207: String-notated bytevectors

by Daphne Preston-Kendal (external notation), John Cowan (procedure design), Wolfgang Corcoran-Mathe (implementation)


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-207@nospamsrfi.schemers.org. To subscribe to the list, follow these instructions. You can access previous messages via the mailing list archive.


To ease the human reading and writing of Scheme code involving binary data that for mnemonic reasons corresponds as a whole or in part to ASCII-coded text, a notation for bytevectors is defined which allows printable ASCII characters to be used literally without being converted to their corresponding integer forms. In addition, this SRFI provides a set of procedures known as the bytestring library for constructing a bytevector from a sequence of integers, characters, strings, and/or bytevectors, and for manipulating bytevectors as if they were strings as far as possible.


Binary file formats are usually not self-describing, and if they are, the descriptive portion is itself binary, which makes it hard for human beings to interpret. To assist with this problem, it is common to have a human-readable section at the beginning of the file, or in some cases at the beginning of each distinct section of the file. For historical reasons and to avoid text encoding complications, it is usual for this human-readable section to be expressed as ASCII text.

For example, ZIP files begin with the hex bytes 50 4B which are the ASCII encoding for the characters "PK", the initials of Phil Katz, the inventor of ZIP format. As another example, the GIF image format begins with 47 49 46 38 39 61, the ASCII encoding for "GIF89a", where "89a" is the format version. A third example is the PNG image format, where the file header begins 89 50 4E 47. The first byte is intentionally non-ASCII, but the next three are "PNG". Furthermore, a PNG file is divided into chunks, each of which contains a 4-byte "chunk type" code. The letters in the chunk type are mnemonics for its purpose, such as "PLTE" for a palette, "bKGD" for a default background color, and "iTXt" for descriptive text in UTF-8.

When bytevectors contain string data of this kind, it is much more tractable for human programmers to deal with them in the form #u8"recursion" than in the form #u8(114 101 99 117 114 115 105 111 110). This is true even when non-ASCII bytes are incorporated into the bytevector: the complete 8-bit PNG file header can be written as #u8"\x89;PNG\r\n\x1A;\n" instead of #u8(0x89 0x50 0x4E 0x47 0x0D 0x0A 0x1A 0x0A).

In addition, this SRFI provides bytevectors with additional procedures that closely resemble those provided for strings. For example, bytevectors can be padded or trimmed, compared case-sensitively or case-insensitively, searched, joined, and split.

In this specification it is assumed that bytevectors are as defined in R7RS-small section 6.9. Implementations may also consider them equivalent to R6RS bytevectors (R6RS 4.3.4) or SRFI 4 u8vectors, depending which kind of homogeneous vectors of unsigned 8-bit integers an implementation supports.


Most of the procedures of this SRFI begin with bytestring- in order to distinguish them from other bytevector procedures. This does not mean that they accept or return a separate bytestring type: bytestrings and bytevectors are exactly the same type.

External notation

The basic form of a string-notated bytevector is:

#u8" content "

To avoid character encoding issues within string-notated bytevectors, only printable ASCII characters (that is, Unicode codepoints in the range from U+0020 to U+007E inclusive) are allowed to be used within the content of a string-notated bytevector. All other characters must be expressed through mnemonic or inline hex escapes, and " and \ must also be escaped as in normal Scheme strings.

Within the content of a string-notated bytevector:

Note: The \| sequence is provided so that string parsing, symbol parsing, and string-notated bytevector parsing can all use the same sequences. However, we give a complete definition of the valid lexical syntax in this SRFI rather than inheriting the native syntax of strings, so that it is clear that #u8"ι" and #u8"\xE000;" are invalid.

When the Scheme reader encounters a string-notated bytevector, it produces a datum as if that bytevector had been written out in full. That is, #u8"A" is exactly equivalent to #u8(65).

A Scheme implementation which supports string-notated bytevectors may not by default use this notation when any of the write family of procedures is called upon a bytevector or upon another datum containing a bytevector. A future SRFI is expected to add a configurable version of the write procedure which may enable the use of this notation in this context.

Formal syntax

The formal syntax of Scheme (defined in R7RS-small 7.1) is amended as follows.


(bytestring arg)

Converts args into a sequence of small integers and returns them as a bytevector as follows:

Otherwise, an error satisfying bytestring-error? is signaled.


(bytestring "lor" #\r #x65 #u8(#x6d)) ⇒ #u8"lorem"
(bytestring "η" #\space #u8(#x65 #x71 #x75 #x69 #x76)) ⇒ error

(make-bytestring list)

If the elements of list are suitable arguments for bytestring, returns the bytevector that would be the result of applying bytestring to list. Otherwise, an error satisfying bytestring-error? is signaled.

(make-bytestring! bytevector at list)

If the elements of list are suitable arguments for bytestring, writes the bytes of the bytevector that would be the result of calling make-bytestring into bytevector starting at index at.

(define bstring (make-bytevector 10 #x20))
(make-bytestring! bstring 2 '(#\s #\c "he" #u8(#x6d #x65)))
bstring ⇒ #u8"  scheme  "


(bytevector->hex-string bytevector)
(hex-string->bytevector string)

Converts between a bytevector and a string containing pairs of hexadecimal digits. If string is not pairs of hexadecimal digits, an error satisfying bytestring-error? is raised.

(bytevector->hex-string #u8"Ford") ⇒ "467f7264"
(hex-string->bytevector "5a6170686f64") ⇒ #u8"Zaphod"

(bytevector->base64 bytevector [digits])
(base64->bytevector string [digits])

Converts between a bytevector and its base-64 encoding as a string. The 64 digits are represented by the characters 0-9, A-Z, a-z, and the symbols + and /. However, there are different variants of base-64 encoding which use different representations of the 62nd and 63rd digit. If the optional argument digits (a two-character string) is provided, those two characters will be used as the 62nd and 63rd digit instead. Details can be found in RFC 4648. If string is not in base-64 format, an error satisfying bytestring-error? is raised. However, characters that satisfy char-whitespace? are silently ignored.

(bytevector->base64 #u8(1 2 3 4 5 6)) ⇒ "AQIDBAUG"
(bytevector->base64 #u8"Arthur Dent") ⇒ "QXJ0aHVyIERlbnQ="
(base64->bytevector "+/     /+") ⇒ #u8(#xfb #xff #xfe)

(bytestring->list bytevector [ start [ end ] ])

Converts all or part of a bytevector into a list of the same length containing characters for elements in the range 32 to 127 and exact integers for all other elements.

(bytestring->list #u8(#x41 #x42 1 2) 1 3) ⇒ (#\B 1)

(make-bytestring-generator arg)

Returns a generator that when invoked will return consecutive bytes of the bytevector that bytestring would create when applied to args, but without creating any bytevectors. The args are validated before any bytes are generated; if they are ill-formed, an error satisfying bytestring-error? is raised.

(generator->list (make-bytestring-generator "lorem"))
  ⇒ (#x6c #x6f #x72 #x65 #x6d)


(bytestring-pad bytevector len char-or-u8)
(bytestring-pad-right bytevector len char-or-u8)

Returns a newly allocated bytevector with the contents of bytevector plus sufficient additional bytes at the beginning/end containing char-or-u8 (which can be either an ASCII character or an exact integer in the range 0-255) such that the length of the result is at least len.

(bytestring-pad #u8"Zaphod" 10 #\_) ⇒ #u8"____Zaphod"
(bytestring-pad-right #u8(#x80 #x7f) 8 0) ⇒ #u8(#x80 #x7f 0 0 0 0 0 0)

(bytestring-trim bytevector pred)
(bytestring-trim-right bytevector pred)
(bytestring-trim-both bytevector pred)

Returns a newly allocated bytevector with the contents of bytevector, except that consecutive bytes at the beginning / the end / both the beginning and the end that satisfy pred are not included.

(bytestring-trim #u8"   Trillian" (lambda (b) (= b #x20)))
  ⇒ #u8"Trillian"
(bytestring-trim-both #u8(0 0 #x80 #x7f 0 0 0) zero?) ⇒ #u8(#x80 #x7f)


(bytestring-replace bytevector1 bytevector2 start1 end1 [start2 end2])

Returns a newly allocated bytevector with the contents of bytevector1, except that the bytes indexed by start1 and end1 are not included but are replaced by the bytes of bytevector2 indexed by start2 and end2.

(bytestring-replace #u8"Vogon torture" #u8"poetry" 6 13)
  ⇒ #u8"Vogon poetry"


To compare bytevectors for equality, use the procedure bytevector=? from either the R6RS library (rnrs bytevectors) or the equivalent R7RS library (scheme bytevector).

(bytestring<? bytevector1 bytevector2)
(bytestring>? bytevector1 bytevector2)
(bytestring<=? bytevector1 bytevector2)
(bytestring>=? bytevector1 bytevector2)

Returns #t if bytevector1 is less than / greater than / less than or equal to / greater than or equal to bytevector2. Comparisons are lexicographical: shorter bytevectors compare before longer ones, all elements being equal.

(bytestring<? #u8"Heart Of Gold" #u8"Heart of Gold") ⇒ #t
(bytestring<=? #u8(#x81 #x95) #u8(#x80 #xa0)) ⇒ #f
(bytestring>? #u8(1 2 3) #u8(1 2)) ⇒ #t


(bytestring-index bytevector pred [start [end]])
(bytestring-index-right bytevector pred [start [end]])

Searches bytevector from start to end / from end to start for the first byte that satisfies pred, and returns the index into bytevector containing that byte. In either direction, start is inclusive and end is exclusive. If there are no such bytes, returns #f.

(bytestring-index #u8(#x65 #x72 #x83 #x6f) (lambda (b) (> b #x7f))) ⇒ 2
(bytestring-index #u8"Beeblebrox" (lambda (b) (> b #x7f))) ⇒ #f
(bytestring-index-right #u8"Zaphod" odd?) ⇒ 4

(bytestring-break bytevector pred)
(bytestring-span bytevector pred)

Returns two values, a bytevector containing the maximal sequence of characters (searching from the beginning of bytevector to the end) that do not satisfy / do satisfy pred, and another bytevector containing the remaining characters.

(bytestring-break #u8(#x50 #x4b 0 0 #x1 #x5) zero?)
  ⇒ #u8(#x50 #x4b)
    #u8(0 0 #x1 #x5)
(bytestring-span #u8"ABCDefg" (lambda (b) (and (> b 40) (< b 91))))
  ⇒ #u8"ABCD"

Joining and splitting

(bytestring-join bytevector-list delimiter [grammar])

Pastes the bytevectors in bytevector-list together using the delimiter, which can be anything suitable as an argument to bytestring. The grammar argument is a symbol that determines how the delimiter is used, and defaults to infix. It is an error for grammar to be any symbol other than these four:

(bytestring-join '(#u8"Heart" #u8"of" #u8"Gold") #x20) ⇒ #u8"Heart of Gold"
(bytestring-join '(#u8(#xef #xbb) #u8(#xbf)) 0 'prefix) ⇒ #u8(0 #xef #xbb 0 #xbf)
(bytestring-join '() 0 'strict-infix) ⇒ error

(bytestring-split bytevector delimiter [grammar])

Divides the elements of bytevector and returns a list of newly allocated bytevectors using the delimiter (an ASCII character or exact integer in the range 0-255 inclusive). Delimiter bytes are not included in the result bytevectors.

The grammar argument is used to control how bytevector is divided. It has the same default and meaning as in bytestring-join, except that infix and strict-infix mean the same thing. That is, if grammar is prefix or suffix, then ignore any delimiter in the first or last position of bytevector respectively.

(bytestring-split #u8"Beeblebrox" #x62) ⇒ (#u8"Bee" #u8"le" #u8"rox")
(bytestring-split #u8(1 0 2 0) 0 'suffix) ⇒ (#u8(1) #u8(2))


(read-textual-bytestring  prefix [ port ])

Reads a string in the external format described in this SRFI from port and return it as a bytevector. If the prefix argument is false, this procedure assumes that "#u8" has already been read from port. If port is omitted, it defaults to the value of (current-input-port). If the characters read are not in the external format, an error satisfying bytestring-error? is raised.

(call-with-port (open-input-string "#u8\"AB\\xad;\\xf0;\\x0d;CD\"")
                (lambda (port)
                  (read-textual-bytestring #t port)))
  ⇒ #u8(#x41 #x42 #xad #xf0 #x0d #x43 #x44)

(write-textual-bytestring bytevector [ port ])

Writes bytevector in the external format described in this SRFI to port. Bytes representing non-graphical ASCII characters are unencoded: all other bytes are encoded with a single letter if possible, otherwise with a \x escape. If port is omitted, it defaults to the value of (current-output-port).

(call-with-port (open-output-string)
                (lambda (port)
                   #u8(#x9 #x41 #x72 #x74 #x68 #x75 #x72 #xa)
                  (get-output-string port)))
  ⇒ "#u8\"\\tArthur\\n\""

(write-binary-bytestring port arg)

Outputs each arg to the binary output port port using the same interpretations as bytestring, but without creating any bytevectors. The args are validated before any bytes are written to port; if they are ill-formed, an error satisfying bytestring-error? is raised.

(call-with-port (open-output-bytevector)
                (lambda (port)
                  (write-binary-bytestring port #\Z #x61 #x70 "hod")
                  (get-output-bytevector port)))
  ⇒ #u8"Zaphod"


(bytestring-error? obj)

Returns #t if obj is an object signaled by any of the following procedures, in the circumstances described above:


There is a sample implementation of the procedures, but not the notation, in the repository of this SRFI.


Daphne Preston-Kendal devised the string notation for bytevectors; John Cowan, the procedure library; Wolfgang Corcoran-Mathe, the sample implementation of the procedures.

The notation is inspired by the notation used in Python since version 2.6 for bytes objects, which are fundamentally similar in purpose to Scheme bytevectors, especially in R7RS. In addition, many of the procedures are closely analogous to those of SRFI 152.

Thanks is also due to the participants in the SRFI mailing list. In particular: Lassi Kortela corrected an embarrassing technical error; Marc Nieper-Wißkirchen explained why the write procedure ought not to be allowed to use this notation by default.

© 2020 Daphne Preston-Kendal, John Cowan, and Wolfgang Corcoran-Mathe.

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