Integrating Megaparsec with Effectful

[janmasrovira]

Is it possible to define an effect that allows an instance of MonadParsec for effectul's Eff?

instance ParsingEff e s :> r => MonadParsec e s (Eff r) where
  ...

For my particular case, it would be enough to have MonadParsec Void Text (Eff r).

So far, I've been unable to do it successfully. This post poses the same question and explains the situation quite well:
https://discourse.haskell.org/t/integrating-megaparsec-with-effectful/9954

[arybczak]

I didn't look closely, but I'm guessing that you should piggyback on ParsecT for everything. https://gist.github.com/CT075/6e0541728d9547e9ddf0b36a8ba50b56 looks generally alright, but I think it should probably use a State effect that keeps track of megaparsec's State s e and use runParserT' to preserve it between operations.

Something like this:

runParsec :: Eff (NonDet : Parsec e s : es) a -> Eff es a
runParsec = evalState initialParserState . runParsec' HANDLER . runNonDetParsec HANDLER . inject {- to hide the State -}

I'm not sure if it works out in the end though. Maybe it can't be written and you need to work with ParsecT .. (Eff es) directly.

[CT075]

Hi, I'm the author of the original linked post and snippet, I wasn't able to make this work.

This was several months ago and I don't remember what I did or did not try, but I didn't think of using a state to hold onto the state returned by runParserT'. Ultimately I decided that working with ParsecT .. (Eff es) ended up being the path of least resistance.

There's some additional subtleties in that using NonDet here causes a divergence the error behavior of ParsecT even if both branches fail (I believe it's something to do with which error gets "caught" and returned to the user).