Applicative Effects

Parallellism

Two independent computations can be combined, giving potential for their parallel execution:

val foobar = foo *! bar

The *! operator is an alias for zipPar method (see Computation API).

The possibility of parallelism, depends on implementation of handlers. Parallelism is possible only when all handlers in the currently used effect stack, are implemented to permit parallelism.

  • If parallelization is possible, 2 fibers1 for foo and bar are implicitly forked. Upon joining, results of each contributing effect are composed, in a similar manner as in composed Applicative Functor2.

  • If parallelization is not possible, zipPar fallbacks to sequential zip.

This condition doesn’t apply to fibers forked & joined explicitly (WIP).

Parallelizability of predefined handlers

Predefined effect Predefined handler for this effect Is the handler parallelizable?
Reader default ✅
Writer local ✅
shared ✅
State local ❌
shared ✅
Error first ❌
all ✅
Choice first ✅
all ✅
Random local ✅
shared ✅
Console default ✅

 

State’s local handler is conceptually similar to standard State monad. State updates are chained linearly. It is inherently impossible to fork-and-join such chain without having to make arbitrary choice about information loss. For this reason, this handler prohibits parallellism. See Everything you didn’t want to know about StateT ⚠️λ3 video for more information.

Error’s first handler short-circuits the computation on the first raised error. This handler prohibits parallellism, because the second computation can’t be run, until it’s determined that the first one completes without error.

Example: Applicative vs. Monadic error

Depending on selected handler, given program will either:

  • Attempt to execute both branches sequentially, but will stop on the first error.

  • Attempt to execute both branches parallelly, and will collect both errors.

import turbolift.!!
import turbolift.effects.ErrorK

case object MyError extends ErrorK[List, String]

val program = MyError.raise("foo") &! MyError.raise("bar")
// program: Computation[Nothing, MyError] = turbolift.Computation@3b247ad

val result1 = program.handleWith(MyError.handlers.first).run
// result1: Either[List[String], Nothing] = Left(value = List("foo"))

val result2 = program.handleWith(MyError.handlers.all).run
// result2: Either[List[String], Nothing] = Left(value = List("foo", "bar"))

  1. Currently, fibers are not exposed to user. 🚧 WIP 🚧 ↩

  2. Turbolift does not use Applicative typeclass. It’s only mentioned as an analogy. ↩

  3. Warning: Haskell code ahead. ↩