cursor.hide(showOnExit = true)
repeat(120) {
a.update(it)
Thread.sleep(25)
}
// Kotter
val barAnim = renderAnimOf(numFrames = 120, 25.milliseconds) { frame ->
for (i in 1..50) {
val hue = ((frame + i) * 3) % 360
hsv(hue, 1f, 1f) { text("━") }
}
}
section {
barAnim(this)
}.runFor(3.seconds)
Surfaced from shared tags and platforms — no rankings paid for.