Haskell in Production

I agree that haskellers have a tendency to disregard the cost of using sophisticated type machinery. Which is unfortunate because haskell's greatest strength (IMO) is its ability to opt in to strong static verification for the 10% of code where it provides the most value.

1. The first do in your snippet does nothing - it's redundant. In the actual code you linked, though, it sequences the `get "/" ...` and the `get "/:short" ...` expressions into a single application initializer. When the app starts up (i.e. when `app` is called), this outer "do" runs through your endpoint-defining expressions and registers each endpoint in turn with the HTTP server.

2. That `get` function takes two arguments: the URI pattern and the action to run if the pattern matches. The whole rest of your snippet _is_ the second argument! Because of the low precedence of the "$" operator, using "$" means you don't have to enclose all that code in parentheses. It is a tiny bit of syntactic sugar that is wildly popular in Haskell-land because Heaven forfend any of your code ends up looking like LISP. ;-)

3. The answer is in the monad being used in this case to sequence these steps together (`ActionT Text IO `). Think of `param "uri"` not as an expression returning a string, but a partially-constructed function that takes an HTTP request as an additional argument. The monad takes care of threading the request into this function when the request is actually handled, and `uri` will be bound to the resulting value.

4. I'm guessing it doesn't. Many type signatures in Haskell are optional, and are added if either 1) there's a genuine ambiguity in the code that needs to be clarified, 2) the author thinks it will make things clearer, or 3) the author hit a bug where the compiler inferred some crazy generic type that the author didn't intend or understand, and they put in some type signatures as sanity checks and constraints on what type is actually expected.

5. Completely agreed, especially since, semantically, it's almost always used as "Error/Success". Silly naming aside, though, it's extremely similar to Maybe, except that the failure case can contain a value (an error message, for example).

6. Think of your entire "action" (the second argument to get) as a function that takes a request and returns a response. Scotty takes care of invoking your action as necessary when a request comes in, and it also handles writing the response out for you.

7. liftIO takes an `IO String` computation as input, and transforms it into an `ActionT Text IO String` computation. This is done about the stupidest way possible: it returns an action that ignores the input request, generates a string, and "returns" it. liftIO is a heavily overloaded function that is defined for lots of monads that can easily incorporate IO computations, usually because they themselves are nothing more than dressed-up IO computations.

8. Yes, -> and is overloaded. = and <- are very different because the latter relies on a monad to define how a value is extracted from the right-hand side and fed to the next part of the computation. The empty pair of parentheses is like void in C - a function with a return type of () doesn't return any value. In the case of our HTML actions, an action of type ActionT Text IO () may have a side effect of writing a lovely HTML response, but it doesn't produce a value like `param "uri"` does - it just does the side effect and it's done. Yes, indentation is important in Haskell, it's less regular (or you might say "more expressive") than in other languages, and in the code you linked it looks like it got jumbled somewhere along the way.