Skip to main content

Caramel - Pipe operator support

Open
ErlangElixirCompilerOcamlfunctional-language
Contribution Type

This contribution is a new feature.

Introduction#

Contribution presentation
Caramel pipe operator

Project#

You can find the Caramel programming language presentation here.

Context#

This contribution adds pipe operator |> support to the Caramel language.

Here is a summary of the different steps in a compiler:

  1. Lexing the source into tokens
  2. Parsing the token stream into an AST (abstract syntax tree)
  3. Validating the AST
  4. Translating the AST

Current behavior#

In many functional programming languages, it is a common way to compose functions into a readable left-to-right "pipeline" of steps of computation.

For example in OCaml:

val (|>) : 'a -> ('a -> 'b) -> 'b

Implement the solution#

code blocks

The code blocks are intentionally incomplete for the sake of readability.
If you want to read the full code you'll find it in the PR link at the top.

changes

This PR being still Open, some parts are likely to change.
I will keep the article updated if any changes are made.

Caramel runtime#

The first idea was to expose the pipe operator as an external that maps to a caramel_runtime:pipe function.

Here is the pipe operator signature:

caramel/stdlib/caramel_runtime.ml
external ( |> ) : 'a -> ('a -> 'b) -> 'b = "pipe"

Here is the pipe operator definition:

caramel/stdlib/caramel_runtime.erl
pipe(A, B) -> B(A).

We will tell the compiler to map |> to our caramel_runtime pipe function:

caramel/compiler/ocaml_to_erlang/names.ml
let ocaml_to_erlang_primitive_op t =
match t with
(* ... *)
| "|>" ->
Name.qualified
~m:(Name.atom (Atom.mk "caramel_runtime"))
~f:(Name.atom (Atom.mk "pipe"))

Finally given this input:

let print_int number = Io.format "~0tp~n" [ number ]
let subtract x y = y - x
let main _ =
let divide x y = y / x in
10 |> subtract 2 |> divide 4 |> print_int

It will output:

-module(pipe).
-export([main/1]).
-export([print_int/1]).
-export([subtract/2]).
-spec print_int(_) -> ok.
print_int(Number) -> io:format(<<"~0tp~n">>, [Number | []]).
-spec subtract(integer(), integer()) -> integer().
subtract(X, Y) -> erlang:'-'(Y, X).
-spec main(_) -> ok.
main(_) ->
Divide = fun
(X, Y) -> erlang:'div'(Y, X)
end,
caramel_runtime:pipe(
caramel_runtime:pipe(
caramel_runtime:pipe(
10, subtract(2)
),
Divide(4)
),
fun print_int/1).

Look how the compiler has replaced our pipes.

The problem was that we needed to first implement partial application to make it work.
Partial application involves passing less than the full number of arguments to a function.

For example:

let add a b = a + b
let addOne = add 1

addOne is the result of partially applying add.
It is a function that takes an integer b and return b + 1.

In our example, subtract and Divide arr called with less arguments that they have to take.

So we went for another solution: rewrite the pipes into SSA (static single assignment).

Static single assignment#

The translation phase is broken down into 3 phases:

  • The AST is lowered into an IR (intermediate representation)
  • The IR is analyzed
  • The IR is translated

SSA introduces a new constraint: all variables are assigned exactly once.

The idea is to go from this:

let f x = x |> add 1 |> div 2 |> mult 3

To something like this:

f(X) ->
Caramel@Tmp1 = add(1, X),
Caramel@Tmp2 = div(2, Caramel@Tmp1),
Caramel@Tmp3 = mult(3, Caramel@Tmp2).

For now we will implement this logic in the ocaml translation side.

The goal is to go from a |> f |> g to let a_f = f a in let a_f_g = g a_f.

Inside caramel/compiler/ocaml_to_erlang/fun.ml there is a match clause for the function application.
We will add a special case for the pipe operator.

| Texp_apply (expr, args) ->
let name =
match
mk_expression expr ~var_names ~modules ~functions ~module_name
with
| Erlang.Ast.Expr_fun_ref { fref_name = n; _ } -> Expr.ident n
| x -> x
in
let args =
List.filter_map
(function
| _, None -> None
| _, Some arg ->
Some
(mk_expression arg ~var_names ~modules ~functions ~module_name))
args
in
Expr.apply name args

The idea is to check if the name variable is the pipe function and then do our logic to bind the result of f x to a new variable with a unique name.

if name = "<PIPE_OPERATOR>"
then (
let f = args[0] in
let x = args[1] in
let f_x = Expr.apply f x in
let new_var = "<NEW_VARIABLE_NAME>" in
let binding = Expr.bind new_var f_x in
Expr.var binding "<NEW_VARIABLE_NAME>"
)
else (
Expr.apply name args
)

Final result#

The final result allows us to compose our function pipeline in this way:

"function1_arg_1"
|> function1 "function1_arg_2"
|> function2 "function2_arg_1"
|> function3

Takeaway#

Problems encountered#

Understanding how the compiler works was the step that took me the most time.
It required me to understand a codebase I'm not used to deal with.

What did I learn ?#

This contribution allowed me to review some concepts about compilers and functional programming languages.