(very breaking) change in Future instances? about cats HOT 5 CLOSED

Note the 2.11 behavior matches that of Future.traverse (https://scastie.scala-lang.org/aU1EoSBnQu2rbWiXtEbcsQ)

from cats.

Right, it does 🤔.
But it doesn't match the traverse with IO: https://scastie.scala-lang.org/1JfGkbQOSTicrUyPaERQhw

Regardless, it still looks like a very breaking change.

from cats.

it still looks like a very breaking change

looks like, but is not. See previous discussion:

• #4176
• #4230

from cats.

The problem is Future, and hence this function test: Int => Future[Unit] is not pure.

When you call test a side effect immediately starts. That is not true of the IO version. When you are working with Future you are working with an impure type and we don't limit the traverse implementation of data types to try to work that.

A PR was created where a lot of energy was spent to optimize traverse implementations: #4498 -- I think it is very likely the change you are seeing is due to that.

I guess what is happening is that since Future has a StackSafeMonad, it is traversing directly (which is what the standard library does), but that is triggering the side-effects in a different order.

from cats.

looks like, but is not

It depends on how you look at it (or how much code you need to fix) 😆.

Thanks for chiming in, everyone! I'll close this one and will go hunting down my traverses :)

from cats.