StackOverflowError in AvroSchemaCodec.encodeToApacheAvro for recursive sealed traits

[EtaCassiopeia]

Description

AvroSchemaCodec.encodeToApacheAvro causes a StackOverflowError when encoding a schema that contains a recursive sealed trait pattern - specifically when a case class within a sealed trait contains a List[SealedTrait] (or similar collection) referring back to the parent sealed trait.

This is similar to #286 which was fixed for ProtobufCodec in PR #389, but the same fix was not applied to AvroSchemaCodec.

Reproduction

Minimal example (Scala 3):

import zio.schema.{DeriveSchema, Schema}
import zio.schema.codec.AvroSchemaCodec

// Recursive sealed trait: Container holds List[Recursive] where Recursive includes Container
sealed trait Recursive
object Recursive:
  case class Container(items: List[Recursive]) extends Recursive
  case class Value(x: Int) extends Recursive
  given schema: Schema[Recursive] = DeriveSchema.gen[Recursive]

// This causes StackOverflowError
AvroSchemaCodec.encodeToApacheAvro(Recursive.schema)

Scala 2 equivalent:

sealed trait Recursive
object Recursive {
  case class Container(items: List[Recursive]) extends Recursive
  case class Value(x: Int) extends Recursive
  implicit val schema: Schema[Recursive] = DeriveSchema.gen[Recursive]
}

Stack Trace

java.lang.StackOverflowError
  at zio.schema.Schema$Enum2.cases(Schema.scala:1024)
  at zio.schema.codec.AvroSchemaCodec$.toAvroEnum(AvroSchemaCodec.scala:547)
  at zio.schema.codec.AvroSchemaCodec$.toAvroSchema(AvroSchemaCodec.scala:280)
  at zio.schema.codec.AvroSchemaCodec$.toAvroSchema(AvroSchemaCodec.scala:284)
  at zio.schema.codec.AvroSchemaCodec$.toAvroRecordField(AvroSchemaCodec.scala:673)
  at zio.schema.codec.AvroSchemaCodec$.extractAvroFields$$anonfun$1(AvroSchemaCodec.scala:594)
  at zio.Chunk.map(Chunk.scala:44)
  at zio.schema.codec.AvroSchemaCodec$.extractAvroFields(AvroSchemaCodec.scala:594)
  at zio.schema.codec.AvroSchemaCodec$.toAvroRecord$$anonfun$1$$anonfun$1(AvroSchemaCodec.scala:612)
  at zio.schema.codec.AvroSchemaCodec$.toAvroRecord(AvroSchemaCodec.scala:604)
  at zio.schema.codec.AvroSchemaCodec$.toAvroSchema(AvroSchemaCodec.scala:281)
  at zio.schema.codec.AvroSchemaCodec$.$anonfun$27(AvroSchemaCodec.scala:577)
  at zio.schema.codec.AvroSchemaCodec$.toAvroEnum(AvroSchemaCodec.scala:566)
  ... (cycle repeats)

The cycle: toAvroEnum → toAvroSchema → toAvroRecord → extractAvroFields → toAvroRecordField → toAvroSchema → toAvroEnum

Expected Behavior

The codec should detect the cycle and either:

1. Use Avro's named type reference mechanism (Avro supports recursive schemas via type name references)
2. Return a Left with an appropriate error message indicating recursive types are not supported

Root Cause

The call chain has no cycle detection. When processing a sealed trait that contains a case class with a field of type List[SealedTrait], it recursively processes the element type, which leads back to the same sealed trait indefinitely.

Suggested Fix

Add cycle detection similar to what was done in PR #389 for ProtobufCodec. Track "in-flight" type names during schema generation and either:

• Return a named reference when encountering an in-flight type (Avro supports this)
• Use a memoization/caching approach to avoid re-processing completed types

Example approach:

private def toAvroSchema(
  schema: Schema[_],
  inFlight: Set[String] = Set.empty
): Either[String, org.apache.avro.Schema] = {
  val typeName = getTypeName(schema)
  if (inFlight.contains(typeName)) {
    // Return named reference for recursive type
    Right(org.apache.avro.Schema.createRecord(typeName, null, namespace, false))
  } else {
    // Process with updated inFlight set
    processSchema(schema, inFlight + typeName)
  }
}

Environment

• zio-schema version: 1.7.6 (latest as of Jan 2026)
• Scala version: 3.3.4
• JDK: 21

Confirmed Test Results

Tested with zio-schema 1.7.6 - all three recursive patterns fail:

Pattern	Result
List[SealedTrait]	StackOverflowError
Option[SealedTrait]	StackOverflowError
Direct field reference	StackOverflowError

Full reproduction project:

// build.sbt
ThisBuild / scalaVersion := "3.3.4"
libraryDependencies ++= Seq(
  "dev.zio" %% "zio-schema" % "1.7.6",
  "dev.zio" %% "zio-schema-avro" % "1.7.6",
  "dev.zio" %% "zio-schema-derivation" % "1.7.6"
)

// src/main/scala/Repro.scala
import zio.schema.{DeriveSchema, Schema}
import zio.schema.codec.AvroSchemaCodec

sealed trait Recursive
object Recursive:
  case class Container(items: List[Recursive]) extends Recursive
  case class Value(x: Int) extends Recursive
  given schema: Schema[Recursive] = DeriveSchema.gen[Recursive]

@main def run(): Unit =
  println(AvroSchemaCodec.encodeToApacheAvro(Recursive.schema))
  // Throws StackOverflowError

Related Issues

• Recursive data-type can't be encoded with ProtobufCodec #286 - Recursive data-type can't be encoded with ProtobufCodec (FIXED in PR Make DynamicValue.fromSchemaAndValue and codecs stack safe #389)
• java.lang.StackOverflowError on DeriveSchema.gen #159 - java.lang.StackOverflowError on DeriveSchema.gen (different issue - macro-level)