Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Enhances stack safety for Eval. #1888

Merged
merged 5 commits into from
Sep 28, 2017
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
157 changes: 102 additions & 55 deletions core/src/main/scala/cats/Eval.scala
Original file line number Diff line number Diff line change
Expand Up @@ -72,28 +72,28 @@ sealed abstract class Eval[+A] extends Serializable { self =>
*/
def flatMap[B](f: A => Eval[B]): Eval[B] =
this match {
case c: Eval.Compute[A] =>
new Eval.Compute[B] {
case c: Eval.FlatMap[A] =>
new Eval.FlatMap[B] {
type Start = c.Start
// See https://issues.scala-lang.org/browse/SI-9931 for an explanation
// of why the type annotations are necessary in these two lines on
// Scala 2.12.0.
val start: () => Eval[Start] = c.start
val run: Start => Eval[B] = (s: c.Start) =>
new Eval.Compute[B] {
new Eval.FlatMap[B] {
type Start = A
val start = () => c.run(s)
val run = f
}
}
case c: Eval.Call[A] =>
new Eval.Compute[B] {
case c: Eval.Defer[A] =>
new Eval.FlatMap[B] {
type Start = A
val start = c.thunk
val run = f
}
case _ =>
new Eval.Compute[B] {
new Eval.FlatMap[B] {
type Start = A
val start = () => self
val run = f
Expand Down Expand Up @@ -203,7 +203,7 @@ object Eval extends EvalInstances {
* which produces an Eval[A] value. Like .flatMap, it is stack-safe.
*/
def defer[A](a: => Eval[A]): Eval[A] =
new Eval.Call[A](a _) {}
new Eval.Defer[A](a _) {}

/**
* Static Eval instance for common value `Unit`.
Expand Down Expand Up @@ -246,83 +246,130 @@ object Eval extends EvalInstances {
val One: Eval[Int] = Now(1)

/**
* Call is a type of Eval[A] that is used to defer computations
* Defer is a type of Eval[A] that is used to defer computations
* which produce Eval[A].
*
* Users should not instantiate Call instances themselves. Instead,
* Users should not instantiate Defer instances themselves. Instead,
* they will be automatically created when needed.
*/
sealed abstract class Call[A](val thunk: () => Eval[A]) extends Eval[A] {
def memoize: Eval[A] = new Later(() => value)
def value: A = Call.loop(this).value
}
sealed abstract class Defer[A](val thunk: () => Eval[A]) extends Eval[A] {

object Call {
def memoize: Eval[A] = Memoize(this)
def value: A = evaluate(this)
}

/**
* Collapse the call stack for eager evaluations.
*/
@tailrec private def loop[A](fa: Eval[A]): Eval[A] = fa match {
case call: Eval.Call[A] =>
loop(call.thunk())
case compute: Eval.Compute[A] =>
new Eval.Compute[A] {
/**
* Advance until we find a non-deferred Eval node.
*
* Often we may have deep chains of Defer nodes; the goal here is to
* advance through those to find the underlying "work" (in the case
* of FlatMap nodes) or "value" (in the case of Now, Later, or
* Always nodes).
*/
@tailrec private def advance[A](fa: Eval[A]): Eval[A] =
fa match {
case call: Eval.Defer[A] =>
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

totally nitpick: call can be renamed to defer and the compute below. No need to address in this PR, I can do in a separate one.

advance(call.thunk())
case compute: Eval.FlatMap[A] =>
new Eval.FlatMap[A] {
type Start = compute.Start
val start: () => Eval[Start] = () => compute.start()
val run: Start => Eval[A] = s => loop1(compute.run(s))
val run: Start => Eval[A] = s => advance1(compute.run(s))
}
case other => other
}

/**
* Alias for loop that can be called in a non-tail position
* from an otherwise tailrec-optimized loop.
*/
private def loop1[A](fa: Eval[A]): Eval[A] = loop(fa)
}
/**
* Alias for advance that can be called in a non-tail position
* from an otherwise tailrec-optimized advance.
*/
private def advance1[A](fa: Eval[A]): Eval[A] =
advance(fa)

/**
* Compute is a type of Eval[A] that is used to chain computations
* FlatMap is a type of Eval[A] that is used to chain computations
* involving .map and .flatMap. Along with Eval#flatMap it
* implements the trampoline that guarantees stack-safety.
*
* Users should not instantiate Compute instances
* Users should not instantiate FlatMap instances
* themselves. Instead, they will be automatically created when
* needed.
*
* Unlike a traditional trampoline, the internal workings of the
* trampoline are not exposed. This allows a slightly more efficient
* implementation of the .value method.
*/
sealed abstract class Compute[A] extends Eval[A] {
sealed abstract class FlatMap[A] extends Eval[A] { self =>
type Start
val start: () => Eval[Start]
val run: Start => Eval[A]

def memoize: Eval[A] = Later(value)

def value: A = {
type L = Eval[Any]
type C = Any => Eval[Any]
@tailrec def loop(curr: L, fs: List[C]): Any =
curr match {
case c: Compute[_] =>
c.start() match {
case cc: Compute[_] =>
loop(
cc.start().asInstanceOf[L],
cc.run.asInstanceOf[C] :: c.run.asInstanceOf[C] :: fs)
case xx =>
loop(c.run(xx.value), fs)
}
case x =>
fs match {
case f :: fs => loop(f(x.value), fs)
case Nil => x.value
}
}
loop(this.asInstanceOf[L], Nil).asInstanceOf[A]
def memoize: Eval[A] = Memoize(this)
def value: A = evaluate(this)
}

private case class Memoize[A](eval: Eval[A]) extends Eval[A] {
var result: Option[A] = None
def memoize: Eval[A] = this
def value: A =
result match {
case Some(a) => a
case None =>
val a = evaluate(this)
result = Some(a)
a
}
}


private def evaluate[A](e: Eval[A]): A = {
type L = Eval[Any]
type M = Memoize[Any]
type C = Any => Eval[Any]

def addToMemo(m: M): C = { a: Any =>
m.result = Some(a)
Now(a)
}

@tailrec def loop(curr: L, fs: List[C]): Any =
curr match {
case c: FlatMap[_] =>
c.start() match {
case cc: FlatMap[_] =>
loop(
cc.start().asInstanceOf[L],
cc.run.asInstanceOf[C] :: c.run.asInstanceOf[C] :: fs)
case mm@Memoize(eval) =>
mm.result match {
case Some(a) =>
loop(Now(a), c.run.asInstanceOf[C] :: fs)
case None =>
loop(eval, addToMemo(mm.asInstanceOf[M]) :: c.run.asInstanceOf[C] :: fs)
}
case xx =>
loop(c.run(xx.value), fs)
}
case call: Defer[_] =>
loop(advance(call), fs)
case m@Memoize(eval) =>
m.result match {
case Some(a) =>
fs match {
case f :: fs => loop(f(a), fs)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This line isn't tested. I am curious how come the random stack safety stress test didn't hit it, is it expected?

case Nil => a
}
case None =>
loop(eval, addToMemo(m) :: fs)
}
case x =>
fs match {
case f :: fs => loop(f(x.value), fs)
case Nil => x.value
}
}

loop(e.asInstanceOf[L], Nil).asInstanceOf[A]
}
}

Expand Down
6 changes: 3 additions & 3 deletions laws/src/main/scala/cats/laws/discipline/Arbitrary.scala
Original file line number Diff line number Diff line change
Expand Up @@ -90,9 +90,9 @@ object arbitrary extends ArbitraryInstances0 {

implicit def catsLawsArbitraryForEval[A: Arbitrary]: Arbitrary[Eval[A]] =
Arbitrary(Gen.oneOf(
getArbitrary[A].map(Eval.now(_)),
getArbitrary[A].map(Eval.later(_)),
getArbitrary[A].map(Eval.always(_))))
getArbitrary[A].map(a => Eval.now(a)),
getArbitrary[() => A].map(f => Eval.later(f())),
getArbitrary[() => A].map(f => Eval.always(f()))))

implicit def catsLawsCogenForEval[A: Cogen]: Cogen[Eval[A]] =
Cogen[A].contramap(_.value)
Expand Down
109 changes: 108 additions & 1 deletion tests/src/test/scala/cats/tests/EvalTests.scala
Original file line number Diff line number Diff line change
@@ -1,11 +1,14 @@
package cats
package tests

import scala.math.min
import cats.laws.ComonadLaws
import cats.laws.discipline.{BimonadTests, CartesianTests, ReducibleTests, SerializableTests}
import cats.laws.discipline.arbitrary._
import cats.kernel.laws.{GroupLaws, OrderLaws}
import org.scalacheck.{Arbitrary, Cogen, Gen}
import org.scalacheck.Arbitrary.arbitrary
import scala.annotation.tailrec
import scala.math.min

class EvalTests extends CatsSuite {
implicit val eqThrow: Eq[Throwable] = Eq.allEqual
Expand Down Expand Up @@ -140,4 +143,108 @@ class EvalTests extends CatsSuite {
isEq.lhs should === (isEq.rhs)
}
}

// the following machinery is all to faciliate testing deeply-nested
// eval values for stack safety. the idea is that we want to
// randomly generate deep chains of eval operations.
//
// there are three ways to construct Eval[A] values from expressions
// returning A (and which are generated by Arbitrary[Eval[A]]):
//
// - Eval.now(...)
// - Eval.later(...)
// - Eval.always(...)
//
// there are four operations that transform expressions returning
// Eval[A] into a new Eval[A] value:
//
// - (...).map(f)
// - (...).flatMap(g)
// - (...).memoize
// - Eval.defer(...)
//
// the O[A] ast represents these four operations. we generate a very
// long Vector[O[A]] and a starting () => Eval[A] expression (which
// we call a "leaf") and then compose these to produce one
// (deeply-nested) Eval[A] value, which we wrap in DeepEval(_).

case class DeepEval[A](eval: Eval[A])

object DeepEval {

sealed abstract class O[A]

case class OMap[A](f: A => A) extends O[A]
case class OFlatMap[A](f: A => Eval[A]) extends O[A]
case class OMemoize[A]() extends O[A]
case class ODefer[A]() extends O[A]

implicit def arbitraryO[A: Arbitrary: Cogen]: Arbitrary[O[A]] =
Arbitrary(Gen.oneOf(
arbitrary[A => A].map(OMap(_)),
arbitrary[A => Eval[A]].map(OFlatMap(_)),
Gen.const(OMemoize[A]),
Gen.const(ODefer[A])))

def build[A](leaf: () => Eval[A], os: Vector[O[A]]): DeepEval[A] = {

def restart(i: Int, leaf: () => Eval[A], cbs: List[Eval[A] => Eval[A]]): Eval[A] =
step(i, leaf, cbs)

@tailrec def step(i: Int, leaf: () => Eval[A], cbs: List[Eval[A] => Eval[A]]): Eval[A] =
if (i >= os.length) cbs.foldLeft(leaf())((e, f) => f(e))
else os(i) match {
case ODefer() => Eval.defer(restart(i + 1, leaf, cbs))
case OMemoize() => step(i + 1, leaf, ((e: Eval[A]) => e.memoize) :: cbs)
case OMap(f) => step(i + 1, leaf, ((e: Eval[A]) => e.map(f)) :: cbs)
case OFlatMap(f) => step(i + 1, leaf, ((e: Eval[A]) => e.flatMap(f)) :: cbs)
}

DeepEval(step(0, leaf, Nil))
}

// we keep this low in master to keep travis happy.
// for an actual stress test increase to 200K or so.
val MaxDepth = 100

implicit def arbitraryDeepEval[A: Arbitrary: Cogen]: Arbitrary[DeepEval[A]] = {
val gen: Gen[O[A]] = arbitrary[O[A]]
Arbitrary(for {
leaf <- arbitrary[() => Eval[A]]
xs <- Gen.containerOfN[Vector, O[A]](MaxDepth, gen)
} yield DeepEval.build(leaf, xs))
}
}

// all that work for this one little test.

test("stack safety stress test") {
forAll { (d: DeepEval[Int]) =>
try {
d.eval.value
succeed
} catch { case (e: StackOverflowError) =>
fail(s"stack overflowed with eval-depth ${DeepEval.MaxDepth}")
}
}
}

test("memoize handles branched evaluation correctly") {
forAll { (e: Eval[Int], fn: Int => Eval[Int]) =>
var n0 = 0
val a0 = e.flatMap { i => n0 += 1; fn(i); }.memoize
assert(a0.flatMap(i1 => a0.map(i1 == _)).value == true)
assert(n0 == 1)

var n1 = 0
val a1 = Eval.defer { n1 += 1; fn(0) }.memoize
assert(a1.flatMap(i1 => a1.map(i1 == _)).value == true)
assert(n1 == 1)

var n2 = 0
val a2 = Eval.defer { n2 += 1; fn(0) }.memoize
assert(Eval.defer(a2).value == Eval.defer(a2).value)
assert(n2 == 1)
}
}
}