Format_tutorial
Principles
Line breaking is based on three concepts:
- boxes : a box is a logical pretty-printing unit, which defines a behaviour of the pretty-printing engine to display the material inside the box.
- break hints: a break hint is a directive to the pretty-printing engine that proposes to break the line here, if it is necessary to properly print the rest of the material. Otherwise, the pretty-printing engine never break lines (except "in case of emergency" to avoid very bad output). In short, a break hint tells the pretty printer that a line break here may be appropriate.
-
indentation rules: When a line break occurs, the pretty-printing engines fixes the indentation (or amount of leading spaces) of the new line using indentation rules, as follows:
- A box can state the extra indentation of every new line opened in its scope. This extra indentation is named box breaking indentation.
- A break hint can also set the additional indentation of the new line it may fire. This extra indentation is named hint breaking indentation.
- If break hint
bh
fires a new line within boxb
, then the indentation of the new line is simply the sum of: the current indentation of boxb
+ the additional box breaking indentation, as defined by boxb
+ the additional hint breaking indentation, as defined by break hintbh
.
Boxes
There are 4 types of boxes. (The most often used is the "hov" box type, so skip the rest at first reading).
-
horizontal box (h box, as obtained by the
open_hbox
procedure): within this box, break hints do not lead to line breaks. -
vertical box (v box, as obtained by the
open_vbox
procedure): within this box, every break hint lead to a new line. -
vertical/horizontal box (hv box, as obtained by the
open_hvbox
procedure): if it is possible, the entire box is written on a single line; otherwise, every break hint within the box leads to a new line. -
vertical or horizontal box (hov box, as obtained by the
open_box
oropen_hovbox
procedures): within this box, break hints are used to cut the line when there is no more room on the line. There are two kinds of "hov" boxes, you can find the details below. In first approximation, let me consider these two kinds of "hov" boxes as equivalent and obtained by calling theopen_box
procedure.
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 vertical or horizontal packing box (as obtained by the
open_hovbox
procedure): break hints are used to cut the line when there is no more room on the line; no new line occurs if there is enough room on the line. -
vertical or horizontal structural box (as obtained by the
open_box
procedure): similar to the "hov" packing box, the break hints are used to cut the line when there is no more room on the line; in addition, break hints that can show the box structure lead to new lines even if there is enough room on the current line.
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 (open_hovbox
), "[(---[(----[(---b)]b)]b)]" is printed as follows:
(--- (---- (---)))
If we replace the packing boxes by structural boxes (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:
- Boxes must be opened and closed consistently (
open_*
andclose_box
must be nested like parentheses). - Never hesitate to open a box.
- 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".
- 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 alet rec
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
ident = expression"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. - 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
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. - 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!
- 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
- "
@[
" open a box (open_box 0). You may precise the type as an extra argument. For instance@[<hov n>
is equivalent toopen_hovbox n
. - "
@]
" close a box (close_box ()
). - "
@
" output a breakable space (print_space ()
). - "
@,
" output a break hint (print_cut ()
). - "
@;<n m>
" emit a "full" break hint (print_break n m
). - "
@.
" end the pretty-printing, closing all the boxes still opened (print_newline ()
).
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
© 1995-2021 INRIA.
https://www.ocaml.org/releases/4.13/htmlman/libref/Format_tutorial.html