Format_tutorial

Using the Format module

Principles

Line breaking is based on three concepts:

Boxes

There are 4 types of boxes. (The most often used is the "hov" box type, so skip the rest at first reading).

Let me give an example. Suppose we can write 10 chars before the right margin (that indicates no more room). We represent any char as a - sign; characters [ and ] indicates the opening and closing of a box and b stands for a break hint given to the pretty-printing engine.

The output "--b--b--" is displayed like this (the b symbol stands for the value of the break that is explained below):

Within a "h" box:

--b--b--

Within a "v" box:

--b
--b
--

Within a "hv" box:

If there is enough room to print the box on the line:

--b--b--

But "---b---b---" that cannot fit on the line is written

---b
---b
---

Within a "hov" box:

If there is enough room to print the box on the line:

--b--b--

But if "---b---b---" cannot fit on the line, it is written as

---b---b
---

The first break hint does not lead to a new line, since there is enough room on the line. The second one leads to a new line since there is no more room to print the material following it. If the room left on the line were even shorter, the first break hint may lead to a new line and "---b---b---" is written as:

---b
---b
---

Printing spaces

Break hints are also used to output spaces (if the line is not split when the break is encountered, otherwise the new line indicates properly the separation between printing items). You output a break hint using print_break sp indent, and this sp integer is used to print "sp" spaces. Thus print_break sp ... may be thought as: print sp spaces or output a new line.

For instance, if b is break 1 0 in the output "--b--b--", we get

within a "h" box:

-- -- --

within a "v" box:

--
--
--

within a "hv" box:

-- -- --

or, according to the remaining room on the line:

--
--
--

and similarly for "hov" boxes.

Generally speaking, a printing routine using "format", should not directly output white spaces: the routine should use break hints instead. (For instance print_space () that is a convenient abbreviation for print_break 1 0 and outputs a single space or break the line.)

Indentation of new lines

The user gets 2 ways to fix the indentation of new lines:

When defining the box: when you open a box, you can fix the indentation added to each new line opened within that box.

For instance: open_hovbox 1 opens a "hov" box with new lines indented 1 more than the initial indentation of the box. With output "---[--b--b--b--", we get:

---[--b--b
     --b--

with open_hovbox 2, we get

---[--b--b
      --b--

Note: the [ sign in the display is not visible on the screen, it is just there to materialise the aperture of the pretty-printing box. Last "screen" stands for:

-----b--b
     --b--

When defining the break that makes the new line. As said above, you output a break hint using print_break sp indent. The indent integer is used to fix the additional indentation of the new line. Namely, it is added to the default indentation offset of the box where the break occurs.

For instance, if [ stands for the opening of a "hov" box with 1 as extra indentation (as obtained by open_hovbox 1), and b is print_break 1 2, then from output "---[--b--b--b--", we get:

   ---[-- --
         --
         --

Refinement on "hov" boxes

The "hov" box type is refined into two categories.

The difference between a packing and a structural "hov" box is shown by a routine that closes boxes and parentheses at the end of printing: with packing boxes, the closure of boxes and parentheses do not lead to new lines if there is enough room on the line, whereas with structural boxes each break hint will lead to a new line. For instance, when printing "[(---[(----[(---b)]b)]b)]", where "b" is a break hint without extra indentation (print_cut ()). If "[" means opening of a packing "hov" box (Format.open_hovbox), "[(---[(----[(---b)]b)]b)]" is printed as follows:

(---
 (----
  (---)))

If we replace the packing boxes by structural boxes (Format.open_box), each break hint that precedes a closing parenthesis can show the boxes structure, if it leads to a new line; hence "[(---[(----[(---b)]b)]b)]" is printed like this:

(---
 (----
  (---
  )
 )
)

Practical advice

When writing a pretty-printing routine, follow these simple rules:

  1. Boxes must be opened and closed consistently (open_* and Format.close_box must be nested like parentheses).
  2. Never hesitate to open a box.
  3. Output many break hints, otherwise the pretty-printer is in a bad situation where it tries to do its best, which is always "worse than your bad".
  4. Do not try to force spacing using explicit spaces in the character strings. For each space you want in the output emit a break hint (print_space ()), unless you explicitly don't want the line to be broken here. For instance, imagine you want to pretty print an OCaml definition, more precisely a let rec
    ident = expression
    value definition. You will probably treat the first three spaces as "unbreakable spaces" and write them directly in the string constants for keywords, and print "let rec" before the identifier, and similarly write = to get an unbreakable space after the identifier; in contrast, the space after the = sign is certainly a break hint, since breaking the line after = is a usual (and elegant) way to indent the expression part of a definition. In short, it is often necessary to print unbreakable spaces; however, most of the time a space should be considered a break hint.
  5. Do not try to force new lines, let the pretty-printer do it for you: that's its only job. In particular, do not use Format.force_newline: this procedure effectively leads to a newline, but it also as the unfortunate side effect to partially reinitialise the pretty-printing engine, so that the rest of the printing material is noticeably messed up.
  6. Never put newline characters directly in the strings to be printed: pretty printing engine will consider this newline character as any other character written on the current line and this will completely mess up the output. Instead of new line characters use line break hints: if those break hints must always result in new lines, it just means that the surrounding box must be a vertical box!
  7. End your main program by a print_newline () call, that flushes the pretty-printer tables (hence the output). (Note that the top-level loop of the interactive system does it as well, just before a new input.)

Printing to stdout: using printf

The format module provides a general printing facility "a la" printf. In addition to the usual conversion facility provided by printf, you can write pretty-printing indications directly inside the format string (opening and closing boxes, indicating breaking hints, etc).

Pretty-printing annotations are introduced by the @ symbol, directly into the string format. Almost any function of the Format module can be called from within a printf format string. For instance

For instance

printf "@[<1>%s@ =@ %d@ %s@]@." "Prix TTC" 100 "Euros";;
Prix TTC = 100 Euros
- : unit = ()

A concrete example

Let me give a full example: the shortest non trivial example you could imagine, that is the lambda calculus :)

Thus the problem is to pretty-print the values of a concrete data type that models a language of expressions that defines functions and their applications to arguments.

First, I give the abstract syntax of lambda-terms:

type lambda =
 | Lambda of string * lambda
 | Var of string
 | Apply of lambda * lambda
;;

I use the format library to print the lambda-terms:

open Format;;

let ident = print_string;;
let kwd = print_string;;
val ident : string -> unit = <fun>
val kwd : string -> unit = <fun>

let rec print_exp0 = function
| Var s ->  ident s
| lam -> open_hovbox 1; kwd "("; print_lambda lam; kwd ")"; close_box ()

and print_app = function
| e -> open_hovbox 2; print_other_applications e; close_box ()

and print_other_applications f =
  match f with
  | Apply (f, arg) -> print_app f; print_space (); print_exp0 arg
  | f -> print_exp0 f

and print_lambda = function
| Lambda (s, lam) ->
      open_hovbox 1;
      kwd "\\"; ident s; kwd "."; print_space(); print_lambda lam;
      close_box()
      | e -> print_app e;;
val print_app : lambda -> unit = <fun>
val print_other_applications : lambda -> unit = <fun>
val print_lambda : lambda -> unit = <fun>

Most general pretty-printing: using fprintf

We use the fprintf function to write the most versatile version of the pretty-printing functions for lambda-terms. Now, the functions get an extra argument, namely a pretty-printing formatter (the ppf argument) where printing will occur. This way the printing routines are more general, since they can print on any formatter defined in the program (either printing to a file, or to stdout, to stderr, or even to a string). Furthermore, the pretty-printing functions are now compositional, since they may be used in conjunction with the special %a conversion, that prints a fprintf argument with a user's supplied function (these user's supplied functions also have a formatter as first argument).

Using fprintf, the lambda-terms printing routines can be written as follows:

open Format;;

let ident ppf s = fprintf ppf "%s" s;;
let kwd ppf s = fprintf ppf "%s" s;;
val ident : Format.formatter -> string -> unit
val kwd : Format.formatter -> string -> unit

let rec pr_exp0 ppf = function
| Var s -> fprintf ppf "%a" ident s
| lam -> fprintf ppf "@[<1>(%a)@]" pr_lambda lam

and pr_app ppf = function
| e -> fprintf ppf "@[<2>%a@]" pr_other_applications e

and pr_other_applications ppf f =
match f with
| Apply (f, arg) -> fprintf ppf "%a@ %a" pr_app f pr_exp0 arg
| f -> pr_exp0 ppf f

and pr_lambda ppf = function
| Lambda (s, lam) ->
fprintf ppf "@[<1>%a%a%a@ %a@]" kwd "\\" ident s kwd "." pr_lambda lam
| e -> pr_app ppf e
;;
val pr_app : Format.formatter -> lambda -> unit
val pr_other_applications : Format.formatter -> lambda -> unit
val pr_lambda : Format.formatter -> lambda -> unit

Given those general printing routines, procedures to print to stdout or stderr is just a matter of partial application:

let print_lambda = pr_lambda std_formatter;;
let eprint_lambda = pr_lambda err_formatter;;
val print_lambda : lambda -> unit
val eprint_lambda : lambda -> unit