The final view on typed sprintf and sscanf

It would be remiss not to mention the dual solution to the problem of
typed sprintf and sscanf sharing the same formatting
specification. The previous message defined the embedded
domain-specific language of formatting specifications in the initial
style, as a data type. The language can also be defined in the final
style. To the end user, the difference is hardly noticeable: all the
tests of the previous message work as they are (modulo a few
adjustments caused by the monomorphism restriction). However, whereas
the initial style required GADT, the final solution is entirely in
Haskell98. One often hears that hardly anything interesting can be
written in Haskell98. I submit that implementing type-indexed terms,
thought to require GADTs or similar dependent-type-like extensions,
ought to count as interesting.

Again, the formulation of the problem and the end-user interface
remain exactly the same as described in the previous message. Here are
a few examples:

> tp1 = sprintf $ lit "Hello world"
> -- "Hello world"
> ts1 = sscanf "Hello world" (lit "Hello world")  ()
> -- Just ()

> tp2 = sprintf (lit "Hello " ^ lit "world" ^ char) '!'
> -- "Hello world!"
> ts2 = sscanf "Hello world!" (lit "Hello " ^ lit "world" ^ char) id
> -- Just '!'

> fmt3 () = lit "The value of " ^ char ^ lit " is " ^ int