If we want to rationalize the notational impact of all these structures, perhaps we should try to recycle the notation we already possess.
- McBride, Paterson, Applicative programming with effects
Effectful is a small macro library that allows you to write monadic code in a more natural style than that afforded by for
-comprehensions, embedding effectful expressions in other effectful expressions rather than explicitly naming intermediate results. The idea is similar to that of the Scala Async library, but generalized to arbitrary monads (not just Future
).
The Effectful library provides two basic dual operations: effectfully: A => M[A]
and unwrap: M[A] => A
(there is also !
, a postfix version of unwrap
). Intuitively, within an effectfully
block, we are allowed to treat impure (or effectful) values as if they were pure. If you think about it, this is exactly what it's like to program in a standard imperative programming language. For example, take this hypothetical code:
if (!db.lookup(key).isDefined)
db.add(key, value);
where db.lookup
and db.add
do the obvious side-effectful things of interacting with a remote database. In order to reify the side-effects of this snippet in the type system, we can define a monad for our database type (or just use IO
). Then, in Scala we could write something like this instead:
for {
optVal <- db.lookup(key)
_ <- optVal map (db.add(key, _)) getOrElse db.pure(())
} yield ()
But this seems to have lost something of the perspicuity of the original. Effectful lets us write it in the original style but with all effects documented in the type system:
effectfully {
if (!db.lookup(key).!.isDefined)
db.add(key, value).!
}
Notice the use of the postfix !
operator to indicate where effects are happening.
In your build.sbt
, add a dependency like so:
libraryDependencies += "org.pelotom" %% "effectful" % "1.0.1"
Now write some code using Effectful:
import scalaz._
import Scalaz._
import effectful._
import language.postfixOps
val xs = List(1,2,3)
val ys = List(true,false)
effectfully { (xs!, ys!) }
// ==> List((1,true), (1,false), (2,true), (2,false), (3,true), (3,false))
Here, the "effect" in question is nondeterminism.
In Scala we have for
-comprehensions as an imperative-looking syntax for writing monadic code, e.g.
for {
x <- foo
y <- bar(x)
z <- baz
} yield (y, z)
Each monadic assignment a <- ma
unwraps a pure value a: A
from a monadic value ma: M[A]
so that it can be used later in the computation. But this is a little less convenient than one might hope--frequently we would like to make use of an unwrapped value without having to explicitly name it. With Effectful we can write it inline, like so:
effectfully { (unwrap(bar(unwrap(foo))), unwrap(baz)) }
or using !
, simply
effectfully { (bar(foo!)!, baz!) }
Writing conditional expressions in for
comprehensions can get hairy fast:
for {
x <- foo
result <- if (x > 12) {
for {
a1 <- bar
a2 <- baz(a1)
} yield a2
} else {
for {
b1 <- boz
b2 <- biz
} yield b1 * b2
}
} yield result
With Effectful we can write this as:
effectfully {
if (foo.! > 12)
baz(bar!)!
else
(boz.! * biz.!)
}
Monadic match
/case
expressions are similarly easier to express with Effectful.
We can even unwrap
monadic values within loops; here's an example using the State
monad:
effectfully {
for (i <- xs; j <- ys)
put(get[Int].! + 2 * i).!
}.run(n)
Compare with a traditional imperative loop that does the same thing, but with side-effects:
var v = n
for (i <- xs; j <- ys)
v = v + 2 * i
Similarly, a for
-comprehension containing unwrap
s will sequence the effects of each monadic action it encounters, yielding all the results:
def fib(n: Int) = effectfully {
for (i <- 1 to n) yield {
val (x, y) = get[(Int, Int)].! // unfortunately we need to remind `get` what type of state it's dealing with
put((y, x + y)).!
x
}
} eval (1, 1)
// fib(20) ==> List(1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, 233, 377, 610, 987, 1597, 2584, 4181, 6765)
Here's an alternate version of the same function using Effectful with the ST
monad.
Just as for
-comprehensions transform your code into flatMap
s and map
s behind the scenes, effectfully
is a macro which transforms code using unwrap
/ !
into calls to bind
and pure
from Scalaz's Monad
type class. So Effectful only works with instances of Monad
(at the moment; there is a proposal to allow using superclasses of Monad
where only their limited functionality is needed).
The transformation of for
-loops and -comprehensions requires that your "iterable" type be an instance of the Scalaz Traverse
type class. Then, the idea is that "effectful" loops and comprehensions (those which contain unwrap
invocations) are transformed in the following way:
t map (x => ...)
becomesunwrap(t traverse (x => effectfully { ... }))
t flatMap (x => ...)
becomesunwrap(t traverse (x => effectfully { ... }) map (_.join))
t foreach (x => ...)
becomesunwrap(t traverse (x => effectfully { ... }) map (_ => ()))
t withFilter {x => ...}
becomesunwrap(t filterM (x => effectfully { ... }))
The flatMap
case implicitly adds the additional requirement that the "iterable" type have a Monad
instance, which it ought to since you're flatMap
ping it! And the withFilter
case just requires that you be traversing a type which has a filterM
method with the appropriate type; Scalaz defines this for List
and, if you import scalaz.std.indexedSeq.indexedSeqSyntax._
, any subtype of scala.collection.immutable.IndexedSeq
.
Within the lexical scope of a effectfully
block, not all invocations of unwrap
/ !
are valid; in particular:
- Function bodies cannot contain
unwrap
calls except in certain limited cases (anonymous functions passed tomap
,flatMap
,foreach
andwithFilter
). - By-name arguments cannot contain
unwrap
calls; these are essentially the same as function bodies. - It makes no sense to use
unwrap
outside of aneffectfully
block.
When unwrap
is used in an unsupported position, it will be flagged with an error.