Polysemy: Design heuristics: Hiding interpreters

In the codebases I have worked on all have multiple global interpreters, most noticeably for production and tests.

From time to time, you want to have the same effect lists, for example to dynamically change the interpretation.

To do so you can inject effects:

Let's take back our Pause effect:

data Pause (m :: Type -> Type) a where
  Pause :: Int -> Pause m ()

makeSem ''Pause

We can have two interpreters:

runPause :: Member (Embed IO) r => InterpreterFor Pause r
runPause =
  interpret $
    \case
      Pause x -> embed $ threadDelay x

runPause :: Member (Embed IO) r => InterpreterFor Pause r
runPause =
  interpret $
    \case
      Pause x -> embed $ threadDelay x

Then we can have two global interpreters:

interpretEmbed :: Sem '[Pause] a -> IO a
interpretEmbed = runM . runPauseEmbed . raiseUnder @(Embed IO)

interpretFinal :: Sem '[Pause] a -> IO a
interpretFinal = runFinal . runPauseFinal . raiseUnder @(Final IO)

raiseUnder injects one effect under the current head.

There are plenty of other functions such as raiseUnder2 and raiseUnder3, which adds 2/3 effects under the head, raise2Under/raise3Under which adds an effect in 2nd/3rd position under the effects' head.

More generally raise_/insertAt can helps you to acheive that with more flexibility.

Finally subsume_ will give you the power to rewrite your effects list (effects order, duplicate or merge effects).

See the full the code here.