This is a re-implementation of the ULID library by Airframe in Kotlin. It follows the ULID spec.

• 128-bit compatibility with UUID
• 1.21e+24 unique ULIDs per millisecond
• Lexicographically sortable!
• Canonically encoded as a 26 character string, as opposed to the 36 character UUID
• Uses Crockford's base32 for better efficiency and readability (5 bits per character)
• Case-insensitive
• No special characters (URL safe)
• Monotonic sort order (correctly detects and handles the same millisecond)

// Constructors returning `ULID` objects
val newRandomUlid: ULID = ULID.newULID()
val newRandomUlidStr: String = ULID.newULIDString()

// Constructors that can fail and return Either<Error, ULID>
val ulidFromString = ULID("01BX5ZZKBKACTAV9WEVGEMMVRZ")
val ulidFromUUID = ULID.fromUUID(UUID.randomUUID())
val ulidFromUUIDString = ULID.fromUUIDString("f93f9fa5-f760-4341-b62b-508def86f087")
val ulidFromBytes = ULID.fromBytes(byteArray)
val ulidFromRawData = ULID.of(unixTimeMillis, randHi, randLow)

// Usage
val ulid = ULID.newULID()
println(ulid)                   // 01F51KNV65BQH43X6FMZ3JJT66
println(ulid.epochMillis)       // 1620330605765
println(ulid.toInstant())       // 2021-05-06T19:50:05.765Z
println(ulid.toUUID())          // 0179433a-ecc5-5de2-41f4-cfa7c72968c6
println(ulid.toBytes().asHex()) // 01 79 43 3a ec c5 5d e2 41 f4 cf a7 c7 29 68 c6
println(ULID.isValid("deadbeef"))                   // false
println(ULID.isValid("01F51KNV65BQH43X6FMZ3JJT66")) // true

ULID generation needs to be synchronized to allow correct handling of sub-millisecond generation. The initial airframe implementation uses synchronized(this) on a static singleton, and the same pattern was followed here. To accommodate the unlikely event of this blocking for a long time, a suspending API without the use of synchronized is available:

import com.kobil.ulid.suspending.newULID

suspend fun businessLogic() {
  val ulid = ULID.newULID() // 'suspendable' construction of ULID here
}

Creating ULIDs is synchronized using a Kotlin Mutex, which offers more fine-grained synchronization control at a higher cost, so using the suspend API will degrade throughput to about the same level as java.util.UUID.

Extensive performance review has not yet been performed, but a basic single-threaded throughput test comparing different existing ULID libraries is discussed here.

The tests were performed on an 2.8 GHz Core i7-7700HQ (MacBook Pro 14,3 2017), each generating 1 million IDs after a warm-up of 1000 IDs.

com.github.guepardoapps.kulid time: 1330 ms
          wvlet.airframe.ulid time: 1040 ms
        de.huxhorn.sulky.ulid time: 9856 ms
               java.util.UUID time: 676  ms
               com.kobil.ulid time: 545  ms

Tests:

./gradlew clean test

Packaging

./gradlew clean assemble

Publishing (don't forget the appropriate credentials in gradle.properties)

./gradlew publishAllPublicationsToNexusRepository