[stdlib] Add allDistinct APIs for iterables, sequences and arrays by DmitryNekrasov · Pull Request #6187 · JetBrains/kotlin

Dmitry Nekrasov (DmitryNekrasov) · 2026-06-09T12:48:02Z

Adds experimental allDistinct() and allDistinctBy { } ("are all elements different from each other?") for Iterable, Sequence, and the object, primitive, and unsigned array families — the dual of allEqual. @ExperimentalStdlibApi, @SinceKotlin("2.4").

Distinctness goes through a hash set, so elements compare by equals/hashCode: NaN equals NaN, and -0.0 is not equal to 0.0 — the same as distinct()/toSet() and the sibling allEqual.

^KT-30270 Fixed

kotlin-safemerge · 2026-06-09T12:48:08Z

Code Owners

Rule	Owners	Approval
`/libraries/stdlib/`, `/libraries/tools/kotlin-stdlib-gen/`	`@DmitryNekrasov`, `@fzhinkin`, `@ilya-g`, `@qwwdfsad`	❌ `@ilya-g`, `@fzhinkin`
`/libraries/tools/binary-compatibility-validator/`	kotlin-libraries `@qwwdfsad` `@ilya-g` `@shanshin` `@DmitryNekrasov` `@sandwwraith` `@fzhinkin`	❌ `@ilya-g`, `@fzhinkin`

Dmitry Nekrasov (DmitryNekrasov) · 2026-06-09T13:09:11Z

/dry-run

Build Server (KotlinBuild) · 2026-06-09T13:10:54Z

THIS IS A DRY RUN

Quality gate is triggered at https://buildserver.labs.intellij.net/build/969309422 — use this link to get full insight.

Quality gate was triggered with the following revisions:

kotlin
Branch: refs/merge/GITHUB-6187/safe-merge
Commit: 1ad454d

Quality gate failed. See https://buildserver.labs.intellij.net/build/969309422 to get full insight.

Dmitry Nekrasov (DmitryNekrasov) · 2026-06-09T17:39:33Z

The failing cases are allDistinctDifferentNaNBits{Double,Float} (and the By variants): they expect two NaNs with different bit patterns to count as duplicates. allDistinct dedups with a HashSet, and on JS the number hashCode disagrees with equals for NaN — equals treats all NaNs as equal, but getNumberHashCode hashes the raw IEEE bits without canonicalizing NaN, so two NaNs with different bits get different hashes, never share a bucket, and the set keeps both. JVM passes because Double.hashCode/Float.hashCode canonicalize NaN via doubleToLongBits. (allEqual is unaffected — it compares with equals, no hashing.)

It's engine-dependent: I checked Float — fails on Node and Chrome (both V8), passes on Safari and Firefox. I think Double takes the same path, I didn't check it across engines. Both js/node and js/browser fail. Canonical NaN and -0.0/0.0 pass everywhere.

Reproducer (https://pl.kotl.in/P6_xrfPbh):

val a = Float.NaN
val b = Float.fromBits(0xFFFC0000.toInt())  // a different NaN bit pattern
println(a.equals(b))                        // true everywhere
println(a.hashCode() == b.hashCode())       // JVM: true; JS: false on Node/Chrome, true on Safari/Firefox

Two options:

Skip these cases on JS with testExceptOn(TestPlatform.Js); they keep running on the JVM, where NaN canonicalization makes the result well-defined. Optionally note the platform-dependence in the KDoc.
File a Kotlin/JS issue to canonicalize NaN in getNumberHashCode — engine-independent and stdlib-side, but it touches every hash-based collection with Double/Float keys, so it's broader than this PR.

I'd take option 1 now; option 2 is a follow-up that would let us drop the guard.

Dmitry Nekrasov (DmitryNekrasov) · 2026-06-09T17:43:11Z

/dry-run

Build Server (KotlinBuild) · 2026-06-09T17:45:06Z

THIS IS A DRY RUN

Quality gate is triggered at https://buildserver.labs.intellij.net/build/969660301 — use this link to get full insight.

Quality gate was triggered with the following revisions:

kotlin
Branch: refs/merge/GITHUB-6187/safe-merge
Commit: 005fd06

Quality gate finished successfully.

Filipp Zhinkin (fzhinkin) · 2026-06-09T18:42:05Z

+@ExperimentalStdlibApi
+public fun BooleanArray.allDistinct(): Boolean {
+    if (size < 2) return true
+    val seen = HashSet<Boolean>()


For booleans, there are only three cases:

the size is < 2

the size is 2 and values are different

in all other cases, there are duplicates

Thank you! Totally agree.

Filipp Zhinkin (fzhinkin) · 2026-06-09T18:54:27Z

+@SinceKotlin("2.4")
+@ExperimentalStdlibApi
+public fun ByteArray.allDistinct(): Boolean {
+    if (size < 2) return true


And if the size is greater than 512, then there are certainly some duplicates :)

Done. The threshold is 256 — the number of distinct Byte values; ShortArray, UShortArray, and CharArray got the same check at 65536.

Filipp Zhinkin (fzhinkin) · 2026-06-09T21:30:00Z

+@ExperimentalStdlibApi
+public fun ByteArray.allDistinct(): Boolean {
+    if (size < 2) return true
+    val seen = HashSet<Byte>()


There's not much we can do for other integer types, but all byte values could be captured using 4 longs, so it might be worth implementing teeny-tiny bitset with a single add operation.

Filipp Zhinkin (fzhinkin) · 2026-06-09T21:38:02Z

+    seen.add(selector(first))
+    do {
+        if (!seen.add(selector(iterator.next()))) return false


selector will be inlined twice (kudos to @qwwdfsad for the hint), needlessly emitting additional bytecode at the allDistinctBy's call-sites.

So it makes sense to rewrite it into something like:

val iterator = iterator() if (!iterator.hasNext()) return true var element: T? = iterator.next() if (!iterator.hasNext() return true val seen = HashSet<K>() while (true) { if (!seed.add(selector(element))) return false if (!iterator.hasNext()) break element = iterator.next() } return true

It would be nice to reduce the size of the loop's preamble, but not sure if there's a reasonable way to achieve it.

Review follow-up for #6187: - BooleanArray: only three outcomes exist (size < 2; a distinct pair; otherwise a guaranteed duplicate), so compute the answer directly with no HashSet. - ByteArray/UByteArray: more than 256 elements can't be all-distinct (pigeonhole), return false right away; values that fit are tracked in a 256-bit set of four Longs instead of a HashSet, avoiding boxing and a hash-table allocation. - ShortArray/UShortArray/CharArray: same pigeonhole shortcut at 65536. No bitset here: zeroing 8 KiB per call would penalize the common small-array case. Wider element types (Int and above) can't benefit: their value domain exceeds the maximum array size, so the size check would never fire. Tests cover the domain boundaries (full domain distinct, same size with a duplicate, domain size + 1), all BooleanArray shapes up to size 2 exhaustively, and byte values that collide in the low six bits of the bitset words. KT-30270

Adds experimental `allDistinct()` and `allDistinctBy { }` ("are all elements different from each other?") for `Iterable`, `Sequence`, and the object, primitive, and unsigned array families — the dual of `allEqual`. `@ExperimentalStdlibApi`, `@SinceKotlin("2.4")`. Distinctness uses `equals`/`hashCode`, so for floating-point elements `NaN` equals `NaN` and `-0.0` is not equal to `0.0`, consistent with `Double.equals` and the existing `allEqual`/`isSorted`. ^KT-30270 Fixed

On JS, Double and Float hashCode don't canonicalize NaN, so a HashSet keeps NaN values with different bit patterns apart.

Only the byte arrays get a bitset: zeroing the 8 KiB needed for the 16-bit types on every call would penalize the common small-array case.

The selector is inlined, so each call site used to receive two copies of the lambda's bytecode.

Dmitry Nekrasov (DmitryNekrasov) · 2026-06-10T07:19:42Z

/dry-run

Build Server (KotlinBuild) · 2026-06-10T07:21:40Z

THIS IS A DRY RUN

Quality gate is triggered at https://buildserver.labs.intellij.net/build/970073740 — use this link to get full insight.

Quality gate was triggered with the following revisions:

kotlin
Branch: refs/merge/GITHUB-6187/safe-merge
Commit: 90f8fc3

Quality gate failed. See https://buildserver.labs.intellij.net/build/970073740 to get full insight.

Dmitry Nekrasov (DmitryNekrasov) self-assigned this Jun 9, 2026

Dmitry Nekrasov (DmitryNekrasov) added the Standard Library label Jun 9, 2026

Dmitry Nekrasov (DmitryNekrasov) force-pushed the rr/stdlib/dmitry.nekrasov/KT-30270-allDistinct branch 3 times, most recently from 31228ac to ebfd3a6 Compare June 9, 2026 13:03

Dmitry Nekrasov (DmitryNekrasov) marked this pull request as draft June 9, 2026 13:04

Dmitry Nekrasov (DmitryNekrasov) force-pushed the rr/stdlib/dmitry.nekrasov/KT-30270-allDistinct branch from ebfd3a6 to ad57343 Compare June 9, 2026 17:41

Filipp Zhinkin (fzhinkin) reviewed Jun 9, 2026

View reviewed changes

Dmitry Nekrasov (DmitryNekrasov) requested a review from ilya-g June 9, 2026 19:35

Dmitry Nekrasov (DmitryNekrasov) marked this pull request as ready for review June 9, 2026 19:35

Filipp Zhinkin (fzhinkin) reviewed Jun 9, 2026

View reviewed changes

Dmitry Nekrasov (DmitryNekrasov) force-pushed the rr/stdlib/dmitry.nekrasov/KT-30270-allDistinct branch 3 times, most recently from e59ec6c to 529aeb5 Compare June 10, 2026 07:13

Dmitry Nekrasov (DmitryNekrasov) added 4 commits June 10, 2026 11:17

[stdlib] Skip allDistinct NaN-bit checks on JS

b67596f

On JS, Double and Float hashCode don't canonicalize NaN, so a HashSet keeps NaN values with different bit patterns apart.

[stdlib] Specialize allDistinct for small-domain element types

d5160ac

Only the byte arrays get a bitset: zeroing the 8 KiB needed for the 16-bit types on every call would penalize the common small-array case.

[stdlib] Call allDistinctBy selector from a single place

5ab237c

The selector is inlined, so each call site used to receive two copies of the lambda's bytecode.

Dmitry Nekrasov (DmitryNekrasov) force-pushed the rr/stdlib/dmitry.nekrasov/KT-30270-allDistinct branch from 529aeb5 to 5ab237c Compare June 10, 2026 07:18

Dmitry Nekrasov (DmitryNekrasov) requested a review from Filipp Zhinkin (fzhinkin) June 10, 2026 07:26

Conversation

Dmitry Nekrasov (DmitryNekrasov) commented Jun 9, 2026

Uh oh!

kotlin-safemerge Bot commented Jun 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Code Owners

Uh oh!

Dmitry Nekrasov (DmitryNekrasov) commented Jun 9, 2026

Uh oh!

Build Server (KotlinBuild) commented Jun 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Dmitry Nekrasov (DmitryNekrasov) commented Jun 9, 2026

Uh oh!

Dmitry Nekrasov (DmitryNekrasov) commented Jun 9, 2026

Uh oh!

Build Server (KotlinBuild) commented Jun 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Filipp Zhinkin (fzhinkin) Jun 9, 2026

Choose a reason for hiding this comment

Uh oh!

Dmitry Nekrasov (DmitryNekrasov) Jun 9, 2026

Choose a reason for hiding this comment

Uh oh!

Dmitry Nekrasov (DmitryNekrasov) Jun 10, 2026

Choose a reason for hiding this comment

Uh oh!

Filipp Zhinkin (fzhinkin) Jun 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Dmitry Nekrasov (DmitryNekrasov) Jun 10, 2026

Choose a reason for hiding this comment

Uh oh!

Filipp Zhinkin (fzhinkin) Jun 9, 2026

Choose a reason for hiding this comment

Uh oh!

Dmitry Nekrasov (DmitryNekrasov) Jun 10, 2026

Choose a reason for hiding this comment

Uh oh!

Filipp Zhinkin (fzhinkin) Jun 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Dmitry Nekrasov (DmitryNekrasov) Jun 10, 2026

Choose a reason for hiding this comment

Uh oh!

Dmitry Nekrasov (DmitryNekrasov) commented Jun 10, 2026

Uh oh!

Build Server (KotlinBuild) commented Jun 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

kotlin-safemerge Bot commented Jun 9, 2026 •

edited

Loading

Build Server (KotlinBuild) commented Jun 9, 2026 •

edited

Loading

Build Server (KotlinBuild) commented Jun 9, 2026 •

edited

Loading

Filipp Zhinkin (fzhinkin) Jun 9, 2026 •

edited

Loading

Filipp Zhinkin (fzhinkin) Jun 9, 2026 •

edited

Loading

Build Server (KotlinBuild) commented Jun 10, 2026 •

edited

Loading