A software engineer website

Polysemy: Design heuristics: Hiding vs Exposing

Gautier DI FOLCO May 14, 2023 [Haskell] #haskell #polysemy #design #effects systems

When it comes to interpreters relying on other effects, you can either hide or expose them.

For example, last year we introduced a Cache effect:

data Cache k v (m :: Type -> Type) a where
  Cached :: k -> Cache k v m v

makeSem ''Cache

runCache :: forall k v r. Ord k => (k -> Sem r v) -> InterpreterFor (Cache k v) r
runCache f =
  evalState @(M.Map k v) mempty
    . reinterpret
      ( \case
          Cached k -> do
            currentCache <- get @(M.Map k v)
            case currentCache M.!? k of
              Nothing -> do
                v <- raise $ f k
                put $ M.insert k v currentCache
                return v
              Just v -> return v

The main motivation are:

Note that you may want to expose it:

On another hand, you have effects relying on other monads, such as a Document relying on BloodHound, or multiples effects relying on a single one:

data DocumentEffect d (m :: Type -> Type) a where
  CreateDocument :: d -> DocumentEffect d m Id
  UpdateDocument :: Id -> (d -> d) -> DocumentEffect d m ()

makeSem ''DocumentEffect

interpreterBH :: forall d r. Member (Embed BH) r => IndexName -> InterpreterFor (DocumentEffect d) r
interpreterBH index =
  interpret $
      CreateDocument doc -> embed @BH $ BH.indexDocument index doc
      UpdateDocument docId f -> embed @BH $ BH.updateDocument index docId f

See the full the code here.