Dear Haskellers, I've been working on variants of the following code: connect_from_file :: String -> IO Connection connect_from_file fi = do s <- BL.readFile fi cred <- (decode s :: Maybe Db_cred) c <- connect . reformat_cred $ cred return c What this code intends to do is read a config from a file in JSON, decode the JSON, and then open a connection based the decoded info. This code doesn't compile for several reasons, but I've left it in this form because I believe it is the most convenient for approaching my question: How should I handle errors here? Both readFile (from Bytestring) and connect (from mysql.simple) throw exceptions when things don't go their way, and decode uses maybe. I see four ways of handling these errors: 0) *All typed errors: *I could "capture" all the exceptions in Either types and use an EitherIO monad or transformer to automatically abort when a Left value appears. Now my function looks like: String -> IO (Either MyErrorType Connection) 1) *All exceptions: *I could throw an exception if Nothing appears during decoding. The function looks the same: String -> IO Connection 2) *A mix*: I could leave the connect and readFile functions alone, but return IO (Nothing) if the decoding doesn't work out. Now we have: String -> IO (Maybe Connection) 3) Not 0)-2) So what is the proper way to do things? Many Thanks, Hillary PS For what it is worth, I have read: 0) https://www.schoolofhaskell.com/user/commercial/content/ exceptions-best-practices . I'd love to get feedback on how to apply this article to my question if it is appropriate. 1) https://lexi-lambda.github.io/blog/2016/06/12/four-months-with-haskell/ . From this article and others, I get the sense that there is great confusion in the community as to how to handle errors correctly, and worse yet, that different approaches are not easily compatible.