The Blag

Logic, Computer Graphics, OCaml, Rust, etc.

20 Mar 16

Format All the Data Structures

Last October, I moved in Nancy to start working as an engineer with Jasmin Blanchette on an exciting project called Nunchaku. Since my domain is formal logic, I spend a lot of time manipulating, transforming, and traversing ASTs (abstract syntax trees). My primary method for debugging is pretty-printing structures; in nunchaku, I even have a --print-all flag to print the AST after each transformation, since the tool is basically a bidirectional pipeline that composes transformations. I will certainly blog on the design of Nunchaku later, but today I want to present the standard Format module from OCaml, that I use for pretty-printing all my data structures.

I will also use containers because it provides some additional utils, in particular the CCFormat module contains predefined pretty-printers and printer combinators that are handy. Every module name starting with CC belongs to containers and not to the standard library. Even then, it is easy, although a bit tedious, to re-write the combinators from CCFormat using only what Format provides, so what I'll explain here can be done with vanilla OCaml, or with, say, the Fmt library.

A quick work on Fmt and CCFormat: both are modules that provide additional facilities on top of Format (combinators for printing lists, options, arrays, etc., ANSI codes for coloring the output) that makes life easier. Indeed, Format provides the pretty-printing algorithm, the support for format strings (same as Printf), basic printers, but does not provide more than the bare minimum for printing data structures, and that is why extension libraries exist.

A first taste

Let us say we have a value of type (int * bool) list, and we want to print it. We could print it "by hand" (using print_endline, print_int, etc. or even using Printf) but it would be nice to have proper alignment.

let l = CCList.init 100 (fun n-> n, n mod 2 = 0);;

Format.printf "l = [@[<hov>%a@]]@."
  CCFormat.(list ~start:"" ~stop:"" (pair int bool)) l;;

And we get something like this (depending on the margin, you might not get the exact same output):

l = [(1, false), (2, true), (3, false), (4, true),
     (5, false), (6, true), (7, false), (8, true),
     (9, false), (10, true), (11, false), (12, true),
     (13, false), (14, true), (15, false), (16, true),
     (17, false), (18, true), (19, false), (20, true),
     (21, false), (22, true), (23, false), (24, true),
     (25, false), (26, true), (27, false), (28, true),
     (29, false), (30, true), (31, false), (32, true),
     (33, false), (34, true), (35, false), (36, true),
     (37, false), (38, true), (39, false), (40, true),
     (41, false), (42, true), (43, false), (44, true),
     (45, false), (46, true), (47, false), (48, true),
     (49, false), (50, true), (51, false), (52, true),
     (53, false), (54, true), (55, false), (56, true),
     (57, false), (58, true), (59, false), (60, true),
     (61, false), (62, true), (63, false), (64, true),
     (65, false), (66, true), (67, false), (68, true),
     (69, false), (70, true), (71, false), (72, true),
     (73, false), (74, true), (75, false), (76, true),
     (77, false), (78, true), (79, false), (80, true),
     (81, false), (82, true), (83, false), (84, true),
     (85, false), (86, true), (87, false), (88, true),
     (89, false), (90, true), (91, false), (92, true),
     (93, false), (94, true), (95, false), (96, true),
     (97, false), (98, true), (99, false), (100, true)]

Nice, but what does this horrible "l = [@[<hov>%a@]]@." mean? It's a formatting string (same as in C printf), but more expressive.

  • The terminating @. prints a newline;
  • The @[<hov>......@] is a box that prints the same as ...., but will align its content either horizontally ("h") or vertically ("v"). The other kinds of boxes are "h" (horizontal), "v" (vertical), and "hv" (behaves either like "h" or like "v", but does not mix them). There is a tutorial about boxes on OCaml's website.
  • the "%a" is similar to "%d", "%s", etc. (which print respectively an integer and a string), but it's used for user-defined printers. For instance, Format.printf "hello %a" has type (Format.formatter -> 'a -> unit) -> 'a -> unit: it takes a first argument that is an 'a printer, then the 'a to print. Luckily (or not), the combinators in CCFormat have similar types, so for instance CCFormat.int : Format.formatter -> int -> unit, and CCFormat.(pair int bool) : Format.formatter -> (int * bool) -> unit. Here, CCFormat.pair is a printer combinator; the stdlib provides a few useful combinators such as Format.pp_print_list.

It is usually convenient, when you define some type t, to also define nearby a value val print : Format.formatter -> t -> unit. This way, it is straightforward to print values of type t when debugging time has come. Note that printers take a formatter as their first argument (to be used, typically, with Format.fprintf) which makes it possible to use them directly on files, etc. without creating a big string first.

Trees, Recursion, and nested Boxes

A case where boxes are really important is for printing recursive structures, such as trees. As far as I know, OCaml itself (the compiler) uses Format to print its intermediate ASTs, the signatures, etc. Let us define a small expression type, as an example:

type expr =
  | Add of expr * expr
  | Fun of bytes * expr
  | Const of int
  | Var of string
  | Let of string * expr * expr
  | App of expr * expr ;;

let e1 = Fun ("x", Add (Var "x", Const 1));;
val e1 : expr = Fun ("x", Add (Var "x", Const 1))

Ok, it does the job of representing a λ-calculus expression (with let-bindings and syntactic sugar for integers, wow!), but the representation of expressions is not very pretty. So let us use Format:

let rec print_expr out = function
  | Const i -> CCFormat.int out i
  | Var s -> CCFormat.string out s
  | Add (e1,e2) ->
    Format.fprintf out "@[<2>%a@ + %a@]" print_expr_inner e1 print_expr_inner e2
  | Let (x,e1,e2) ->
    Format.fprintf out "@[<v>@[<2>let %s =@ @[%a@] in@]@ %a@]"
      x print_expr e1 print_expr e2
  | App (e1,e2) ->
    Format.fprintf out "@[<2>%a@ %a@]" print_expr_inner e1 print_expr_inner e2
  | Fun (x,e) ->
    Format.fprintf out "@[<2>fun %s ->@ @[%a@]@]" x print_expr_inner e
and print_expr_inner out e = match e with
  | Const _ | Var _ -> print_expr out e
  | Add _ | Fun _ | App _ | Let _ ->
    Format.fprintf out "(@[%a@])" print_expr e
;;

where:

  • the argument out is the formatter we write into,
  • CCFormat.int is short for Format.pp_print_int,
  • the second printer is used to wrap an expression in parenthesis when needed to avoid ambiguities,
  • we put boxes around composite expressions to enforce proper indentation,
  • "@," is a break (can be replaced by a newline + indentation if it pleases the pretty-printer),
  • "@ " is either a space or a break,
  • "@[<2>" starts a box with indentation 2.

This is a very simple printer, and it could be improved, but it contains the gist of how to write a printer for recursive structures. Let see how it performs:

(* define a bigger expression *)
let e2 =
  Let ("f", e1,
    Let ("x",
      App (Var "f", App (Var "f", Const 0)),
      Let ("result",
        Let ("g",
          Fun ("y", App(Var "f", Add (Var "x", Var "y"))),
          App (Var "g", (Add (Const 40, Var "x")))),
        Add (Const 0, Var "result"))))
;;

(* print it: see the indentation *)
Format.printf "@[<2>e2 =@ %a@]@." print_expr e2;;
e2 =
  let f = fun x -> (x + 1) in
  let x = f (f 0) in
  let result = let g = fun y -> (f (x + y)) in
               g (40 + x) in
  0 + result
- : unit = ()

For a more realistic example, here is the main printer in Nunchaku. It's not perfect (it prints too many parenthesis) but works pretty well, as the following example shows. Just imagine if there was no indentation…

a pretty-printed term

The toplevel

The toplevel itself (or utop, you should probably use it!) uses Format for printing values. There is a pragma #install_printer to make it use your own printer:

type complex = Complex.t = { re: float; im: float; };;
let pp_complex out c =
  Format.fprintf out "%.2f + %.2fi" c.re c.im;;
#install_printer pp_complex;;

[ {re=1.0; im=(atan 1. *. 4.)}
; {re=0.; im=37.}
];;
- : complex list = [1.00 + 3.14i; 0.00 + 37.00i]

Yes, the toplevel will use the custom printer even inside structures such as lists.

Deriving printers

It is possible to derive printers automatically, using ppx_deriving.show (and possibly camlp4-based things, in versions anterior to 4.02, but ppx is the future). This makes printers even easier to use as, in general, you don't have to write anything to use them on your types (unless you want some specific behavior). It is as simple as:

#require "ppx_deriving.show";;

type foo =
  | A of int
  | B of string * bool list
  [@@deriving show] ;;

type bar = {
  foo1 : foo;
  foo2 : foo option
} [@@deriving show];;

let b = {foo1=A 42; foo2=Some (B ("a string", [true; false])); };;

Format.asprintf "b=%a" pp_bar b;;
- : string = "b={ foo1 = (A 42); foo2 = (Some B (\"a string\", [true; false])) }"

A word on performance

Adding GADTs to OCaml made it possible to clean the old magic code dealing with format strings that, once, gave Format a reputation for slowness (see the awesome work of Benoit Vaugon). I hope that flambda (the new optimization pass in OCaml 4.03) will reduce the performance overhead even further.

Conclusion

Format is great and you should use it! :)