166: Monadic Formatting

by Alex Shinn


This SRFI is currently in final status. Here is an explanation of each status that a SRFI can hold. To provide input on this SRFI, please send email to srfi-166@nospamsrfi.schemers.org. To subscribe to the list, follow these instructions. You can access previous messages via the mailing list archive.

Table of Contents


A library of procedures for formatting Scheme objects to text in various ways, and for easily concatenating, composing and extending these formatters efficiently without resorting to capturing and manipulating intermediate strings.

This SRFI is an updated version of SRFI 159, primarily with the difference that state variables are hygienic.

Summary of differences from SRFI 159:


There are several approaches to text formatting. Concatenating strings to display is not acceptable, since it doesn't scale to very large output. The simplest realistic idea, and what people resort to in typical portable Scheme, is to interleave display and write and manual loops, but this is both extremely verbose and doesn't compose well. A simple concept such as padding space can't be achieved directly without somehow capturing intermediate output.

The traditional approach in other languages is to use templates - typically strings, though in theory any object could be used and indeed Emacs's mode-line format templates allow arbitrary sexps. Templates can use either escape sequences (as in C's printf and Common Lisp's format) or pattern matching (as in Visual Basic's Format, Perl6's form, and SQL date formats). The primary disadvantage of templates is the relative difficulty (usually impossibility) of extending them, their opaqueness, and the unreadability that arises with complex formats. Templates are not without their advantages, but they are already addressed by other libraries such as SRFI 28 and SRFI 48.

Another important aspect of formatting is state. Common Lisp format provides a "fresh-line" format spec which outputs a newline only if the output stream is not already at the beginning of a line. C++ iostreams allow changing the radix and floating-point precision for numeric output, not just for a single value but as a persistent setting for all future output. Custom formatters which could manipulate their own state would allow for many new possibilities.

This SRFI takes a combinator approach to solving both problems. Formatters are defined, which are called to produce their output as needed, composed with other formatters, and refer to and update arbitrary state. The primary goal of this SRFI is to have a maximally expressive and extensible formatting library. The next most important goal is scalability — to be able to handle arbitrarily large output and not build intermediate results except where necessary. The third goal is brevity and ease of use.


show each each-in-list
displayed written written-shared written-simply
escaped maybe-escaped
numeric numeric/comma numeric/si numeric/fitted
nl fl space-to tab-to nothing
joined joined/prefix joined/suffix
joined/last joined/dot joined/range
padded padded/right padded/both
trimmed trimmed/right trimmed/both
trimmed/lazy fitted fitted/right fitted/both
fn with with! forked call-with-output
port row col width output writer
string-width substring/width
pad-char ellipsis
radix precision decimal-sep decimal-align
sign-rule comma-rule comma-sep
word-separator? ambiguous-is-wide?
pretty pretty-shared pretty-simply pretty-with-color
columnar tabular wrapped wrapped/list wrapped/char
justified from-file line-numbers
string-terminal-width string-terminal-width/wide
substring-terminal-width substring-terminal-width/wide
upcased downcased
as-red as-blue as-green as-cyan
as-yellow as-magenta as-white as-black
as-bold as-italic as-underline
as-color as-true-color
on-red on-blue on-green on-cyan
on-yellow on-magenta on-white on-black
on-color on-true-color

Types and Naming Conventions

We introduce two new types, formatters, which are disjoint from any type except possibly procedures, and state variables, which are distinct from any type except possibly SRFI 39 parameters. These are in fact identical to the SRFI 165 computations and computation environment variables, respectively, though knowledge of SRFI 165 is not required to use this SRFI.

In the prototypes below the following naming conventions imply type restrictions:


The SRFI is divided into a core implementation and three utility libraries, which could be defined portably in terms of the core but are provided as convenience extensions. The libraries are as follows:

  (srfi 166)           ; composite of all of the following
  (srfi 166 base)      ; all bindings not in one of the following
  (srfi 166 pretty)    ; all bindings in Pretty Printing
  (srfi 166 columnar)  ; all bindings in Columnar Formatting
  (srfi 166 unicode)   ; all bindings in Unicode
  (srfi 166 color)     ; all bindings in Formatting with Color


(show output-dest fmt ...)

The entry point for all formatting. Applies the fmt formatters in sequence, accumulating the output to output-dest. As with SRFI 28 format, output-dest can be an output port, #t to indicate the current output port, or #f to accumulate the output into a string and return that as the result of show.

Each fmt should be a formatter as discussed below. As a convenience, non-formatter arguments are also allowed and are formatted as if wrapped with displayed, described below, so that

    (show #f "π = " (with ((precision 2)) (acos -1)) nl)
would return the string "π = 3.14\n".

As mentioned, formatters are an opaque type and cannot directly be applied outside of show. Custom formatters are built on the existing formatters, and as first-class objects may be named or computed dynamically, so that:

  (let ((~.2f (lambda (x) (with ((precision 2)) x))))
    (show #f "π = " (~.2f (acos -1)) nl))
produces the same result. For typical uses you only need to combine the existing high-level formatters described in the succeeding sections, but see the section Higher Order Formatters and State for control flow and state-manipulation primitives.

The return value of show is the accumulated string if output-dest is #f and unspecified otherwise.

Formatting Objects

(displayed obj)

If obj is a formatter, returns obj as is. Otherwise, outputs obj using display semantics. Specifically, strings are output as if by write-string and characters are written as if by write-char. Other objects are output as with written (including nested strings and chars inside obj). This is the default behavior for top-level formats in show, each and most other high-level formatters.

It is an error if obj is a procedure which is not a formatter.

(written obj)

Outputs obj using write semantics. Uses the current numeric formatting settings to the extent that the written result can still be passed to read, possibly with loss of precision. Specifically, the current radix is used if set to any of 2, 8, 10 or 16, and the fixed-point precision is used if specified and the radix is 10.

(show #f (written (cons 0 1)))
=> "(0 . 1)"
(show #f 1.5 " " (with ((precision 0)) 1.5))
=> "1.5 2"
(show #f 1/7 " " (with ((precision 3)) 1/7)
             " " (with ((precision 20)) 1/7))
=> "1/7 0.143 0.14285714285714285714"

Implementations should allow arbitrary precision for exact rational numbers. For example:

(show #f (with ((precision 50)) 1/3))
=> "0.33333333333333333333333333333333333333333333333333"

As a less obvious example, using string-segment from SRFI 152, the following code returns the first 100 Fibonacci numbers:

(map string->number
      (show #f (with ((precision 2500))
                 (/ 1000 (- #e1e50 #e1e25 1))))

If you don't know the type of an object and want to print it out for debugging purposes, you should always wrap it with written or one of its variants, in case the object is itself a formatter. Note that, for debugging, a convenient idiom is to wrap the object(s) in a quasiquote list:

(define (add x y)
  (show #t `(add x: ,x y: ,y) nl)
  (+ x y))
(written-shared obj)

Like written, but using data labels for shared structures among all pairs and vectors, analogous to write-shared.

(written-simply obj)

Like written, but doesn't handle shared structures, analogous to write-simply. Infinite loops can still be avoided if used inside a formatter that truncates data (see trimmed-lazy below).

(escaped str [quote-ch esc-ch renamer])

Outputs the string str, escaping any quote or escape characters. If esc-ch, which defaults to #\\, is #f, escapes only the quote-ch, which defaults to #\", by doubling it, as in SQL strings and CSV values. If renamer is provided, it should be a procedure of one character which maps that character to its escape value, e.g. #\newline => #\n, or #f if there is no escape value.

(show #f (escaped "hi, bob!"))
=> "hi, bob!"

(show #f (escaped "hi, \"bob!\""))
=> "hi, \"bob!\""
(maybe-escaped str pred [quote-ch esc-ch renamer])

Like escaped, but first checks if any quoting is required (by the existence of either any quote or escape characters, or any character matching pred), and if so outputs the string in quotes and with escapes. Otherwise outputs the string as is. This is useful for quoting symbols and CSV output, etc.

(show #f (maybe-escaped "foo" char-whitespace? #\"))
=> "foo"
(show #f (maybe-escaped "foo bar" char-whitespace? #\"))
=> "\"foo bar\""
(show #f (maybe-escaped "foo\"bar\"baz" char-whitespace? #\"))
=> "\"foo\"bar\"baz\""

Formatting Numbers

(numeric num [radix precision sign-rule comma-rule comma-sep decimal-sep])

Formats a single number num. You can optionally specify any radix from 2 to 36 (even if num isn't an integer). precision forces a fixed-point format.

A sign-rule of #t indicates to output a plus sign (+) for positive integers. However, if sign-rule is a pair of two strings, it means to wrap negative numbers with the two strings. For example, ("(" . ")") prints negative numbers in parentheses, financial style: -1.99 => (1.99).

comma-rule is an integer specifying the number of digits between commas, or a list of integers representing the number of digits between each successive comma, with the first being the least significant digits and the last repeating.

comma-sep is the character to use for commas, defaulting to #\,.

decimal-sep is the character to use for decimals, defaulting to #\., or to #\, (European style) if comma-sep is already #\..

These parameters may seem unwieldy, but they can also take their defaults from state variables, described below, if any are omitted.

(numeric/comma num [comma-rule radix precision sign-rule])

Shortcut for numeric to print with commas.

(show #f (numeric/comma 123456789))
=> "123,456,789"

(show #f (numeric/comma 123456789 2))
=> "1,23,45,67,89"

(show #f (numeric/comma 123456789 '(3 2)))
=> "12,34,56,789"
(numeric/si num [base separator])

Abbreviates num with an SI suffix as in the -h or --si option to many GNU commands. The base defaults to 1000, using suffix names k, M, G, etc. If the base is 1024, the suffixes are Ki, Mi, Gi, etc. (note the capital "Ki" in this case). It is an error to specify a base other than 1000 or 1024. If separator is provided, it is inserted after the number, before any suffix, for example to allow a space.

(show #f (numeric/si 608))
=> "608"
(show #f (numeric/si 608) "B")
=> "608B"
(show #f (numeric/si 608 1000 " ") "B")
=> "608 B"
(show #f (numeric/si 3986))
=> "4k"
(show #f (numeric/si 3986 1024) "B")
=> "3.9KiB"
(show #f (numeric/si 1.23e-6) "m")
=> "1.2µm"
(show #f (numeric/si 1.23e-6 1000 " ") "m")
=> "1.2 µm"

See https://en.wikipedia.org/wiki/Metric_prefix for the complete list of abbreviations.

(numeric/fitted width n . args)

Like numeric, but if the result doesn't fit in width using the current precision, output instead a string of hashes rather than showing an incorrectly truncated number. For example

(show #f (with ((precision 2)) (numeric/fitted 4 1.25)))
=> "1.25"
(show #f (with ((precision 2)) (numeric/fitted 4 12.345)))
=> "#.##"
(show #f (with ((precision 0)) (numeric/fitted 2 123.45)))
=> "##"

Formatting Space


Outputs a newline.

(show #f nl)
=> "\n"

Short for "fresh line," outputs a newline only if we're not already at the start of a line.

(show #f fl)
=> ""
(show #f "hi" fl)
=> "hi\n"
(show #f "hi" nl fl)
=> "hi\n"
(space-to column)

Outputs spaces up to the given column. If the current column is already >= column, does nothing. The character used for spacing is the current value of pad-char, described below, which defaults to space. Columns are zero-based.

(show #f "a" (space-to 5) "b")
=> "a    b"
(show #f "a" (space-to 0) "b")
=> "ab"
(tab-to [tab-width])

Outputs spaces up to the next tab stop, using tab stops of width tab-width, which defaults to 8. If already on a tab stop, does nothing. If you want to ensure you always tab at least one space, you can use (each " " (tab-to width)). Columns are zero-based.

(show #f (tab-to 5) "b")
=> "b"
(show #f "a" (tab-to 5) "b")
=> "a    b"
(show #f "abcdefghi" (tab-to 5) "b")
=> "abcdefghi b"

Outputs nothing (useful in combinators and as a default noop in conditionals).

(show #f "a" nothing "b")
=> "ab"


(each fmt ...)

Applies each fmt in sequence, as in the top-level of show.

(show #f (each "a" "b"))
=> "ab"
(each-in-list list-of-fmts)

Equivalent to (apply each list-of-fmts) but may be more efficient.

(joined mapper list [sep])

Formats each element elt of list with (mapper elt), inserting sep in between. sep defaults to the empty string, but can be any format or string.

(show #f (joined displayed '(a b c) ", "))
=> "a, b, c"
(joined/prefix mapper list [sep])
(joined/suffix mapper list [sep])
(show #f (joined/prefix displayed '(usr local bin) "/"))
=> "/usr/local/bin"
(show #f (joined/suffix displayed '(1 2 3) nl))
=> "1\n2\n3\n"
Like joined, but inserts sep before/after every element.
(joined/last mapper last-mapper list [sep])

Like joined, but the last element of the list is formatted with last-mapper instead.

(show #f (joined/last displayed
                      (lambda (last) (each "and " last))
                      '(lions tigers bears)
                      ", "))
=> "lions, tigers, and bears"
(joined/dot mapper dot-mapper list [sep])

Like joined, but if the list is a dotted list, then formats the dotted value with dot-mapper instead.

(show #f
      (joined/dot displayed
		  (lambda (dot) (each ". " dot))
		  '(1 2 . 3)
		  " ")
=> "(1 2 . 3)"
(joined/range mapper start [end sep])

Like joined, but counts from start (inclusive) to end (exclusive), formatting each integer in the range with mapper. If end is #f or unspecified, produces an infinite stream of output.

(show #f (joined/range displayed 0 5 " "))
=> "0 1 2 3 4"

Padding and Trimming

Formatters for ensuring output expands to, does not exceed, or fits exactly within a specified width. Width is measured with the string-width state variable, and trimming is then done by calling substring/width state variable on the desired left and right widths. The default values for these are string-length and substring, indicating the widths are are equivalent to string indexes, however this may have different semantics as in terminal-aware discussed below, and other extensions could be imagined such as enforcing trimming on word boundaries.
(padded width fmt ...)
(padded/right width fmt ...)
(padded/both width fmt ...)

Analogs of SRFI-13 string-pad, these add extra space to the left, right or both sides of the output generated by the fmts to pad it to width. If width is exceeded, has no effect. padded/both will include one more extra space on the right side of the output if the difference is odd.

padded/right is guaranteed not to accumulate any intermediate data.

The padding can be controlled with the pad-char state variable described below, defaulting to space.

Note these are column-oriented padders, so won't necessarily work with multi-line output (padding doesn't seem a likely operation for multi-line output).

(show #f (padded 5 "abc"))
=> "  abc"
(show #f (padded/right 5 "abc"))
=> "abc  "
(show #f (padded/both 5 "abc"))
=> " abc "
(trimmed width fmt ...)
(trimmed/right width fmt ...)
(trimmed/both width fmt ...)

Analogs of SRFI-13 string-trim, truncates the output of the fmts to force it in under width columns. trimmed truncates on the left, trimmed/right on the right, and trimmed/both truncates on both the left and right, truncating 1 more on the right if the width isn't even. If width is not exceeded, is equivalent to each.

If a truncation ellipsis is set, then when any truncation occurs, trimmed and trimmed/right will prepend and append the ellipsis, respectively. trimmed/both will both prepend and append. The length of the ellipsis will be considered when truncating the original string, so that the total width will never be longer than width. It is an error if width is less than the length of ellipsis, or double the length for /both.

If the state variable substring/preserve is not #f, then this is called on the left and/or right portions of the output which have been excluded by trim, and the result of this is output in place of the trimmed text. The default is #f (do nothing), but it can be useful to override this to preserve control sequences which are zero-width regardless but affect the state of the output stream. In particular substring-terminal/preserve as enabled in terminal-aware preserves ANSI control sequences and bidirectional overrides.

For example, consider

(show #f (with ((ellipsis "…")) (trimmed/both 5 "abcdef")))
=>  "…bcd…"
Here we note that the output width of 6 exceeds the requested width of 5 and trimming must be done. Since the ellipsis width is 1, we must split the trimming to remove one character from the left and two from the right. Thus, the output is computed as:
(let ((str "abcdef"))
  (each (substring/preserve (substring/width str -1 1))
        (substring/width str 1 4)
        (substring/preserve (substring/width str 4 6))))

Additional examples:

(show #f (trimmed 5 "abcde"))
=>  "abcde"
(show #f (trimmed 5 "abcdef"))
=>  "bcdef"
(show #f (trimmed/right 5 "abcdef"))
=>  "abcde"
(show #f (trimmed/both 5 "abcdef"))
=>  "abcde"
(show #f (trimmed/both 4 "abcdef"))
=>  "bcde"
(show #f (with ((ellipsis "...")) (trimmed 5 "abcdef")))
=>  "...ef"
(show #f (with ((ellipsis "...")) (trimmed/right 5 "abcdef")))
=>  "ab..."
(show #f (with ((ellipsis "_")) (trimmed/both 5 "abcdef")))
=>  "_bcd_"
(show #f (trimmed/right 2 "日本語"))
=>  "日本"
(show #f (terminal-aware (trimmed/right 2 "日本語")))
=>  "日"
(trimmed/lazy width fmt ...)

A variant of trimmed which generates each fmt in left-to-right order, and truncates and terminates immediately if more than width characters are generated. It does not output ellipsis. Thus this is safe to use with an infinite amount of output, e.g. from written-simply on an infinite list.

(fitted width fmt ...)
(fitted/right width fmt ...)
(fitted/both width fmt ...)

A combination of padded and trimmed, ensures the output width is exactly width, truncating if it goes over and padding if it goes under.

Pretty Printing

The homoiconic nature of Lisp makes pretty printing an indispensable utility for presenting code in a readable format.
(pretty obj)

Pretty-prints obj. The result should be identical to written except possibly for differences in whitespace to make the output resemble formatted source code. Implementations should print vectors and data lists (lists that don't begin with a (nested) symbol) in a tabular format when possible to reduce vertical space. As with written, cyclic structure must be detected and represented with datum labels.

(pretty-shared obj)

The same as pretty but using data labels for shared structures among all pairs and vectors, analogous to write-shared.

(pretty-simply obj)

The same as pretty, but without using any datum labels.

(pretty-with-color obj)

Equivalent to pretty, but may optionally include ANSI control sequences (as in Formatting with Color below) to provide syntax highlighting. In such a case, the raw output may not be directly parseable with read.

Columnar Formatting

The following procedures are provided in the (srfi 166 columnar) library.

Although tab-to, space-to and padding/trimming can be used to manually align columns to produce table-like output, these can be tedious to use. The optional extensions in this section make this easier.

(columnar column ...)

Formats each column side-by-side, i.e. as though each were formatted separately and then the individual lines concatenated together. The current line width (from the width state variable) is divided evenly among the columns (setting their state variables accordingly), and all but the last column are right-padded. For example

(show #t (columnar (displayed "abc\ndef\n")
                   (displayed "123\n456\n")))
abc     123
def     456
assuming a 16-char width (the left side gets half the width, or 8 spaces, and is left aligned). Note that we explicitly use displayed instead of the strings directly. This is because columnar treats raw strings as literals inserted into the given location on every line, to be used as borders, for example:
(show #t (columnar "/* " (displayed "abc\ndef\n")
                   " | " (displayed "123\n456\n")
                   " */"))
would output
/* abc | 123 */
/* def | 456 */
Padding ensures alignment only under the assumption that no columns are wider than their allocated width. You can use wrapping or trimming to enforce the underlying width.

You may also prefix any column with any of the symbols 'left, 'right or 'center to control the justification. The symbol 'infinite can be used to indicate the column generates an infinite stream of output.

You can further prefix any column with a width modifier. Any positive integer is treated as a fixed width, ignoring the available width. Any real number between 0 and 1 indicates a fraction of the available width (after subtracting out any fixed widths). Columns with unspecified width divide up the remaining width evenly. If the extra space does not divide evenly, it is allocated column-wise left to right, e.g. if the width of 78 is divided among 5 columns, the column widths become 16, 16, 16, 15, 15 in order. Note that if explicit widths are used, columnar may not take up the full available width.

The value of the col state variable is reset to 0 at the start of each line for each formatter (i.e. they each format as though they were the only column).

Note that columnar builds its output incrementally, interleaving calls to the column formatters until each has produced a line, then concatenating that line together and outputting it. When a formatter has been exhausted, it contributes empty lines until all non-infinite columns are exhausted, at which point the output is complete. This is important because as noted above, some columns may produce an infinite stream of output, and in general you may want to format data larger than can fit into memory. Thus columnar would be suitable for line numbering a file of arbitrary size, or implementing the Unix yes(1) command, etc.

The degenerate case of no columns produces a single blank line.

(tabular column ...)

Equivalent to columnar except that each column is padded at least to the minimum width required on any of its lines. Thus

(show #t (tabular "|" (each "a\nbc\ndef\n") "|"
                      (each "123\n45\n6\n") "|"))
|a  |123|
|bc |45 |
|def|6  |
This makes it easier to generate tables without knowing widths in advance. However, because it requires generating the entire output in advance to determine the correct column widths, tabular cannot format a table larger than would fit in memory.

Note that since tabular computes explicit widths for all columns, it will use the most compact width for unspecified columns and not necessarily consume the full available width.

(wrapped fmt ...)

Behaves like each, except text is accumulated and lines are wrapped to fit in the current width as in the Unix fmt(1) command. Specifically, words are tokenized by splitting on all characters which satisfy the predicate in the parameter word-separator?, which defaults to char-whitespace?. Words are grouped into lines separating them by space, and line breaks are introduced to minimize the sum of the cube of trailing whitespace on every line while ensuring no line exceeds width (as measured with the string-width state variable).

The last line is not appended with a newline, so that in the trivial case of a single line this is equivalent to each (but reducing whitespace).

(wrapped/list list-of-strings)

Like wrapped, but taking a pre-tokenized list of strings.

(wrapped/char fmt ...)

Like wrapped, but splits simply on individual characters as the current width is reached on each line. Thus there is nothing to optimize and this formatter doesn't buffer output, as we only need look ahead one character at a time to check its width.

(justified <format> ...)

Like wrapped except the lines are full-justified.

(define func
  '(define (fold kons knil ls)
     (let lp ((ls ls) (acc knil))
       (if (null? ls) acc (lp (cdr ls) (kons (car ls) acc))))))

(define doc
    "The fundamental list iterator.  Applies KONS to each "
    "element of LS and the result of the previous application, "
    "beginning with KNIL.  With KONS as CONS and KNIL as '(), "
    "equivalent to REVERSE."))

(show #t (columnar (pretty func) " ; " (justified doc)))
(define (fold kons knil ls)          ; The   fundamental   list   iterator.
  (let lp ((ls ls) (acc knil))       ; Applies  KONS  to  each  element  of
    (if (null? ls)                   ; LS  and  the  result of the previous
        acc                          ; application,  beginning  with  KNIL.
        (lp (cdr ls)                 ; With  KONS  as CONS and KNIL as '(),
            (kons (car ls) acc)))))  ; equivalent to REVERSE.
(from-file pathname)

Displays the contents of the file pathname one line at a time, so that in typical formatters such as columnar only constant memory is consumed, making this suitable for formatting files of arbitrary size.

(line-numbers [start])

A convenience utility, just formats an infinite stream of numbers (in the current radix) beginning with start, which defaults to 1.

The Unix nl(1) utility could be implemented as:

(show #t (columnar 4 'right 'infinite (line-numbers)
                   " " (from-file "read-line.scm")))
which might output:
   2 (define (read-line . o)
   3   (let ((port (if (pair? o) (car o) (current-input-port))))
   4     (let lp ((res '()))
   5       (let ((c (read-char port)))
   6         (if (or (eof-object? c) (eqv? c #\newline))
   7             (list->string (reverse res))
   8             (lp (cons c res)))))))

Formatting with Color

The following procedures are provided in the (srfi 166 color) library.

(as-red fmt ...)
(as-blue fmt ...)
(as-green fmt ...)
(as-cyan fmt ...)
(as-yellow fmt ...)
(as-magenta fmt ...)
(as-white fmt ...)
(as-black fmt ...)
(as-bold fmt ...)
(as-italic fmt ...)
(as-underline fmt ...)

Outputs the formatters fmt ... colored or (boldened, italicized or underline) with ANSI control sequences, for use when formatting to a terminal.

(on-red fmt ...)
(on-blue fmt ...)
(on-green fmt ...)
(on-cyan fmt ...)
(on-yellow fmt ...)
(on-magenta fmt ...)
(on-white fmt ...)
(on-black fmt ...)

Outputs the formatters fmt ... with ANSI control sequences to set the background color, for use when formatting to a terminal.

(as-color red green blue fmt ...)

Each of red, green, blue should be an exact integer in the range [0, 5], representing the corresponding components of an RGB color model. Outputs the formatters colored accordingly using 8-bit color ANSI control sequences.

(as-true-color red green blue fmt ...)

The 24-bit True Color equivalent of as-color, taking a range of [0, 255] for each component, for use with terminals supporting this.

(on-color red green blue fmt ...)

The equivalent of as-color, setting the background color.

(on-true-color red green blue fmt ...)

The equivalent of as-true-color, setting the background color.

It is up to the caller to ensure that the terminals support these ANSI control sequences.


The following procedures are provided in the (srfi 166 unicode) library.

(terminal-aware fmt ...)

Equivalent to

    (fn (ambiguous-is-wide?)
      (with ((string-width (if ambiguous-is-wide?
             (substring/width (if ambiguous-is-wide?
             (substring/preserve substring-terminal/preserve))
        fmt ...))

Padding, trimming and tabbing, etc. will generally not do the right thing in the presence of zero-width and double-width Unicode characters, or ANSI control sequences. This formatter overrides the string-width and substring/width state variables used in column tracking to do the right thing in such cases, considering Unicode double or full width characters as 2 characters wide (as they typically are in fixed-width terminals), while treating combining and non-spacing characters as 0 characters wide.

;; 3 characters padded to 5
(show #f (with ((pad-char #\〜)) (padded/both 5 "日本語")))
=> "〜日本語〜"

;; the 3 characters have a terminal width of 6 so are not padded
(show #f (terminal-aware (with ((pad-char #\〜)) (padded/both 5 "日本語"))))
=> "日本語"
(string-terminal-width str)

A utility function which returns the integer number of columns str would require in a terminal, according to the following rules:

  1. non-spacing characters (format control characters with the property Cf, or non-spacing marks with the property Mn) count as 0 columns
  2. characters with the East Asian Wide (W) or East Asian Fullwidth (F) properties, according to Unicode TR #11, count as 2 columns
  3. characters with the Halfwidth (H) or Narrow (Na) property should count as 1 column
  4. characters with the Neutral (N) non-East Asian property also count as 1 column
  5. characters with the Ambiguous (A) property are 1 columns
  6. ANSI terminal control sequences, as output by the color formatters above, count as 0 columns
  7. the tab character is implementation defined

Implementations should support the properties from at least the current Unicode specification at the time of writing of this SRFI, 12.0.0.

(string-terminal-width/wide str [start end])

Equivalent to string-terminal-width, except that ambiguous characters are counted as 2 columns, as they are in certain Japanese environments (notably kterm, fonts such as MS Gothic and the system fonts of many Japanese feature phones).

(substring-terminal-width str from to)

Returns the substring of str, starting from the first index where the total string width exceeds from width, inclusive, to the first index exceeding to width, exclusive, using the notion of width as defined in string-terminal-width. Note this naturally groups grapheme clusters together by excluding leading modifiers and including trailing modifiers. If str includes only single-width characters, this definition is equivalent to substring. A negative start can be used to effectively include all leading zero-width characters.

(substring-terminal-width "abc" 0 6)
=> "abc"
(substring-terminal-width "abc" 0 4)
=> "ab"
(substring-terminal-width "abc" 2 6)
=> "bc"
(substring-terminal-width "abc" 1 4)
=> "ab"
(substring-terminal-width "abc" 1 5)
=> "ab"
(substring-terminal-width "abc" 2 4)
=> "b"
(substring-terminal-width "abc" 2 3)
=> ""
(substring-terminal-width "abc" -1 2)
=> "a"
(substring-terminal-width/wide str from to)

Equivalent to substring-terminal-width, except that ambiguous characters are counted as 2 columns

(substring-terminal-preserve str)

Returns only the substring sequences of str which would have non-local implications for rendering the text in a terminal. Specifically, preserves ANSI color control sequences, as well as the directional formatting characters described in the Unicode Bidirectional Algorithm.

(upcased fmt ...)
(downcased fmt ...)

Runs the formatters fmt ..., but with all output translated as if first passed to string-upcase or string-downcase, respectively. Note these should also work correctly when combined with the ANSI control sequences from formatting with color, which includes ASCII letters in the control sequences.

Note there should be no internal buffering, which may have an effect on context-sensitive casing. For example, if an implementation correctly supports "ς" as the proper downcased form of a Greek sigma "Σ" at the end of a word, it may assume that the end of a string is in fact the end of a word, even if later succeeded by a word constituent character.

(show #f (upcased "abc"))
=> "ABC"

(show #f (downcased "ΜΈΛΟΣ"))
=> "μέλος"

(show #f (downcased "ΜΈΛΟΣ" "Μ"))
=> unspecified

Higher Order Formatters and State

Formatters up to this point have been simple accumulators of output, with no control flow or handling of state. Both of these are provided by fn and with for getting and setting state, respectively.

(fn ((id state-var) ...) expr ... fmt)

Short for "function," this is the analog to lambda. Returns a formatter which on application evaluates each expr and fmt in left-to-right order, in a lexical environment extended with each identifier id bound to the current value of the state variable evaluated by state-var. The result of the fmt is then applied as a formatter.

As a convenience, any (id state-var) list may be abbreviated as simply id, indicating id is bound to the state variable of the same identifier. Note this would then shadow the state variable in any nested functions.

(show #f "column: " (fn (col) col))
=> "column: 8"

(show #f "column: " (fn ((col1 col))
                     (each col1 ", " (fn ((col2 col)) col2))))
=> "column: 8, 11"

The trivial case of no state variables is often useful to allow for lazy applications of formatters, needed for conditional formatting and loops. For example:

(show #t (let lp ((ls ls))
           (if (pair? ls)
               (each (car ls) (lp (cdr ls)))
would eagerly create a formatter concatenating every element of ls before starting to accumulate any output, whereas
(show #t (let lp ((ls ls))
           (if (pair? ls)
               (each (car ls) (fn () (lp (cdr ls))))
would lazily apply the formatters one at a time.
(with ((state-var value) ...) fmt ...)

Conceptually the formatting equivalent of parameterize, temporarily altering state variables. Applies each of the formatters fmt with each state-var bound to the corresponding value. The resulting state is then updated to restore each state-var to its original value.

(with! (state-var value) ...)

Similar to with but does not restore the original values, changing the value of each state-var for any remaining formatters in a sequence.

As the current formatting state can be captured or reentered with continuations, with! should be used with caution, and may produce unexpected output in some cases.

(forked fmt1 fmt2)

Calls fmt1 on (a conceptual copy of) the current state, then fmt2 on the same original state as though fmt1 had not been called.

(call-with-output formatter mapper)

A utility, calls formatter on a copy of the current state (as with forked), accumulating the results into a string. Then calls the formatter resulting from (mapper result-string) on the original state.

State Variables

(make-state-variable name default [immutable])

Returns a new state variable suitable for use in fn and with, etc. The name should be a string and is strictly for debugging purposes. default is the default value when referenced in fn if the value has not be set. If immutable is true, the state variable can only be dynamically bound with with, and not set with with!.

The following state variables have predefined meanings with the formatters in this SRFI. These are all exported by the (srfi 166 base) library.


The textual port output is written to. This can be overridden to capture intermediate output. If any output is made to port by parallel computations and/or side-effecting Scheme procedures during the dynamic extent of a call to show, then the values of row and col are unspecified.


The current row of output, starting at 0 regardless of what may previously have been written to port.


The current column of output, used for padding and spacing, etc., starting at 0 regardless of what may previously have been written to port.


The current line width, used for columnar, wrapping and pretty-printing. The default is implementation-defined.


The underlying standard formatter for writing a single string. The default value outputs the string to port while tracking the current row and col. This can be overridden both to capture intermediate output and perform transformations on strings before outputting, but should generally wrap the existing output to preserve expected behavior. You should not write to port except via output. The default output procedure is exported as output-default.


The mapper for automatic formatting of non-string/char values in top-level show, each and other formatters. The default value is implementation-defined, but should format in sexp notation. One could override this to format other programming languages.


A procedure taking three args: (string [start end]), where start and end are optional, defaulting to the 0 and (string-length string) respectively. Returns the length in columns of that string within the given range. The default treats each character as a width of 1, returning (- end start).


A procedure taking three args: (string from to), which returns the substring of string whose width is between from and to. The default value is substring, where from and to correspond directly to indexes. This should generally be updated in conjunction with string-width.


A procedure taking three args with the same semantics as substring/width: (string from to), which returns any control characters or sequences which have non-local implications and thus should not be removed by a trimmed operation. The default value is #f, indicating nothing needs to be preserved, but can be overriden as in substring-terminal/preserve.


The character used by space-to, tab-to and other padding formatters.

(define (print-table-of-contents alist)
  (define (print-line x)
    (each (car x) (space-to 72) (padded 3 (cdr x))))
  (show #t (with ((pad-char #\.))
             (joined/suffix print-line alist nl))))

 '(("An Unexpected Party" . 29)
   ("Roast Mutton" . 60)
   ("A Short Rest" . 87)
   ("Over Hill and Under Hill" . 100)
   ("Riddles in the Dark" . 115)))
would output
An Unexpected Party.....................................................29
Roast Mutton............................................................60
A Short Rest............................................................87
Over Hill and Under Hill...............................................100
Riddles in the Dark....................................................115

The string used when truncating as described in trimmed, default the empty string.


The radix for numeric output, defaulting to 10, as used in numeric and written.


The precision for numeric output, as described in numeric and written. The precision specifies the number of digits written after the decimal point. If the numeric value to be written out requires more digits to represent it than precision, the written representation is chosen which is closest to the numeric value and representable with the specified precision. If the numeric value falls on the midpoint of two such representations, it is implementation-dependent which representation is chosen.

When the numeric value is an inexact floating-point number, there is more than one interpretation of this "rounding". One is to take the effective value the floating-point number represents (e.g. if we use binary floating-point numbers, we take the value of (* sign mantissa (expt 2 exponent))), and compare it to the two closest numeric representations of the given precision. Another way is to obtain the default notation of the floating-point number and apply rounding to it. The former (we call it effective rounding) is consistent with most floating-point number operations, but may lead to a non-intuitive result than the latter (we call it notational rounding). For example, 5.015 can't be represented exactly in binary floating-point numbers. With IEEE754 floating-point numbers, the floating point number closest to 5.015 is smaller than exact 5.015, i.e. (< 5.015 5015/1000) => #t. With effective rounding with precision 2, it should result in "5.01". However, users who look at the notation may be confused by "5.015" not being rounded up as they usually expect. With notational rounding the implementation chooses "5.02" (if it also adopts round-half-to-infinity or round-half-up rule). It is up to the implementation to choose which interpretation to adopt.


The decimal separator for floating point output, default ".".


Specifies an alignment for the decimal place when formatting numbers, useful for outputting tables of numbers.

(define (print-angles x)
  (joined numeric (list x (sin x) (cos x) (tan x)) " "))

(show #t (with ((decimal-align 5) (precision 3))
           (joined/suffix print-angles (iota 5) nl)))
would output
 0.000    0.000    1.000    0.000
 1.000    0.842    0.540    1.557
 2.000    0.909   -0.416   -2.185
 3.000    0.141   -0.990   -0.142
 4.000   -0.757   -0.654    1.158

Additional vars used for formatting as described in formatting numbers.


A character predicate used to tokenize words for wrapped and justify. Defaults to char-whitespace?. More flexibility is available with wrapped/list.


Use to choose between string-terminal-width and string-terminal-width/wide when formatting with terminal-aware. The default value is implementation-defined. A reasonable approach might be to check if the TERM environment variable is kterm when writing to a terminal. One could also check if the recipient of an email were using a .jp email address, however frequent users of the ambiguous characters in such environments are likely to have changed their fonts.


An environment which may optionally be used for hints in the pretty printing formatters, defaulting to (interaction-environment).


A sample implementation in portable R7RS will be available at https://github.com/ashinn/chibi-scheme/blob/master/lib/srfi/166.sld, and included files, depending on SRFI 1, 69, 117, 130, 165. It is mostly the same as in SRFI 159. Two alternative implementations are also available, one by Adam Nelson at https://github.com/ar-nelson/schemepunk/tree/show, and one by Marc Nieper-Wißkirchen at https://gitlab.com/nieper/show.

Note columnar and trimmed/lazy rely on first-class continuations, however an implementation written in CPS-style would not require this.


The author would like to thank everyone who provided feedback for SRFI 159 and SRFI 166, in particular Marc Nieper-Wißkirchen for detailed feedback and bug reports and work on an alternate implementation, Adam Nelson for his implementation, Jim Rees who provided many bug fixes early on, John Cowan for early editorial comments, and Arthur Gleckler for his hard work and super-human patience in waiting for me to get my act together.


      Alex Shinn, John Cowan, Arthur Gleckler, Revised^7 Report on the Algorithmic Language Scheme
      Guy L. Steele Jr., Common Lisp Hyperspec
      Scott G. Miller, SRFI 28 - Basic Format Strings
      Ken Dickey, SRFI 48 - Intermediate Format Strings
      Alex Shinn, SRFI 159 - Combinator Formatting
      C++ iomanip
      Damian Conway, Perl6 Exegesis 7 - formatting
      Alex Shinn, fmt - Combinator Formatting
      Mark Davis et al., Unicode® Standard Annex #9 - Unicode Bidirectional Algorithm
      Ken Lunde, Unicode® Standard Annex #11 - East Asian Width


Copyright (C) Alex Shinn 2020. All Rights Reserved.

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice (including the next paragraph) shall be included in all copies or substantial portions of the Software.


Editor: Arthur A. Gleckler