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:
- Setting colors and text decorations (e.g. underline, bold)
- Handling user input
- Creating timers and animations
- Seamlessly repainting terminal text when values change
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.
// build.gradle.kts (kotlin script)
plugins {
kotlin("jvm")
}
repositories {
mavenCentral()
}
dependencies {
implementation("com.varabyte.kotter:kotter-jvm:1.1.2")
testImplementation("com.varabyte.kotterx:kotter-test-support-jvm:1.1.2")
}
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(), // Linux
mingwX64(), // Windows
macosArm64(), // Mac M1
macosX64(), // Mac Legacy
).forEach { nativeTarget ->
nativeTarget.apply {
binaries {
executable {
entryPoint = "main"
}
}
}
}
sourceSets {
val commonMain by getting {
dependencies {
implementation("com.varabyte.kotter:kotter:1.1.2")
}
}
val commonTest by getting {
dependencies {
implementation("com.varabyte.kotterx:kotter-test-support:1.1.2")
}
}
}
}
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://s01.oss.sonatype.org/content/repositories/snapshots/")
// ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
}
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 section
s. Your own app may only ever contain a single section
and that's
fine! But if you have multiple section
s, 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 calling scopedState
for 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 liveVarOf
method is provided by the session
block. For many remaining examples, we'll elide the
session
boilerplate, 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
var numFilledSegments by liveVarOf(0)
section {
text("[")
for (i in 0 until BAR_LENGTH) {
text(if (i < numFilledSegments) "*" else "-")
}
text("]")
}.run {
var percent = 0
while (percent < 100) {
delay(Random.nextLong(10, 100))
percent += Random.nextInt(1,5)
numFilledSegments = ((percent / 100f) * 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 signal
before you reach waitForSignal
, then waitForSignal
will 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 { input = input.filter { it.isLetter() } }
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 { input = input.filter { it.isLetter() } }
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) }
// ^^^^^^^^^^^^^^^^^^^^
// 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.
viewMap
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
customFormat
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--
if (blinkCount == 0L) {
repeat = false
}
}
/* ... */
}
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, onFinishing
is registered directly against the underlying
section
and not inside the run
block, 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("\\", "|", "/", "-"), 125.milliseconds)
val thinkingAnim = textAnimOf(listOf("", ".", "..", "..."), 500.milliseconds)
section {
if (!finished) { text(spinnerAnim) } else { text("✓") }
text(" Searching for files")
if (!finished) { text(thinkingAnim) } else { text("... Done!") }
}.run {
doExpensiveFileSearching()
finished = true
}
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.values().size, 250.milliseconds) { i ->
color(Color.values()[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 lengths,
for example. To render it, you need to call offscreen.createRenderer
and then use renderer.renderNextRow
to render
out each line at a time.
Here, we use offscreen
to render the header effect described above:
section {
val buffer = offscreen {
textLine("Multi-line"); textLine("Header"); textLine("Example")
}
val headerLen = buffer.lineLengths.maxOrNull() ?: 0
val renderer = buffer.createRenderer()
repeat(headerLen) { text('=') }; textLine()
while (renderer.hasNextRow()) {
renderer.renderNextRow()
textLine()
}
repeat(headerLen) { 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("", ".", "..", "..."), 500.milliseconds)
section {
textLine()
if (!isFinished) {
textLine("Searching$searchingAnim")
}
else {
textLine("Finished searching")
}
}.run {
aside {
textLine("Matches found so far:")
textLine()
}
fileWalker.findFiles("*.txt") { file ->
aside { textLine(" - ${file.name}") }
}
isFinished = true
}
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 { // Auto set to row=0, col=1
textLine("Cell2")
}
// Third cell in a grid with two columns creates a new row
cell { // Auto set to row=1, col=0
textLine("Cell3")
}
cell { // Auto set to row=1, col=1
textLine("Cell4")
}
// You can explicitly set the row and column if you want. Rows and columns
// are 0-indexed.
cell(row = 2, col = 1) { // Jump over cell row=2,col=0
textLine("Cell6")
}
}
}.run()
Tip
There is also a Cols.uniform
method for when you want to create multiple columns of the same width. For example,
instead of Cols(6, 6)
which we used above, you could also call Cols.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 colSpan
is 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 = 2) // Force three rows to be created
cell(row = 0, col = 0, colSpan = 3)
}
// 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 = 3) // Force four rows to be created
cell(row = 1, col = 1, rowSpan = 2, colSpan = 2)
}
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") }
}
// +-+-+-+
// |2|3|4|
// +-+-+-+
// |1|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).
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 section
s.
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 { /* ... */ }
}
fun Session.secondSection() { /* ... */ }
fun Session.thirdSection() { /* ... */ }
// Later ...
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 also 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
Run.Lifecycle
(each associated with the scopes discussed above).
Note
No lifecycles are provided for offscreen
or aside
blocks at the moment. Feel free to open up an issue with a
use-case requiring additional lifecycles if you run into one.
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 Run
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 = Run.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:
- always overwrite:
data[key] = value
- add if first time:
data.tryPut(key, value) // returns true if added, false otherwise
- add if first time but always run some follow-up logic:
data.putIfAbsent(key, provideInitialValue = { value }) { ... logic using value ... }
- This is essentially a shortcut for calling
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.
- This is essentially a shortcut for calling
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 cache 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:
- I don't have to worry about multiple sections
println
ing at the same time - who likes clobbered text? - Kotter handles repainting by moving the terminal cursor around, which would fail horribly if multiple sections tried doing this at the same time.
- Kotter embraces the idea of a dynamic, active section preceded by a bunch of static history. If two dynamic blocks wanted to be active at the same time, what would that even mean?
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 section
s 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 run in an interactive way, or even that you won't be called in a legacy terminal (e.g. on Windows) that doesn't support ANSI virtual codes.
For example, debugging this project with IntelliJ as well as running within Gradle are two such environments where interactivity isn't available! Since in that case, IntelliJ/Gradle are already consuming the interactivity themselves, and running your program in a more limited environment.
Kotter will attempt to detect if your console does not support the features it uses, and if not, it will open up a virtual terminal instead. This fallback gives your application better cross-platform support.
To modify the logic to ALWAYS open the virtual terminal, you can set the terminal
parameter in session
like
this:
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()
) {
/* ... */
}
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:
- You're trying to run Kotter in an environment that is both non-interactive AND doesn't have any graphical system enabled (like ssh-ing to some remote machine).
- You are in a non-interactive environment, and you constructed your session in a way that excludes using a
VirtualTerminal
. - You are in a non-interactive environment, and you are using Kotlin/Native (which does not provide a
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:
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
}
}
}
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:
- Trivial to do.
- Easy to share.
- Access to the whole JVM ecosystem.
Cons:
- User must extract files on their machine, a step they aren't used to worrying about.
- User must have Java installed.
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:
- Trivial for the user to install (although this may install Java as a side effect).
- Access to the whole JVM ecosystem.
Cons:
- Everyone has their favorite package manager, and you won't be able to satisfy everyone.
- Some package managers take a lot of effort to set up.
- Some users won't want to use a package manager, so you'll need to include other options anyway.
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:
- No Java required on the user's machine.
- Potentially small final binary size.
- Relatively easy to set up.
Cons:
- Will require multiple host machines if you want to build binaries for all platform targets.
- No access to the broader JVM ecosystem.
- Native targets don't provide a virtual terminal.
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 ▲).
Pros:
- No Java required on the user's machine.
- Access to the whole JVM ecosystem.
Cons:
- Requires JDK14 or newer.
- Will require multiple host machines if you want to build binaries for all platform targets.
- May be complicated to set up.
- May require your CI have access to two different JDKs, one for compiling your code and one for running the jpackage step, in case you are intentionally compiling your code with an older JDK version.
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:
- No Java required on the user's machine.
- Great flexibility. User can use the entire JVM library ecosystem while still producing a binary that can run anywhere.
- Potentially small final binary size (after using UPX).
Cons:
- Will require multiple host machines if you want to build binaries for all platform targets.
- GraalVM can be annoying to install.
- GraalVM can be very fussy.
- Can be frustrating to chase down runtime exceptions caused by a misconfigured compile.
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 terminal
parameter 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:
-
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.
- Using Kotlin v1.3 or older for some reason? You're out of luck.
- Want to upgrade Kotlin without updating Kotter? You're out of luck.
- Want to update Kotter without also upgrading Kotlin? You might be out of luck.
-
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.
- For a concrete example, see the compiler demo. A compiler might generate hundreds (or thousands!) of history lines. We definitely don't want to rerender all of those every frame.
-
Compose encourages using a set of powerful layout primitives, namely
Box
,Column
, andRow
, 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.
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(null) {
for (i in 1..20) {
delay(250)
count++
}
}
}
// Kotter
session {
var count by liveVarOf(0)
section {
textLine("The count is: $count")
}.run {
for (i in 1..20) {
delay(250)
count++
}
}
}
Comparisons with Mosaic are included in the examples/mosaic folder.
Mordant is another Kotlin console API. And honestly, it's awesome. It's worth a look!
Mordant provides an API where you instantiate a Terminal
object and issue commands to it directly. It's very clean!
The library also provides markdown support and builders for complex tables, which are really nice features that don't
currently exist in Kotter. It has a few other opinionated components, such as an animated progress bar and an input
prompter that requires the answer be one of a few choices.
If you are mostly rendering output text, Mordant may honestly result in more streamlined code. It also handles falling back better if you run it inside a terminal that does not support interactive mode. (Kotter will open a virtual terminal if it can, or crash if it can't).
You may still prefer using Kotter for cases where you plan to have a lot of interactive elements, such as several animations running side by side in parallel, or if you want keypress handling, or if you want to want the ability to manage timers, or if you want to show interactive prompts with rich auto-completion behavior.
Additionally, Kotter's split between the section
block and run
blocks benefit more and more in increasingly complex
scenarios. It can be nice to have a clear separation of the rendering logic from background computation logic.
Still, for concreteness, let's take a few examples from Mordant's README and show them side-by-side with equivalent Kotter implementations:
Multiple styles
// Mordant
val t = Terminal()
t.println("${red("red")} ${white("white")} and ${blue("blue")}")
// Kotter
section {
red { text("red") }; text(' ')
white { text("white")}; text(" and ")
blue { textLine("blue") }
}.run()
Nest 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()
Reuse 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
// 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)
Prompting for input
// Mordant
val t = Terminal()
val response = t.prompt("Choose a size", choices=listOf("small", "large"))
t.println("You chose: $response")
// Kotter
val choices = listOf("small", "large")
var choice: String = ""
section {
text("Choose a size [${choices.joinToString()}]: "); input(Completions(*choices))
}.runUntilInputEntered {
onInputEntered {
if (input !in choices) rejectInput() else choice = input
}
}
section {
text("You chose: $choice")
}.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.