Run2 scala variance

Scala Variance Annotations (vs Python TypeVars)

Python TypeVar → Scala Variance

T_co    = TypeVar("T_co",    covariant=True)     # +T in Scala
T_contra = TypeVar("T_contra", contravariant=True) # -T in Scala
T       = TypeVar("T")                            # T  in Scala (invariant)

Variance Rules

Annotation	Meaning	T can appear in	Subtyping
+T	Covariant	Output (return types)	F[Dog] <: F[Animal]
-T	Contravariant	Input (parameters)	F[Animal] <: F[Dog]
T	Invariant (default)	Both	No subtype relation

Common Variance Problem: Covariant T in Function Parameter Position

Python doesn't enforce variance rules, but Scala does at compile time:

# Python allows this — T_co (covariant) used in contravariant (input) position
class TokenContainer(Generic[T_co]):
    def map_tokens(self, func: Callable[[T_co], str]) -> list[str]: ...

Scala compiler REJECTS:

class TokenContainer[+T] {
  def mapTokens(f: T => String): List[String]  // ERROR: covariant T in contravariant position
}

Scala fix — use a lower bound:

class TokenContainer[+T](items: Seq[T]) {
  def mapTokens[B >: T](f: B => String): List[String] = items.map(f).toList
  // B >: T: B is a supertype of T. f accepts B, and T values can be passed as B.
}

Covariant Container (read-only)

final class TokenContainer[+T](items: Seq[T]) {
  private val _items: Vector[T] = items.toVector
  def getAll: Vector[T]         = _items
  def size: Int                 = _items.size
  def mapTokens[B >: T](f: B => String): List[String] = _items.map(f).toList
}

• Vector[T] is itself covariant, so returning it from +T class is fine.
• getAll returns Vector[T] — covariant position ✓

Contravariant Sink (write-only)

final class TokenSink[-T] {
  private var _received: List[Any] = Nil   // List[Any] avoids variance issue in var
  def receive(item: T): Unit = _received = _received :+ (item: Any)
  def drain(): List[Any] = {
    val result = _received
    _received = Nil
    result
  }
}

• List[Any] instead of List[T] — a var List[T] with -T would be invalid
• receive(item: T) — contravariant position ✓

Invariant Handler (read + write)

final class BivariantHandler[T](private var value: T) {
  def get: T               = value
  def set(v: T): Unit      = value = v
  def transform(f: T => T): T = { value = f(value); value }
}

• Both get (output) and set (input) use T → must be invariant

Abstract Generic Tokenizer

abstract class BaseTokenizer[T] {
  def tokenize(value: T): Token                              // abstract
  def tokenizeBatch(values: Iterable[T]): Iterator[Token] = // concrete default
    values.iterator.map(tokenize)
}

• Iterable[T] is covariant, so tokenizeBatch works correctly.
• Concrete subclasses implement tokenize.

Concrete Subclass Pattern

final class StringTokenizer(
  encoding: String = "UTF-8",
  normalizer: String => String = identity
) extends BaseTokenizer[String] {
  def tokenize(value: String): Token = Token(normalizer(value), TokenType.STRING)
  def tokenizeBytes(bytes: Array[Byte]): Token = tokenize(new String(bytes, encoding))
}