Java: Rethink Domain Primitives with Valhalla

(dfa1.github.io)

30 points | by dfa11 a day ago ago

12 comments

  • mrkeen 16 hours ago ago

    Oh cool. It's newtype from Haskell 1.3 (1996).

    https://www.haskell.org/definition/from12to13.html#newtype

  • dfa11 a day ago ago

    Valhalla removes the historical performance tax on tiny domain objects. This article argues that value classes finally make domain‑safe primitives practical in real systems — strong invariants, flat layout, no wrapper overhead.

    I’d love feedback from folks who’ve experimented with Valhalla or have opinions on modelling domain types in Java.

    • exabrial 11 hours ago ago

      > Value classes cannot be null.

      Won't this be an issue for JPA/JSONB? `null` is a an important and value in JSON and nearly every database, and for modeling the real world.

      • dfa11 6 hours ago ago

        True: this compiles and the test passes today on JDK 27-jep401ea3:

          record Person(String name, Age age, Iban iban) {}
        
          // {"name": "SuperMario", "age": null, "iban": null}
          Person result = mapper.readValue(json, Person.class);
        
          assertThat(result.age()).isNull();   // passes — value class field, null
          assertThat(result.iban()).isNull();  // passes — value class field, null
        
          Jackson bypasses the constructor for null JSON tokens (getNullValue() short-circuits), so the
          validation in new Age(...) never runs. Both Age and Iban are value classes; both fields are null.
        
        What do you think? Of course, there are limitations as Person cannot be flattened anymore but I guess Valhalla will ship more JEPs later?
      • mrkeen 11 hours ago ago

        It is not.

        Null is the thing which it isn't. A null value is evidence that your modelling does not fit the domain.

        • exabrial 10 hours ago ago

          Unfortunately, the simple fact is that one can be handed a JSON document with non-existent keys, and one can be handed a database row with a null column. Often, the programmer may handed a schema from a third party they have no control over: opening a semantics debate is likely not a good course of action.

          I guess since basic types, like an int or double, cannot be null, I understand why these cannot be null. This unfortunately limits their usefulness, but it's a carryover from the underlying properties of the basic type.

          • joe_mwangi 8 hours ago ago

            Value types will be optionally null. What java will introduce to the tooling is narrowing of nullness types. Hence Foo! <: Foo? <: Foo. This will assist in enabling safe domains or scope in code that are null-restricted with ease. Hence we can model around such a type system rule.

          • mrkeen 10 hours ago ago

            I agree that it is unfortunate.

    • rf15 16 hours ago ago

      I've been working in Java for more than 20 years now, the last 10 with unperformant, yet performance critical enterprise slop.

      I can only really say one thing here: they are really late to the party with this but it's good. The performance boost will be welcome, and would probably change ORMs forever if applied right.

      Nullability rules should of course be also part of this domain-constricted types concept, since it would solve a lot of accidental mistakes. Because that's what a type system is for, right? Validation of programmer input.

      • dfa11 4 hours ago ago

        yeah 100% agree! In some critical area, I'm forced to drop all the "good things" and pack 2 shorts into 1 int to build a tuple(short,short)...