Some form of `contextalias Foo = context(Bar, Baz,...

# language-evolution

[Youssef Shoaib [MOD]]

Some form of contextalias Foo = context(Bar, Baz, ...) thing for context parameters would be really nice! The alternative to this is a song-and-dance boilerplate-y ritual of having Bar and Baz be interfaces, having interface Foo: Bar, Baz, ... and then packing them together using interface delegation. Or of course one can repeatedly mention them together, but it'd be really nice if we can have that idea of a set of related context parameters that one might want to use (of course, it might be hard to name them, but perhaps it can only work with context(_: Foo) for now). This is nice too if you have transformers on Baz for instance since then you can do:

context(_: Foo)
fun foo() = withTransformendBaz {
  otherFoo()
}
context(_: Foo)
fun otherFoo()

and thus otherFoo will get the initial Bar and the transformed Baz. Note that this example translates to:

context(_: Bar, _: Baz)
fun foo() = withTransformendBaz {
  otherFoo()
}
context(_: Bar, _: Baz)
fun otherFoo()

and so it truly should be just a simple macro-expansion with no special semantics or anything.

[phldavies]

I can certainly see the use for this. The downside being overuse of a common "context" on all functions, especially when most of the constituent parts may not be required. From the flip side, would you expect a function requiring the aliased context to be callable should the individual contexts be available? If so, you're also obscuring the true signature somewhat.

[Youssef Shoaib [MOD]]

Yes I'd expect it to be callable given the elements. It feels like if the IDE just inline expands the signature to show the constituent elements then it should be fine. Overuse of a common context is a very real problem, but at least IDEA can very easily give a warning for that and even offer a refactoring to only keep the necessary elements. With the alternative mentioned of using interfaces, IDEA won't be able to give you any warnings for that because there's no indication that the interface exists solely for its elements.

[dmitriy.novozhilov]

cc @Alejandro Serrano.Mena

[Youssef Shoaib [MOD]]

Since Alejandro is getting mentioned, a great example of this pattern is IorRaise in Arrow. It's currently composed of a Raise and an accumulate function, which could be its own standalone Accumulate interface. Then, you could have contextalias IorRaise<E> = context(Raise<E>, Accumulate<E>).
Another, much simpler example, is ScopedRaiseAccumulate which is just RaiseAccumulate + CoroutineScope. In fact, RaiseAccumulate<E> is really just Raise<E> + Raise<Nel<E>>, so ScopedRaiseAccumulate<E> is context(Raise<E>, Raise<Nel<E>>, CoroutineScope) . You can see how some of these interfaces are solely there to combine 2, while some also kind-of gain meaning on their own.
Having this feature makes it really easy to fix the recover issue with Arrow builders because replacing Raise becomes really simple instead of having to recreate a Raise subclass. I'd be happy to expand more on that issue if you'd like