Skainet Java Interop

skainet-java-interop

Rules for making SKaiNET features usable from Java without surfacing Kotlin-specific machinery. Java is the first guest language after Kotlin, so its ergonomics are a constraint — not an afterthought.

When to use

• Adding a new Java-callable surface (factory, ops utility, builder).
• Editing files under skainet-*/src/jvmMain/kotlin/sk/ainet/java/.
• Reviewing whether a Kotlin API can be called idiomatically from Java.
• Porting a Kotlin test into the skainet-test-java module to prove the Java surface.

When NOT to use

• Writing the underlying Kotlin implementation — kotlin.
• Editing tests against the Java surface — skainet-testing (the Java JUnit pattern is documented there).
• Picking the source set or KMP target — kmp.
• Adding the Gradle dependency on skainet-test-java — gradle-multimodule.

Hard rules

1. @file:JvmName("...") on every Java-callable file. The file's first non-blank line MUST be @file:JvmName("...") followed by package sk.ainet.java. The JvmName matches what Java users will type as the class identifier (SKaiNET, TensorJavaOps, <Feature>JavaOps).
2. Container shape is object, never top-level functions. A Java-callable surface MUST be public object Name { ... }. Never expose top-level public fun foo(...) to Java callers — Java sees them as static methods on a synthetic <FileName>Kt class.
3. Every member carries @JvmStatic. Without it, Java users have to type SKaiNET.INSTANCE.context() instead of SKaiNET.context(). This rule applies to every public function and every property accessor on the Java-facing object.
4. Default arguments require @JvmOverloads. Java doesn't see Kotlin defaults; @JvmOverloads synthesises overloads with progressively fewer parameters from right to left. Apply it to every @JvmStatic function that has a default value.
5. Forbidden in Java-facing signatures:
   • value class types (Java sees the boxed form, defeats the purpose)
   • inline fun (the bytecode shape is unfriendly)
   • Result<T> (no Kotlin runtime in pure Java contexts; consumers can't unwrap)
   • Receiver-typed lambdas (X.() -> Y) — Java users cannot supply them
   • Sequence<T> (no Java-native iteration); use Iterable<T> or a primitive array instead
   • Variance tokens that complicate Java generics — flatten to Tensor<*, *> for any tensor crossing the boundary
6. Tensors crossing the boundary use Tensor<*, *>. Java callers cannot easily express Tensor<DType, Any?> parameters; expose Tensor<*, *> and cast internally with @Suppress("UNCHECKED_CAST"). See TensorJavaOps.add for the canonical pattern.
7. Primitive arrays beat List<Float>. Use FloatArray, IntArray, DoubleArray, ByteArray — Java sees them as primitive arrays (float[], int[], …), zero boxing.
8. DType is passed by instance, not KClass. Java callers don't have ergonomic access to KClass<T>; expose DType instances (DType.fp32(), DType.int32()) and resolve to KClass internally.
9. @JvmOverloads for any defaults; no vararg of objects in Java-facing positions unless backed by a primitive array.
10. Mirror every Java-facing method with a Java JUnit 5 test in skainet-test-java. The presence of a test in that module is the canonical proof that Java consumers can call the API. New @JvmStatic member ⇒ new @Test in skainet-test-java.

Workflow

1. Find the closest sibling under skainet-*/src/jvmMain/kotlin/sk/ainet/java/. Mirror its @file:JvmName, package declaration, KDoc-with-Java-example header, and member shape.
2. Define the new object (or extend an existing one).
3. For each member: @JvmStatic is mandatory; @JvmOverloads if any parameter has a default; primitive arrays for bulk data; Tensor<*, *> for tensors crossing the boundary.
4. Implement the member by casting the wildcard tensor to the internal generic form (Tensor<DType, Any?>) and delegating to the Kotlin API.
5. Write a JUnit 5 test under skainet-test-java/src/test/java/sk/ainet/java/<Feature>JavaOpsTest.java proving the surface compiles and runs from Java.
6. Self-verify with the checklist below before reporting done.

Self-verification checklist

Before declaring the change complete:

• File starts with @file:JvmName("...") (line 1).
• File is in package sk.ainet.java.
• Container is a public object.
• Every public member has @JvmStatic.
• Every member with a default has @JvmOverloads.
• No value class, no Result<T>, no Sequence<T>, no receiver-typed lambdas in any signature.
• Tensors cross the boundary as Tensor<*, *>.
• DType crosses as DType instance, not KClass<DType>.
• A JUnit 5 test exists under skainet-test-java/.../<Feature>JavaOpsTest.java exercising at least one method of the new surface.

Canonical examples

Entry-point factory:

@file:JvmName("SKaiNET")

package sk.ainet.java

import sk.ainet.context.DirectCpuExecutionContext
import sk.ainet.context.ExecutionContext
import sk.ainet.lang.tensor.Shape
import sk.ainet.lang.tensor.Tensor
import sk.ainet.lang.types.DType
import sk.ainet.lang.types.FP32
import kotlin.reflect.KClass

public object SKaiNET {

    @JvmStatic
    public fun context(): ExecutionContext =
        DirectCpuExecutionContext.create()

    @JvmStatic
    public fun tensor(ctx: ExecutionContext, shape: IntArray, dtype: DType, data: FloatArray): Tensor<*, *> {
        @Suppress("UNCHECKED_CAST")
        val kclass = dtype::class as KClass<DType>
        return ctx.fromFloatArray<DType, Any?>(Shape(*shape), kclass, data)
    }

    @JvmStatic
    @JvmOverloads
    public fun zeros(ctx: ExecutionContext, shape: IntArray, dtype: DType = FP32): Tensor<*, *> {
        @Suppress("UNCHECKED_CAST")
        val kclass = dtype::class as KClass<DType>
        return ctx.zeros<DType, Any?>(Shape(*shape), kclass)
    }
}
// from: SKaiNET/skainet-backends/skainet-backend-cpu/src/jvmMain/kotlin/sk/ainet/java/SKaiNET.kt:1-102

Ops facade — every member @JvmStatic, defaults via @JvmOverloads:

@file:JvmName("TensorJavaOps")

package sk.ainet.java

public object TensorJavaOps {

    @JvmStatic
    public fun add(a: Tensor<*, *>, b: Tensor<*, *>): Tensor<*, *> {
        @Suppress("UNCHECKED_CAST")
        val ta = a as Tensor<DType, Any?>
        @Suppress("UNCHECKED_CAST")
        val tb = b as Tensor<DType, Any?>
        return ta.ops.add(ta, tb)
    }

    @JvmStatic
    public fun matmul(a: Tensor<*, *>, b: Tensor<*, *>): Tensor<*, *> { /* ... */ }

    @JvmStatic
    @JvmOverloads
    public fun softmax(a: Tensor<*, *>, dim: Int = -1): Tensor<*, *> { /* ... */ }
}
// from: SKaiNET/skainet-lang/skainet-lang-core/src/jvmMain/kotlin/sk/ainet/java/TensorJavaOps.kt:1-322

Mirror Java JUnit test that proves the surface from a Java caller:

package sk.ainet.java;

import org.junit.jupiter.api.BeforeAll;
import org.junit.jupiter.api.Test;
import sk.ainet.context.ExecutionContext;
import sk.ainet.lang.tensor.Tensor;
import sk.ainet.lang.types.DType;

import static org.junit.jupiter.api.Assertions.*;

class TensorJavaOpsTest {
    private static ExecutionContext ctx;

    @BeforeAll
    static void setUp() { ctx = SKaiNET.context(); }

    @Test
    void add() {
        Tensor<?, ?> a = SKaiNET.tensor(ctx, new int[]{2, 2}, DType.fp32(),
                new float[]{1f, 2f, 3f, 4f});
        Tensor<?, ?> b = SKaiNET.tensor(ctx, new int[]{2, 2}, DType.fp32(),
                new float[]{10f, 20f, 30f, 40f});

        Tensor<?, ?> c = TensorJavaOps.add(a, b);
        assertArrayEquals(new int[]{2, 2}, c.getShape().getDimensions());

        float[] result = c.getData().copyToFloatArray();
        assertArrayEquals(new float[]{11f, 22f, 33f, 44f}, result, 1e-6f);
    }
}
// from: SKaiNET/skainet-test/skainet-test-java/src/test/java/sk/ainet/java/TensorJavaOpsTest.java:1-38

Related skills

• The underlying Kotlin API the facade wraps — see ../kotlin/SKILL.md.
• Source-set placement (always jvmMain for facades) — see ../kmp/SKILL.md.
• Wiring skainet-test-java and adding the Java test — see ../gradle-multimodule/SKILL.md and ../skainet-testing/SKILL.md.
• Tensor construction the facade exposes — see the skainet-data-dsl skill (in the sibling consumer plugin).
• The Java caller's perspective on this same surface — see the skainet-java-consumer skill (in the sibling consumer plugin).

Anti-patterns

// WRONG — top-level fun; Java sees it as TensorOpsKt.add
@file:JvmName("TensorOps")
package sk.ainet.java

public fun add(a: Tensor<*, *>, b: Tensor<*, *>): Tensor<*, *> = ...

// RIGHT — public object + @JvmStatic
@file:JvmName("TensorJavaOps")
package sk.ainet.java

public object TensorJavaOps {
    @JvmStatic
    public fun add(a: Tensor<*, *>, b: Tensor<*, *>): Tensor<*, *> = ...
}

// WRONG — defaults without @JvmOverloads; Java cannot omit `dim`
@JvmStatic
public fun softmax(a: Tensor<*, *>, dim: Int = -1): Tensor<*, *> = ...

// RIGHT — @JvmOverloads
@JvmStatic
@JvmOverloads
public fun softmax(a: Tensor<*, *>, dim: Int = -1): Tensor<*, *> = ...

// WRONG — exposing KClass to Java
@JvmStatic
public fun tensor(ctx: ExecutionContext, shape: IntArray, dtype: KClass<DType>, data: FloatArray): Tensor<*, *> = ...

// RIGHT — DType instance
@JvmStatic
public fun tensor(ctx: ExecutionContext, shape: IntArray, dtype: DType, data: FloatArray): Tensor<*, *> = ...

// WRONG — Sequence + receiver lambda
@JvmStatic
public fun forEachLayer(model: Module<*, *>, action: Module<*, *>.(Layer) -> Unit)

// RIGHT — plain Iterable + java.util.function.BiConsumer (or specialised interface)
@JvmStatic
public fun forEachLayer(model: Module<*, *>, action: java.util.function.BiConsumer<Module<*, *>, Layer>)

References

• references/jvm-annotations.md — every @Jvm* annotation we use, when to apply each.
• references/forbidden-signatures.md — exhaustive list of Kotlin shapes that MUST NOT cross the Java boundary, with rewrites.