Scala generics variance

Scala Generics and Variance

Core Concepts

Scala's type system is stricter than Python. Python allows flexible runtime type dispatch, while Scala enforces compile-time variance constraints. When translating Python generics to Scala:

Variance Modes in Scala

1. Invariant (T) - Default, exact type matching required

   • Can appear in both input (contravariant position) and output (covariant position)
   • Use for mutable containers or bidirectional types

2. Covariant (+T) - Subtypes allowed in output

   • Can return subtypes
   • Cannot accept as input (read-only, immutable)
   • Use for producers: class Container[+T], function returns

3. Contravariant (-T) - Supertypes allowed in input

   • Can accept supertypes
   • Cannot return (write-only)
   • Use for consumers: class Handler[-T], function parameters

Translation Patterns

Python TypeVar → Scala Type Parameter

Python (flexible):

T = TypeVar("T")                    # Invariant
T_co = TypeVar("T_co", covariant=True)
T_contra = TypeVar("T_contra", contravariant=True)

Scala (strict):

class Invariant[T]           // No annotation
class Container[+T]          // Covariant (producer)
class Handler[-T]            // Contravariant (consumer)

Python Generic Class → Scala Implementation

Python (duck typing):

class BaseTokenizer(ABC, Generic[T]):
    @abstractmethod
    def tokenize(self, value: T) -> Token:
        pass

Scala (type-safe):

abstract class BaseTokenizer[T] {
  def tokenize(value: T): Token
}

Covariant Container (Read-only)

Python:

class TokenContainer(Generic[T_co]):
    def __init__(self, items: Sequence[T_co]) -> None:
        self._items: tuple[T_co, ...] = tuple(items)

    def get_all(self) -> tuple[T_co, ...]:
        return self._items

Scala (producer, covariant):

class TokenContainer[+T](items: Seq[T]) {
  private val _items = items.toVector

  def getAll: Vector[T] = _items

  def mapTokens[S](func: T => String): List[String] =
    _items.map(func).toList
}

Why +T: The class only returns T values (covariant position), never accepts them as input.

Contravariant Handler (Write-only)

Python:

class TokenSink(Generic[T_contra]):
    def __init__(self) -> None:
        self._received: list[Any] = []

    def receive(self, item: T_contra) -> None:
        self._received.append(item)

Scala (consumer, contravariant):

class TokenSink[-T] {
  private var _received: List[Any] = List()

  def receive(item: T): Unit =
    _received = _received :+ item

  def drain(): List[Any] = {
    val result = _received
    _received = List()
    result
  }
}

Why -T: The class only accepts T values as input (contravariant position), never returns them typed as T.

Invariant Bidirectional (Get and Set)

Python:

class BivariantHandler(Generic[T]):
    def __init__(self, default: T) -> None:
        self._value: T = default

    def get(self) -> T:
        return self._value

    def set(self, value: T) -> None:
        self._value = value

Scala (both input and output):

class BivariantHandler[T](default: T) {
  private var _value: T = default

  def get: T = _value

  def set(value: T): Unit =
    _value = value

  def transform(func: T => T): T = {
    _value = func(_value)
    _value
  }
}

Why invariant (no +/-): Method both accepts T (set) and returns T (get), so requires exact type match.

Method Type Parameters vs Class Type Parameters

Type parameters can also appear at method level:

// Covariant class can use covariant type parameter in methods
class Container[+T](items: Seq[T]) {
  // This works: method output can depend on T
  def map[U](f: T => U): Container[U] = ???

  // This also works due to covariance bound
  def filtered[U >: T](pred: U => Boolean): Container[T] = ???
}

// Invariant class can use any method-level variance
class BivariantHandler[T](default: T) {
  def transform[U](func: T => U): U = func(_value)
}

Variance Bounds (Upper/Lower Bounds)

Python (implicit via unions and optional):

T_contra = TypeVar("T_contra", contravariant=True)
def process(item: T_contra) -> None: ...

Scala (explicit):

class Handler[-T] {
  def process(item: T): Unit = ???
}

// Method with lower bound: can accept T or any supertype
def acceptSuper[T >: MyBase](item: T): Unit = ???

// Method with upper bound: can accept T or any subtype
def acceptSub[T <: MyBase](item: T): Unit = ???

Key Rules to Remember

1. Covariant (+T) = read-only, immutable producers

   • Return T, never accept T as input
   • Example: immutable containers, function returns

2. Contravariant (-T) = write-only consumers

   • Accept T, never return typed as T
   • Example: callback handlers, event listeners

3. Invariant (no +/-) = bidirectional, mutable

   • Both accept and return T
   • Example: mutable containers, references

4. Default to invariant when unsure; Scala will tell you if variance is wrong (compile-time checking!)

Common Mistakes

❌ Don't use +T in mutable containers:

// WRONG: can't mutate
class MutableBag[+T] { def add(item: T) = ??? }

// RIGHT: use invariant
class MutableBag[T] { def add(item: T) = ??? }

❌ Don't mix variance in a single method:

// WRONG: T used in both positions
class Container[+T] { def swap(item: T) = ??? }

// RIGHT: use invariant or clarify purpose
class Container[T] { def swap(item: T): Unit = ??? }

✓ Use method-level type parameters for flexibility:

// OK: preserves covariance on class level
class Container[+T](items: Seq[T]) {
  def collect[U >: T](f: T => Option[U]): Seq[U] = ???
}