package ppx_let

ppx_let

A ppx rewriter for monadic and applicative let bindings, match expressions, and if expressions.

Overview

The aim of this rewriter is to make monadic and applicative code look nicer by writing custom binders the same way that we normally bind variables. In OCaml, the common way to bind the result of a computation to a variable is:

let VAR = EXPR in BODY

ppx_let simply adds two new binders: let%bind and let%map. These are rewritten into calls to the bind and map functions respectively. These functions are expected to have

val map  : 'a t -> f:('a -> 'b)   -> 'b t
val bind : 'a t -> f:('a -> 'b t) -> 'b t

for some type t, as one might expect.

These functions are to be provided by the user, and are generally expected to be part of the signatures of monads and applicatives modules. This is the case for all monads and applicatives defined by the Jane Street's Core suite of libraries. (see the section below on getting the right names into scope).

Parallel bindings

ppx_let understands parallel bindings as well. i.e.:

let%bind VAR1 = EXPR1 and VAR2 = EXPR2 and VAR3 = EXPR3 in BODY

The and keyword is seen as a binding combination operator. To do so it expects the presence of a both function, that lifts the OCaml pair operation to the type t in question:

val both : 'a t -> 'b t -> ('a * 'b) t

Some applicatives have optimized map functions for more than two arguments. These applicatives will export functions like map4 shown below:

val map4: 'a t -> 'b t -> 'c t -> 'd t -> f:('a -> 'b -> 'c -> 'd -> 'r) -> 'r t

In order to use these optmized functions, ppx_let provides the let%mapn syntax, which picks the right map{n} function to call based on the amount of applicatives bound by the syntax.

Match statements

We found that this form was quite useful for match statements as well. So for convenience ppx_let also accepts %bind and %map on the match keyword. Morally match%bind expr with cases is seen as let%bind x = expr in match x with cases.

If statements

As a further convenience, ppx_let accepts %bind and %map on the if keyword. The expression if%bind expr1 then expr2 else expr3 is morally equivalent to let%bind p = expr1 in if p then expr2 else expr3.

Function statements

We accept function%bind and function%map too.

let f = function%bind
  | Some a -> g a
  | None -> h

is equivalent to

let f = fun temp ->
  match%bind temp with
  | Some a -> g a
  | None -> h

While statements

We also expand while%bind expr1 do expr2 done as

let rec loop () =
  if%bind expr1
  then (
    let%bind () = expr2 in
    loop ())
  else return ()
in loop ()

Note that this form will (potentially) evaluate the textual form of expr1 multiple times!

We do not support while%map, as that cannot be implemented without bind.

Syntactic forms and actual rewriting

ppx_let adds seven syntactic forms

let%bind P = M in E

let%map  P = M in E

let%sub P = M in E

match%bind M with P1 -> E1 | P2 -> E2 | ...

match%map  M with P1 -> E1 | P2 -> E2 | ...

if%bind M then E1 else E2

if%map  M then E1 else E2

while%bind M do E done

that expand into

bind M ~f:(fun P -> E)


map  M ~f:(fun P -> E)

sub M ~f:(fun P -> E)

bind M ~f:(function P1 -> E1 | P2 -> E2 | ...)

map  M ~f:(function P1 -> E1 | P2 -> E2 | ...)

bind M ~f:(function true -> E1 | false -> E2)

map  M ~f:(function true -> E1 | false -> E2)

let rec loop () = bind M ~f:(function true -> bind E ~f:loop | false -> return ()) in loop ()

respectively.

As with let, let%bind and let%map also support multiple parallel bindings via the and keyword:

let%bind P1 = M1 and P2 = M2 and P3 = M3 and P4 = M4 in E

let%map  P1 = M1 and P2 = M2 and P3 = M3 and P4 = M4 in E

that expand into

let x1 = M1 and x2 = M2 and x3 = M3 and x4 = M4 in
bind
  (both x1 (both x2 (both x3 x4)))
  ~f:(fun (P1, (P2, (P3, P4))) -> E)

let x1 = M1 and x2 = M2 and x3 = M3 and x4 = M4 in
map
  (both x1 (both x2 (both x3 x4)))
  ~f:(fun (P1, (P2, (P3, P4))) -> E)

respectively. (Instead of x1, x2, ... ppx_let uses variable names that are unlikely to clash with other names)

Unlike let%map and let%bind, let%sub does not permit multiple bindings via the and keyword.

As with let, names introduced by left-hand sides of the let bindings are not available in subsequent right-hand sides of the same sequence.

Getting the right names in scope

The description of how the %bind and %map syntax extensions expand left out the fact that the names bind, map, both, and return are not used directly., but rather qualified by Let_syntax. For example, we use Let_syntax.bind rather than merely bind.

This means one just needs to get a properly loaded Let_syntax module in scope to use %bind and %map. The intended way to do this is to create a module Let_syntax with a signature like:

module Let_syntax : sig
  module Let_syntax : sig
    val bind : ...
    val map : ...
    ...
  end
  ...
end

and then use open Let_syntax to make the inner Let_syntax module available.

Alternatively, the extension can use values from a Let_syntax module other than the one in scope. If you write %map.A.B.C instead of %map, the expansion will use A.B.C.Let_syntax.Let_syntax.map instead of Let_syntax.map (and similarly for all extension points).

For monads, Core.Monad.Make produces a submodule Let_syntax of the appropriate form.

For applicatives, the convention for these modules is to have a submodule Let_syntax of the form:

module Let_syntax : sig
  module Let_syntax : sig
    val return : 'a -> 'a t
    val map    : 'a t -> f:('a -> 'b) -> 'b t
    val both   : 'a t -> 'b t -> ('a * 'b) t
    module Open_on_rhs : << some signature >>
  end
end

The Open_on_rhs submodule is used by variants of %map and %bind called %map_open and %bind_open. It is locally opened on the right hand sides of the rewritten let bindings in %map_open and %bind_open expressions. For match%map_open and match%bind_open expressions, Open_on_rhs is opened for the expression being matched on.

Open_on_rhs is useful when programming with applicatives, which operate in a staged manner where the operators used to construct the applicatives are distinct from the operators used to manipulate the values those applicatives produce. For monads, Open_on_rhs contains return.

let%sub

let%sub is a form almost equivalent to let%bind but calling a function called [sub] instead of [bind]. The intended use case is for things which have a "bind-like" operation with a type like:

val sub : 'a t -> f:('a s -> 'b t) -> 'b t

(e.g. a relative monad) The name comes from the quintessential example of such an operation: substitution of terms for variables. We didn't want to just use [let%bind] for such functions as it might confuse people.

There is one large difference between let%sub and let%bind stemming from the difference in the expected signatures of sub and bind. Since the value passed into f is not totally "unwrapped", it cannot be directly destructured. Because accessing the components of complex structures is often desirable, let%sub does the extra work. The below code snippet

let%sub a, b = c in
BODY

gets roughly translated to

let%sub temp_var = c in
let%sub a = return (map ~f:(fun (a, _) -> a) temp_var) in
let%sub b = return (map ~f:(fun (_, b) -> a) temp_var) in
BODY

The one potentially unexpected part of this is the usage of return followed immediately by let%sub, which seems like a no-op. Why not do this instead?

let%sub temp_var = c in
let a = map ~f:(fun (a, _) -> a) temp_var in
let b = map ~f:(fun (_, b) -> a) temp_var in
BODY

The difference is that the second option binds a and b to the computations that map temp_var to its components, but the first option binds a and b to the components after the mappings have occurred. Conceptually this means that for the first, correct desugaring, reusing a and b does not duplicate the mapping computation, but for the second desugaring, every usage of a or b refers to a duplicate of its computation.

match%sub

Rather than depending on a bind operation, match%sub depends on the presence of a switch function (the concrete Bonsai version is shown below). Note that the type of the expression being matched is Value.t rather than Computation.t.

val switch
  :  match_:int Value.t
  -> branches:int
  -> with_:(int -> 'a Computation.t)
  -> 'a Computation.t

Example:

let f (either_value : (_, _) Either.t Value.t) page1 page2 =
  let open Bonsai.Let_syntax in
  match%sub either_value with
  | First (a, b) -> page1 a b
  | Second x -> page2 x
;;

expands to roughly the following

let f (either_value : (_, _) Either.t Value.t) page1 page2 =
  let open Bonsai.Let_syntax in
  let%sub either_value = return either_value in
  Let_syntax.switch
    ~match_:
      (match%map either_value with
       | First (_, _) -> 0
       | Second _ -> 1)
    ~branches:2
    ~with_:(function
      | 0 ->
        let%sub a =
          return
            (match%map either_value with
             | First (a, _) -> a
             | _ -> assert false)
        in
        let%sub b =
          return
            (match%map either_value with
             | First (_, b) -> b
             | _ -> assert false)
        in
        page1 a b
      | 1 ->
        let%sub x =
          Bonsai.Let_syntax.return
            (match%map either_value with
             | Second x -> x
             | _ -> assert false)
        in
        page2 x
      | _ -> assert false)
;;

let%arr

One way of looking at arrow programs (i.e. programs involving let%sub) is that let%map...and builds a computation that does work, and let%sub saves the work to a variable so that other computations can share the result.

A common mistake is to forget to save a computation; when the computation gets used twice, the work must be done twice, rather than getting shared between the two occurrences. The form let%arr...and aims to eliminate such mistakes via the type-system. It extends the Let_syntax module with an arr function that lifts a function to an arrow.

val arr : 'a Value.t -> f:('a -> 'b) -> 'b Computation.t

The signature is the same as map except the result is a Computation.t, and not a Value.t. The implementation of arr is equivalent to return (map x ~f). Roughly the only thing you can do with a Computation.t is use let%sub to gain access to a Value.t handle to the computation. Thus, using arr forces the user to use as much sharing as possible. Of course, you can always duplicate work by copy-pasting code, but it won't happen accidentally.

An unrelated distinction of let%arr is that it tracks Lexing.position. That is, the signature of arr is actually the following:

val arr : ?here:Lexing.position -> 'a Value.t -> f:('a -> 'b) -> 'b Computation.t

The let%arr form is as a testing ground for source location tracking. It's likely that eventually all the other binding forms will also track location.