ParAff does not run thing in Parallel

[garyb]

Exemplified by:

parTraverse (later' 2000 <<< log) ["hello", "goodbye"]

👀

[garyb]

Actually, something else might be going on here - the above is still broken though.

[paf31]

My guess would be that

derive newtype instance applicativeParAff :: Applicative (ParAff e)

is wrapping up the superclass instance for Apply. We should probably disallow that actually, for this reason.

[garyb]

Ah interesting, thanks! That does sound suspect. This was driving me a bit crazy, since this works fine:

sequential $ (\x y -> unit) <$> (parallel $ later' 2000 $ log "foo") <*> (parallel $ later' 2000 $ log "bar")

What is it we should diallow, deriving newtype instances for any class underneath a normal instance in the hierarchy?

[garyb]

That is exactly it, thanks so much!

[natefaubion]

I think this is pretty interesting though. I assume what's happening is that, since the dictionary is just coerced, if you ask for an Applicative and call apply, you'll get the entailed Apply instance. The Apply instance declared is different than the Apply instance you get through entailment on the derived Applicative. I'd consider this a bug in newtype deriving.

[jdegoes]

💯

[garyb]

purescript/purescript#2399