This module is intended for high-performance programming. Concretely, the ExpandCase class can be used to force GHC to avoid compiling Reader, Endo or State-based code to unnecessary closures. You can observe its usage in Mikan.TypeChecking.Free.Generic. For a smaller example and some explanation, consider the following.

f :: Bool -> Endo Int
f b = case b of
  True  -> mempty
  False -> Endo (+ 10)

The desired -O1 output should be the following (ignoring newtype casts):

f = \b n -> case b of
  True  -> n
  False -> n + 10

A typical undesired output would be

f = \b -> case b of
  True  -> \n -> n
  False -> \n -> n + 10

Returning closures can be better or worse, depending on the program context. However, in high-performance situations we almost never want to return closures, and GHC is not nearly reliable enough at getting rid of the closures.

Using ExpandCase, we can write code as follows.

f :: Bool -> Endo Int
f b = expand \ret -> case b of
  True  -> ret mempty
  False -> ret $ Endo (+10)

Here, expand immediately introduces a lambda abstraction, and the ret continuation applies the "body" of the definition to the freshly abstracted variable. Hence, we get something like the following as an intermediate piece of Core:

f :: Bool -> Endo Int
f b = Endo \n -> case b of
  True  -> appEndo mempty n
  False -> appEndo (Endo (+10)) n

which is then reliably optimized to

f :: Bool -> Endo Int
f b = Endo \n -> case b of
  True  -> n
  False -> n + 10

NOTE: if you want to use this module, it is very strongly recommended that you check the Core of your code!