Kotlin Cheatsheet - Kotlin Syntax & Coroutine Reference

The priority is clear. ?. is the default: name?.length evaluates to null for the whole expression instead of throwing. ?: supplies a fallback, and val len = name?.length ?: 0 chains the two together — the most common null-safe idiom in a call chain. !! is a lie told to the compiler: it asserts non-null and throws NPE if that turns out to be false, converting a compile-time check into a runtime crash. Justify !! only when a non-null check happens on the immediately preceding line or the value is guaranteed by an immutable construction; everywhere else ?. or ?: does the job.

Only the return value differs. launch gives back a Job, meaning "this coroutine started and finished" — no result. async gives back a Deferred, which you can await to obtain the value. The rule is plain: use launch when nobody consumes the coroutine body's result; use async when you need a return value and genuinely want to run several concurrently and collect them. Both start on the current context by default, so wrap blocking work in withContext(Dispatchers.IO) { ... }. In tests, use runBlocking { ... } to let the coroutine finish before asserting rather than really sleeping with delay(1000).

Two dimensions decide it: whether the receiver is it or this, and whether you get the lambda result back or the object back. let { it.foo() } and with(obj) { foo() } both return the lambda result, with let using it for a chained transformation. apply { this.x = 1 } returns the object itself and its receiver is this, which makes it the right pick for initialising a batch of properties. also { it.log() } also returns the original object but binds it, so you can slot in a side effect and keep chaining. run { compute() } is this-receiver plus returned result, handy for calling several members on a non-null object. Mnemonic: apply to initialise, let to transform, also for side effects.

first() throws NoSuchElementException when nothing matches, while firstOrNull { it.id == 1 } hands back null, so the second is the safe choice whenever the data may be incomplete — and it composes naturally with a following null check. Alongside it sit the one-pass collection operations: groupBy { it.category } buckets by any key into a Map, partition { it > 0 } splits in a single pass and returns a Pair (first = matching, second = not), distinct() deduplicates by equals, take(5) with drop(3) implements paging, and sortedBy { it.age } returns ascending order that reversed can flip. All of these return new collections and leave the original untouched, so watch the memory cost of intermediates on large data.

A sealed class Result models a restricted hierarchy: every subclass lives in the same file, and when used with a when expression the compiler checks that branches are exhaustive, so you need no else. That makes it ideal for mutually exclusive states like success, failure, and loading. enum class Color { RED, GREEN, BLUE } is a fixed set of constants and comes with its own value list. data class User(val name: String, val age: Int) fits pure data models — the compiler generates equals, hashCode, toString and copy(), so a single changed field is expressed as val b = a.copy(age = 30). Remember to mark a base class open if it should be subclassed.