Add maybe types by odersky · Pull Request #26814 · scala/scala3

odersky · 2026-08-15T15:30:05Z

Trial balloon for a new kind of maybe type T?.

bishabosha · 2026-08-24T13:04:02Z

        val year  = y.parseInt?
-        provided(1 <= day && day <= 31)
-        provided(1 <= month && month <= 12)
+        if 1 <= day && day <= 31


bishabosha · 2026-08-24T13:12:32Z

+        val month = m.parseInt.withErr(s"malformed month: $m")?
+        val year  = y.parseInt.withErr(s"malformed year: $y")?
+        if 1 <= day && day <= 31 else s"day $day outside allowed range 1..31"
+        if 1 <= month && month <= 12 else s"month $month outside allowed range 1..12"


bishabosha · 2026-08-24T13:13:53Z

+    case Invoice(customerId: String, amount: BigInt)
+    case Refund(invoiceId: String, reason: String)
+
+  def validateJson(json: JsonDict): Result[InvoiceOrRefund, List[String]] =


@odersky this is the example of the cumulative error scope

Ichoran · 2026-08-26T17:19:05Z

+  inline def apply[T](x: T): Maybe[T, Nothing] = {
+    if x == null then new Valid(null)
+    else if x.isInstanceOf[Valid] then new Valid(x)
+    else x


This encoding declares (A ? E) ? F to be isomorphic to A ? (E | F) because Ok(Err(e)) falls through to the else x case. If this is what we intend, the type system should know about it. Otherwise, we need to either statically know that the actual T is disjoint from A ? E for all A, E; or we need a test for Valid or Fail, so that we encode Ok(Err(e)) as Valid(Fail(e)).

Indeed. It's fixed in the latest commit.

- Enabled by experimental.magic - Independent of language.postfixOps - Highest precedence

ExtractDependencies fell into a "NoDenotation cannot be cast to ClassDenotation" assertioin violation. When recompiling after some changes. This fix avoids that.

Otherwise the maybe logic does not work correctly. It would be nice if we could link explicit-nulls with magic, but that does not work since explicit-nulls is a global flag that has to be set when the compiler starts up. We can't even change it from run to run.

Special cases needed for subtyping, erasure, and pattern matching.

The erasure of Maybe changes after bootstrapped. So anything that touches it can compile only with the bootstrapped compiler.

Also: Drop $ in front of Maybe, no harm in using it directly, since it is a sealed trait. Also: Fix exhaustivity checking for Ok patterns

- Drop OK unapply call, which is known to be the identity - Simplify leading null test for maybe types

Also, move back bootstrapped library and tests to regular. Since we now inline Ok.unapply, we are no longer affected by difference in erasure between bootstrapped and non-bootstrapped.

Plus a lot more tests

Needed to avoid erasure difference between bootstrapped and non-bootstrapped compilers.

Also: drop overfitted condition by Claude in space engine.

To reduce added linecount we drop a bunch of tests that are straightforward Option -> ? substitutions.

That ensures Ok(x) == Ok(x) Err(y) == Err(y)

The correct location is in run-bootstrapped

Comes with the following bug fixes: 1. Maybe arguments should not be boxed. 2. Capture sets of module vals need to be interpolated 3. When reporting overriding errors under -explain, use isSubTypeWhenFrozen 4. Interpolate self types of modules downwards 5. Make Tuple a Pure trait (fails i13968-maybe.scala otherwise)

Flag an error only if two different source versions are implied. Reason: We might have set the source to future and a language import might also imply source future. That should be OK.

This used to also include nulls but that makes no sense: null.isInstanceOf does not diverge, it just returns false. The ok-tostring test shows that always false should not yield a warning, since this might come from inlined code.

I have no idea what goes on here. MimaFilter error messages are not specific enough to figure out whether they mean forwards or backwards compatibility.

odersky · 2026-09-02T12:59:39Z

Superseded by #26956.

odersky force-pushed the add-maybe-types branch 5 times, most recently from 339631c to d740b53 Compare August 21, 2026 14:44

odersky force-pushed the add-maybe-types branch 2 times, most recently from 5744e79 to cd3d5d3 Compare August 23, 2026 20:33

bishabosha reviewed Aug 24, 2026

View reviewed changes

odersky force-pushed the add-maybe-types branch 2 times, most recently from 7d57d29 to 2bdf8fe Compare August 24, 2026 14:30

Ichoran reviewed Aug 26, 2026

View reviewed changes

odersky force-pushed the add-maybe-types branch 2 times, most recently from 2a34add to ed7cedc Compare August 27, 2026 16:48

odersky added 15 commits August 27, 2026 22:30

Allow ? postfix operator

513610f

- Enabled by experimental.magic - Independent of language.postfixOps - Highest precedence

Desugar to Magic types

b1e59f7

Fixing a crash when recompiling

c6f4b04

ExtractDependencies fell into a "NoDenotation cannot be cast to ClassDenotation" assertioin violation. When recompiling after some changes. This fix avoids that.

Handling maybe types

6750311

Special cases needed for subtyping, erasure, and pattern matching.

Move some of library and tests into bootstrapped directory

0c96f97

The erasure of Maybe changes after bootstrapped. So anything that touches it can compile only with the bootstrapped compiler.

Print maybe types with ?

cd7db97

Treat magic as non-viral

36106c9

Prepare for binary maybes

6196452

Also: Drop $ in front of Maybe, no harm in using it directly, since it is a sealed trait. Also: Fix exhaustivity checking for Ok patterns

Binary maybe

0e06020

Optimizations for maybe matching

f0305e0

- Drop OK unapply call, which is known to be the identity - Simplify leading null test for maybe types

Fixes for handling orelse types

25bbd87

Adapt Space logic

705c0af

Also, move back bootstrapped library and tests to regular. Since we now inline Ok.unapply, we are no longer affected by difference in erasure between bootstrapped and non-bootstrapped.

Fix maybe widening in TypeComparer

c1017e5

Plus a lot more tests

Make Err constructor inline

0d59a3a

Needed to avoid erasure difference between bootstrapped and non-bootstrapped compilers.

odersky added 13 commits August 27, 2026 22:30

Allow widening also for binary orelse types

4284f7c

Direct style core infrastructure

15c1aa6

Also: drop overfitted condition by Claude in space engine.

Drop mostly redundant tests

2158165

To reduce added linecount we drop a bunch of tests that are straightforward Option -> ? substitutions.

Make newSyntax setting depend on language imports

08a8945

Set source version to future under magic

be83aa4

Allow if without then in maybe blocks

6bd84c8

Disallow maybes with wildcard arguments

f8826fc

Optimizations of code patterns around maybes

51bcc19

Make Valid and Fail case classes

4e01f4d

That ensures Ok(x) == Ok(x) Err(y) == Err(y)

Fix reduce for Unit equality

0e20fb8

Optimize Ok(...) pattern match code

00191a1

Drop spurious pos/maybe-translation.scala file

341ae82

The correct location is in run-bootstrapped

odersky force-pushed the add-maybe-types branch from 4023769 to 9e8ce45 Compare August 27, 2026 20:35

Update MiMaFilters and error code snippets

028aacf

odersky force-pushed the add-maybe-types branch from 9e8ce45 to 028aacf Compare August 27, 2026 21:09

odersky added 4 commits August 28, 2026 19:50

Relax duplicate source version errors

19fe064

Flag an error only if two different source versions are implied. Reason: We might have set the source to future and a language import might also imply source future. That should be OK.

Fix parenthesization when printing of maybe and result types

0f8749b

OK.apply method to account for OK(Err(_))

7951895

odersky force-pushed the add-maybe-types branch from 6238514 to 116e4b6 Compare August 28, 2026 22:25

Try to fix MimaFilters

001c53b

I have no idea what goes on here. MimaFilter error messages are not specific enough to figure out whether they mean forwards or backwards compatibility.

odersky force-pushed the add-maybe-types branch from 116e4b6 to 001c53b Compare August 29, 2026 08:16

Add "maybe" language import

27a6595

odersky force-pushed the add-maybe-types branch 2 times, most recently from 0ddea2f to be3fd7c Compare August 31, 2026 06:49

Make everything available under experimental.errorHandling

db5d26d

odersky force-pushed the add-maybe-types branch from be3fd7c to db5d26d Compare August 31, 2026 06:59

odersky closed this Sep 2, 2026

Navigation Menu

Uh oh!

Add maybe types - #26814

Add maybe types#26814
odersky wants to merge 40 commits into
scala:mainfrom
dotty-staging:add-maybe-types

odersky commented Aug 15, 2026

Uh oh!

bishabosha Aug 24, 2026

Uh oh!

bishabosha Aug 24, 2026

Uh oh!

bishabosha Aug 24, 2026

Uh oh!

Ichoran Aug 26, 2026 •

edited

Loading

Uh oh!

odersky Aug 28, 2026

Uh oh!

odersky commented Sep 2, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

odersky commented Aug 15, 2026

Uh oh!

bishabosha Aug 24, 2026

Choose a reason for hiding this comment

Uh oh!

bishabosha Aug 24, 2026

Choose a reason for hiding this comment

Uh oh!

bishabosha Aug 24, 2026

Choose a reason for hiding this comment

Uh oh!

Ichoran Aug 26, 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!

odersky Aug 28, 2026

Choose a reason for hiding this comment

Uh oh!

odersky commented Sep 2, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Ichoran Aug 26, 2026 •

edited

Loading