Chapter Four: Lifting

Something that we're going to sidestep almost entirely is a concept called monad transformers. We rarely use them directly in practice, even though they're all through our dependencies, or isolated to type class instances or more semantic helper functions. Remember that the purpose of this guide is to give a pragmatic up-and-running style introduction to being productive in Haskell. No one asks how the plumbing inside TypeScript promises work, but it's easy to follow along with the high-level syntax and concepts.
Levels
Your application will sometimes consist of several nested data types. For example:
1
Either Error (Maybe [a])
Copied!
You probably actually care about the a values, but need to work with type class instances for Either, Maybe, and List. We can think of these as "levels" of wrappers:
1
+--------Either--------+
2
|          +--Maybe--+ |
3
|          |         | |
4
| Error    |  [ a ]  | |           
5
|          |         | |
6
|          +---------+ |
7
+----------------------+
Copied!
Mixing Levels
Clearly there's a lot of layers that you can think about. If all you care about is the outermost Either, then it's pretty straightforward. To access the inner Maybe, you need to find a way to get into Either structure, do what you need at that level, and get back out again. You can have all of your functions wrap and unwrap the various layers, but that composes badly and leads to a lot of boilerplate.
The solution is to lift a function through the layers. Take something that works on one layer, and make it work on another. We can do this manually, but for all of the common data types, there are type classes to help us do this in a straightforward way. By far the most common one is liftIO.
Everything in a Haskell application occurs inside the IO type, which is where the horrible, unpredictable, tainted imperative world touches your beautiful, pristine functional clockwork.
Haskell doesn't expose the IO constructor (i.e. it's not exported, something you can do in your modules, too), so we want to avoid touching IO directly as much as possible because it's impossible to get fully out of it. At the end of the day your function will get used in a IO context (even if you're several layers down). Most of the Haskell style is creating clean, pure, high-level DSLs, and calling those inside IO.
MonadIO
IO actions occur frequently. Accessing CLI input, talking to the database, and making a network request are all things that you'll want to do on a regular basis. If the monad that you're working in also has a MonadIO instance, you get access to liftIO. Fission uses MonadRIO very frequently, which inherits from MonadIO. The best way to read liftIO is as "turn an IO function into one that happens in my current wrapper."
This example uses an IO-enabled wrapper: m. It's done this way to make it match easily with any other MonadIO instance, rather than using IO directly.
Fission.Timestamp
1
addM :: MonadIO m => Unstamped r -> m r
2
addM record = do
3
  now <- liftIO getCur
rentTime
4
  return $ add now record
Copied!
Specifically RIO
Let's look at something that uses a concrete wrapping type, in this case RIO.
This example uses RIO directly because it's used while setting up the database pool for our ambient config on app startup, and it's more convenient phrased this way for this scenario.
Fission.Storage.SQLite
1
connPool :: HasLogFunc cfg => DB.Path -> RIO cfg DB.Pool
2
connPool (DB.Path {getPath 
= path}) = do
3
  logDebug $ "Establishing DB pool for " <> displayShow path
4
​
5
  rawPool <- liftIO $ createPool (sqliteOpen path) seldaClose 4 2 10
6
  logDebug $ "DB pool stats: " <> displayShow rawPool
7
​
8
  return $ DB.Pool rawPool
Copied!
The function createPool :: IO (Pool SeldaConnection) needs to be brought into RIO. Remember that IO doesn't export a constructor for us to explicitly destructure and repackage! We need a way to turn IO a into RIO cfg a.
RIO stands for "Reader + IO." Clearly this has a MonadIO instance! 
 We can always write this function manually, but by far the easiest is the liftIO instance for RIO. The upshot is that it makes IO actions compatible with the rest of your RIO function.

Chapter Three: Ambient Config

Chapter Five: Error Handling

Last modified 2yr ago

Export as PDF

Copy link

Contents