Add AtomicMap

build.sbt

@@ -1148,7 +1148,12 @@ lazy val std = crossProject(JSPlatform, JVMPlatform, NativePlatform)
         ProblemFilters.exclude[DirectMissingMethodProblem](
           "cats.effect.std.Mutex#ConcurrentImpl.EmptyCell"),
         ProblemFilters.exclude[DirectMissingMethodProblem](
-          "cats.effect.std.Mutex#ConcurrentImpl.LockQueueCell")
+          "cats.effect.std.Mutex#ConcurrentImpl.LockQueueCell"),
+        // #4424, refactored private classes
+        ProblemFilters.exclude[IncompatibleMethTypeProblem](
+          "cats.effect.std.AtomicCell#AsyncImpl.this"),
+        ProblemFilters.exclude[IncompatibleMethTypeProblem](
+          "cats.effect.std.AtomicCell#ConcurrentImpl.this")
       )
   )
   .jsSettings(

docs/std/atomic-cell.md

@@ -5,10 +5,6 @@ title: Atomic Cell
 
 A synchronized, concurrent, mutable reference.
 
-Provides safe concurrent access and modification of its contents, by ensuring only one fiber
-can operate on them at the time. Thus, all operations except `get` may semantically block the
-calling fiber.
-
 ```scala mdoc:silent
 abstract class AtomicCell[F[_], A] {
   def get: F[A]
@@ -20,34 +16,45 @@ abstract class AtomicCell[F[_], A] {
 }
 ```
 
+Provides safe concurrent access and modification of its contents, by ensuring only one fiber
+can operate on them at the time. Thus, all operations except `get` may semantically block the
+calling fiber.
+
 ## Using `AtomicCell`
 
 The `AtomicCell` can be treated as a combination of `Mutex` and `Ref`:
+
 ```scala mdoc:reset:silent
-import cats.effect.{IO, Ref}
-import cats.effect.std.Mutex
+import cats.effect.IO
+import cats.effect.std.AtomicCell
 
 trait State
-class Service(mtx: Mutex[IO], ref: Ref[IO, State]) {
-  def modify(f: State => IO[State]): IO[Unit] = 
-    mtx.lock.surround {
-      for {
-        current <- ref.get
-        next <- f(current)
-        _ <- ref.set(next) 
-      } yield ()
-    }
+
+class Service(cell: AtomicCell[IO, State]) {
+  def modify(f: State => IO[State]): IO[Unit] =
+    cell.evalUpdate(f)
 }
 ```
 
-The following is the equivalent of the example above:
+### Example
+
+Imagine a random data generator,
+that requires running some effectual operations _(e.g. checking a database)_
+to produce a new value.
+In that case, it may be better to block than to repeat the operation.
+
 ```scala mdoc:reset:silent
 import cats.effect.IO
 import cats.effect.std.AtomicCell
 
-trait State
-class Service(cell: AtomicCell[IO, State]) {
-  def modify(f: State => IO[State]): IO[Unit] = 
-    cell.evalUpdate(current => f(current))
+trait Data
+
+class RandomDataGenerator(cell: AtomicCell[IO, Data]) {
+  // Generates a new random value.
+  def next: IO[Data] =
+    cell.evalUpdateAndGet(generate)
+
+  private def generate(previous: Data): IO[Data] =
+    ???
 }
 ```

docs/std/atomic-map.md

@@ -0,0 +1,78 @@
+---
+id: atomic-map
+title: Atomic Map
+---
+
+A total map from `K` to `AtomicCell[F, V]`.
+
+```scala mdoc:silent
+import cats.effect.std.AtomicCell
+
+trait AtomicMap[F[_], K, V] {
+  /**
+   * Access the AtomicCell for the given key.
+   */
+  def apply(key: K): AtomicCell[F, V]
+}
+```
+
+It is conceptually similar to a `AtomicMap[F, Map[K, V]]`, but with better ergonomics when
+working on a per key basis. Note, however, that it does not support atomic updates to
+multiple keys.
+
+Additionally, it also provide less contention: since all operations are performed on
+individual key-value pairs, the pairs can be sharded by key. Thus, multiple concurrent
+updates may be executed independently to each other, as long as their keys belong to
+different shards.
+
+## Using `AtomicMap`
+
+You can think of a `AtomicMap` like a `MapRef` that supports effectual updates by locking the underlying `Ref`.
+
+```scala mdoc:reset:silent
+import cats.effect.IO
+import cats.effect.std.AtomicMap
+
+trait State
+trait Key
+
+class Service(am: AtomicMap[IO, Key, State]) {
+  def modify(key: Key)(f: State => IO[State]): IO[Unit] =
+    am(key).evalUpdate(f)
+}
+```
+
+### Example
+
+Imagine a parking tower,
+where users have access to specific floors,
+and getting a parking space involves an effectual operation _(e.g. a database call)_.
+In that case, it may be better to block than repeat the operation,
+but without blocking operations on different floors.
+
+```scala mdoc:reset:silent
+import cats.effect.IO
+import cats.effect.std.AtomicMap
+
+trait Car
+trait Floor
+trait ParkingSpace
+
+class ParkingTowerService(state: AtomicMap[IO, Floor, List[ParkingSpace]]) {
+  // Tries to park the given Car in the solicited Floor.
+  // Returns either the assigned ParkingSpace, or None if this Floor is full.
+  def parkCarInFloor(floor: Floor, car: Car): IO[Option[ParkingSpace]] =
+    state(key = floor).evalModify {
+      case firstFreeParkingSpace :: remainingParkingSpaces =>
+        markParkingSpaceAsUsed(parkingSpace = firstFreeParkingSpace, car).as(
+          remainingParkingSpaces -> Some(firstFreeParkingSpace)
+        )
+
+      case Nil =>
+        IO.pure(List.empty -> None)
+    }
+
+  private def markParkingSpaceAsUsed(parkingSpace: ParkingSpace, car: Car): IO[Unit] =
+    ???
+}
+```

std/shared/src/main/scala/cats/effect/std/AtomicCell.scala

@@ -153,35 +153,29 @@ object AtomicCell {
       of(M.empty)(F)
   }
 
-  private[effect] def async[F[_], A](init: A)(implicit F: Async[F]): F[AtomicCell[F, A]] =
-    Mutex.apply[F].map(mutex => new AsyncImpl(init, mutex))
-
-  private[effect] def concurrent[F[_], A](init: A)(
-      implicit F: Concurrent[F]): F[AtomicCell[F, A]] =
-    (Ref.of[F, A](init), Mutex.apply[F]).mapN { (ref, m) => new ConcurrentImpl(ref, m) }
+  private[effect] def async[F[_], A](
+      init: A
+  )(
+      implicit F: Async[F]
+  ): F[AtomicCell[F, A]] =
+    Mutex.apply[F].map(mutex => new AsyncImpl(init, lock = mutex.lock))
 
-  private final class ConcurrentImpl[F[_], A](
-      ref: Ref[F, A],
-      mutex: Mutex[F]
+  private[effect] def concurrent[F[_], A](
+      init: A
   )(
       implicit F: Concurrent[F]
-  ) extends AtomicCell[F, A] {
-    override def get: F[A] = ref.get
-
-    override def set(a: A): F[Unit] =
-      mutex.lock.surround(ref.set(a))
+  ): F[AtomicCell[F, A]] =
+    (Ref.of[F, A](init), Mutex.apply[F]).mapN { (ref, mutex) =>
+      new ConcurrentImpl(ref, lock = mutex.lock)
+    }
 
+  // Provides common implementations for derived methods that depend on F being an applicative.
+  private[effect] sealed abstract class CommonImpl[F[_], A](
+      implicit F: Applicative[F]
+  ) extends AtomicCell[F, A] {
     override def modify[B](f: A => (A, B)): F[B] =
       evalModify(a => F.pure(f(a)))
 
-    override def evalModify[B](f: A => F[(A, B)]): F[B] =
-      mutex.lock.surround {
-        ref.get.flatMap(f).flatMap {
-          case (a, b) =>
-            ref.set(a).as(b)
-        }
-      }
-
     override def evalUpdate(f: A => F[A]): F[Unit] =
       evalModify(a => f(a).map(aa => (aa, ())))
 
@@ -192,12 +186,33 @@ object AtomicCell {
       evalModify(a => f(a).map(aa => (aa, aa)))
   }
 
-  private final class AsyncImpl[F[_], A](
+  private[effect] final class ConcurrentImpl[F[_], A](
+      ref: Ref[F, A],
+      lock: Resource[F, Unit]
+  )(
+      implicit F: Concurrent[F]
+  ) extends CommonImpl[F, A] {
+    override def get: F[A] =
+      ref.get
+
+    override def set(a: A): F[Unit] =
+      lock.surround(ref.set(a))
+
+    override def evalModify[B](f: A => F[(A, B)]): F[B] =
+      lock.surround {
+        ref.get.flatMap(f).flatMap {
+          case (a, b) =>
+            ref.set(a).as(b)
+        }
+      }
+  }
+
+  private[effect] final class AsyncImpl[F[_], A](
       init: A,
-      mutex: Mutex[F]
+      lock: Resource[F, Unit]
   )(
       implicit F: Async[F]
-  ) extends AtomicCell[F, A] {
+  ) extends CommonImpl[F, A] {
     @volatile private var cell: A = init
 
     override def get: F[A] =
@@ -206,17 +221,14 @@ object AtomicCell {
       }
 
     override def set(a: A): F[Unit] =
-      mutex.lock.surround {
+      lock.surround {
         F.delay {
           cell = a
         }
       }
 
-    override def modify[B](f: A => (A, B)): F[B] =
-      evalModify(a => F.pure(f(a)))
-
     override def evalModify[B](f: A => F[(A, B)]): F[B] =
-      mutex.lock.surround {
+      lock.surround {
         F.delay(cell).flatMap(f).flatMap {
           case (a, b) =>
             F.delay {
@@ -225,14 +237,93 @@ object AtomicCell {
             }
         }
       }
+  }
 
-    override def evalUpdate(f: A => F[A]): F[Unit] =
-      evalModify(a => f(a).map(aa => (aa, ())))
+  /**
+   * Allows seeing a `AtomicCell[F, Option[A]]` as a `AtomicCell[F, A]`. This is useful not only
+   * for ergonomic reasons, but because some implementations may save space.
+   *
+   * Setting the `default` value is the same as storing a `None` in the underlying `AtomicCell`.
+   */
+  def defaultedAtomicCell[F[_], A](
+      atomicCell: AtomicCell[F, Option[A]],
+      default: A
+  )(
+      implicit F: Applicative[F]
+  ): AtomicCell[F, A] =
+    new DefaultedAtomicCell[F, A](atomicCell, default)
 
-    override def evalGetAndUpdate(f: A => F[A]): F[A] =
-      evalModify(a => f(a).map(aa => (aa, a)))
+  private[effect] final class DefaultedAtomicCell[F[_], A](
+      atomicCell: AtomicCell[F, Option[A]],
+      default: A
+  )(
+      implicit F: Applicative[F]
+  ) extends CommonImpl[F, A] {
+    override def get: F[A] =
+      atomicCell.get.map(_.getOrElse(default))
 
-    override def evalUpdateAndGet(f: A => F[A]): F[A] =
-      evalModify(a => f(a).map(aa => (aa, aa)))
+    override def set(a: A): F[Unit] =
+      if (a == default) atomicCell.set(None) else atomicCell.set(Some(a))
+
+    override def evalModify[B](f: A => F[(A, B)]): F[B] =
+      atomicCell.evalModify { opt =>
+        val a = opt.getOrElse(default)
+        f(a).map {
+          case (result, b) =>
+            if (result == default) (None, b) else (Some(result), b)
+        }
+      }
+  }
+
+  implicit def atomicCellOptionSyntax[F[_], A](
+      atomicCell: AtomicCell[F, Option[A]]
+  )(
+      implicit F: Applicative[F]
+  ): AtomicCellOptionOps[F, A] =
+    new AtomicCellOptionOps(atomicCell)
+
+  final class AtomicCellOptionOps[F[_], A] private[effect] (
+      atomicCell: AtomicCell[F, Option[A]]
+  )(
+      implicit F: Applicative[F]
+  ) {
+    def getOrElse(default: A): F[A] =
+      atomicCell.get.map(_.getOrElse(default))
+
+    def unset: F[Unit] =
+      atomicCell.set(None)
+
+    def setValue(a: A): F[Unit] =
+      atomicCell.set(Some(a))
+
+    def modifyValueIfSet[B](f: A => (A, B)): F[Option[B]] =
+      evalModifyValueIfSet(a => F.pure(f(a)))
+
+    def evalModifyValueIfSet[B](f: A => F[(A, B)]): F[Option[B]] =
+      atomicCell.evalModify {
+        case None =>
+          F.pure((None, None))
+
+        case Some(a) =>
+          f(a).map {
+            case (result, b) =>
+              (Some(result), Some(b))
+          }
+      }
+
+    def updateValueIfSet(f: A => A): F[Unit] =
+      evalUpdateValueIfSet(a => F.pure(f(a)))
+
+    def evalUpdateValueIfSet(f: A => F[A]): F[Unit] =
+      atomicCell.evalUpdate {
+        case None => F.pure(None)
+        case Some(a) => f(a).map(Some.apply)
+      }
+
+    def getAndSetValue(a: A): F[Option[A]] =
+      atomicCell.getAndSet(Some(a))
+
+    def withDefaultValue(default: A): AtomicCell[F, A] =
+      defaultedAtomicCell(atomicCell, default)
   }
 }