What is the relationship between Haskell FreeT and Coroutine
In the article "Coroutine Pipelines" in Monad.Reader Issue 19, the author defines the general type Coroutine
:
newtype Coroutine f m a = Coroutine
{ resume :: m (Either (f (Coroutine f m a)) a)
}
I noticed that this type is very similar to the type FreeT
from free
package:
data FreeF f a b = Pure a | Free (f b)
newtype FreeT f m a = FreeT
{ runFreeT :: m (FreeF f a (FreeT f m a))
}
It seems that they are FreeT
also Coroutine
isomorphic. Here are the functions that map functions to each other:
freeTToCoroutine
:: forall f m a. (Functor f, Functor m) => FreeT f m a -> Coroutine f m a
freeTToCoroutine (FreeT action) = Coroutine $ fmap f action
where
f :: FreeF f a (FreeT f m a) -> Either (f (Coroutine f m a)) a
f (Pure a) = Right a
f (Free inner) = Left $ fmap freeTToCoroutine inner
coroutineToFreeT
:: forall f m a. (Functor f, Functor m) => Coroutine f m a -> FreeT f m a
coroutineToFreeT (Coroutine action) = FreeT $ fmap f action
where
f :: Either (f (Coroutine f m a)) a -> FreeF f a (FreeT f m a)
f (Right a) = Pure a
f (Left inner) = Free $ fmap coroutineToFreeT inner
I have the following questions:
- What is the relationship between types
FreeT
andCoroutine
? Why didn't the author of Coroutine Pipelines use typeCoroutine
instead ofCoroutine
typeCoroutine
? - Is there some deeper relationship between free monads and coroutines? The types do not appear to be isomorphic.
-
Why aren't there popular streaming libraries in Haskell based on
FreeT
?Basic datatype in :
pipes
Proxy
data Proxy a' a b' b m r = Request a' (a -> Proxy a' a b' b m r ) | Respond b (b' -> Proxy a' a b' b m r ) | M (m (Proxy a' a b' b m r)) | Pure r
Basic datatype in :
conduit
Pipe
data Pipe l i o u m r = HaveOutput (Pipe l i o u m r) (m ()) o | NeedInput (i -> Pipe l i o u m r) (u -> Pipe l i o u m r) | Done r | PipeM (m (Pipe l i o u m r)) | Leftover (Pipe l i o u m r) l
I am guessing it would be possible to create datatypes
Proxy
orPipe
based onFreeT
, so I wonder why this is not done? Is this for performance reasons?The only hint
FreeT
I've seen in popular streaming libraries is pipes-group , which it usesFreeT
for group in streams.
source to share
To answer the second question, let's simplify the problem first by looking Free
. Free f a
allows you to construct f
-shaped AST a
-values ββfor subsequent reduction (aka, interpretation). Comparing monad transformers in the article with the non-free-free designs, we can simply choose Identity
to m
, as is usually done for the construction of the base of their monads transformers: Free f = FreeT Identity f
.
The first Monad Reader article features the removed monk Identity
trampling transformer, so let's start by looking at the unlighted version with elided:
data Trampoline a = Return a | Bounce (Trampoline a)
If you compare this with Free
data Free f r = Pure r | Free (f (Free f r))
and feel a little, we can see that all we really need to do is "remove" the structure f
as we previously "removed" the structure m
. So we have Trampoline = Free Identity
, again because it Identity
has no structure. This, in turn, means that this trampoline is FreeT Identity Identity
: a kind of degenerate coroutine with a trivial form and no way to use effects to determine whether to bounce or return. So the difference between this trampoline and the monad-trampoline transformer: the transformer allows you to jump using m
-actions.
With a bit of work, we can also see that generators and consumers are free monads for certain options f
, respectively, ((,) a)
and ((->) a)
. Their free versions of monad transformers similarly allow them to intersperse m
-actions (for example, the generator can ask for user input before committing). Coroutine
summarizes both the f
form of the AST (fixed on f ~ Identity
for Trampoline) and the type of effects that can alternate (fixed with no effects or m ~ Identity
) for Free
. That's for sure FreeT m f
.
Intuitively, if it Free f
is a monad for a pure construction of f
-shaped AST, then it FreeT m f
is a monad for constructing f
-shaped AST, alternating with the effects provided m
. In case you're squinting a bit, this is exactly what coroutines are: a complete generalization that parameterizes an initialized computation in both the shape of the AST constructed and the type of effects used to construct it.
source to share