Scala: ExecutionContext for future for-comprehension

Question

When I make a future, or apply methods like onSuccess and map, I can specify ExecutionContext for them.

For example,

val f = future {
  // code
} executionContext

f.map(someFunction)(executionContext)

f onSuccess {
  // code
} executionContext

However, if I use a for-comprehension of future, how can I specify ExecutionContext for the yield part?

for {
  f <- future1
  g <- future2
} yield {
  // code to be executed after future1 onSuccess and future2 onSuccess
  // What ExecutionContext runs this code?
} // (executionContext) here does not work

And, what ExecutionContext runs the code in yield if not specified?

---

EDIT

OK. Thanks to answers, I found something.
If I don't define or import implicit ExecutionContext (like Implicits.global), the for-comprehension does not compile. That means, for-comprehension uses implicit ExecutionContext.

Then, how can I use for-comprehension without implicit ExecutionContext, i.e. how to specify?

Covenantee

See my answer for the underlying reason of why the for comprehensions won't compile if you don't have an implicit specified. – Alginate

@flavian In simple situation, that works. But if there are two ExecutionContext to use for for-comprehension, how can I specify? It will show errors like ambiguous implicit values. I can block each time I define or import implicit vals or defs & for-comprehension, but is there another way? – Covenantee

just to be absolutely sure - do you mean a single for-comprehension with multiple generators (in which case flavian's answer works) or multiple consequent for-comprehensions? – Alginate

@TheTerribleSwiftTomato I meant a single for-comprehension with multiple generators. The solution was not that simple, but Scala doesn't seem to support explicit ExecutionContext for for-comprehension... – Covenantee

Answer 1

The ExecutionContext parameter is actually implicit. That means you can:

import scala.concurrent.ExecutionContext

implicit val context = ExecutionContext.fromExecutor(//etc)
for {
  f <- future1
  g <- future2
} yield {
  // code to be executed after future1 onSuccess and future2 onSuccess
  // What ExecutionContext runs this code?: the one above.
}

You also have a default, namely scala.concurrent.ExecutionContext.Implicits.global. This has as many threads as the processors on the running machine.

It won't be used by all Futures by default, you still have to import it.

Update: If you really want to specifiy, although it's not recommended, you can unwrap the for yield

val combined = futureA.flatMap(x => futureB)(context)

Pigeonwing

What if there are two implicit ExecutionContexts in scope and you need to select one? – Canterbury

@Canterbury you can use selective aliasing like this: import com.bla.bla.{ contextToDiscard => _, _ }. This will import just one of them into scope if it's a package import causing the problems. Another way is to move imports that bring in the executors to a lower scope. – Pigeonwing

Answer 2

Since for comprehensions are "mapped" to map/flatMap operations, and the ExecutionContext parameters of those are implicit, I guess you can try to add an implicit val in the local scope:

implicit val myContext:ExecutionContext = ...

.

I don't believe there is a "default" implicit ExecutionContext, but the most commonly used one is ExecutionContext.Implicits.global .

Perlis