charsets — character set standards and internationalization
This manual page gives an overview on different character set standards and how they were used on Linux before Unicode became ubiquitous. Some of this information is still helpful for people working with legacy systems and documents.
Standards discussed include such as ASCII, GB 2312, ISO 8859, JIS, KOI8-R, KS, and Unicode.
The primary emphasis is on character sets that were actually used by locale character sets, not the myriad others that could be found in data from other systems.
ASCII (American Standard Code For Information Interchange) is the original 7-bit character set, originally designed for American English. Also known as US-ASCII. It is currently described by the ISO 646:1991 IRV (International Reference Version) standard.
Various ASCII variants replacing the dollar sign with other currency symbols and replacing punctuation with non-English alphabetic characters to cover German, French, Spanish, and others in 7 bits emerged. All are deprecated; glibc does not support locales whose character sets are not true supersets of ASCII.
As Unicode, when using UTF-8, is ASCII-compatible, plain ASCII text still renders properly on modern UTF-8 using systems.
ISO 8859 is a series of 15 8-bit character sets, all of which have ASCII in their low (7-bit) half, invisible control characters in positions 128 to 159, and 96 fixed-width graphics in positions 160-255.
Of these, the most important is ISO 8859-1 ("Latin Alphabet No .1" / Latin-1). It was widely adopted and supported by different systems, and is gradually being replaced with Unicode. The ISO 8859-1 characters are also the first 256 characters of Unicode.
Console support for the other 8859 character sets is available under Linux through user-mode utilities (such as setfont(8)) that modify keyboard bindings and the EGA graphics table and employ the "user mapping" font table in the console driver.
Here are brief descriptions of each set:
Latin-1 covers many West European languages such as Albanian, Basque, Danish, English, Faroese, Galician, Icelandic, Irish, Italian, Norwegian, Portuguese, Spanish, and Swedish. The lack of the ligatures Dutch Ĳ/ĳ, French œ, and old-style „German“ quotation marks was considered tolerable.
Latin-2 supports many Latin-written Central and East European languages such as Bosnian, Croatian, Czech, German, Hungarian, Polish, Slovak, and Slovene. Replacing Romanian ș/ț with ş/ţ was considered tolerable.
Latin-3 was designed to cover of Esperanto, Maltese, and Turkish, but 8859-9 later superseded it for Turkish.
Latin-4 introduced letters for North European languages such as Estonian, Latvian, and Lithuanian, but was superseded by 8859-10 and 8859-13.
Cyrillic letters supporting Bulgarian, Byelorussian, Macedonian, Russian, Serbian, and (almost completely) Ukrainian. It was never widely used, see the discussion of KOI8-R/KOI8-U below.
Was created for Arabic. The 8859-6 glyph table is a fixed font of separate letter forms, but a proper display engine should combine these using the proper initial, medial, and final forms.
Was created for Modern Greek in 1987, updated in 2003.
Supports Modern Hebrew without niqud (punctuation signs). Niqud and full-fledged Biblical Hebrew were outside the scope of this character set.
This is a variant of Latin-1 that replaces Icelandic letters with Turkish ones.
Latin-6 added the Inuit (Greenlandic) and Sami (Lappish) letters that were missing in Latin-4 to cover the entire Nordic area.
Supports the Thai alphabet and is nearly identical to the TIS-620 standard.
This set does not exist.
Supports the Baltic Rim languages; in particular, it includes Latvian characters not found in Latin-4.
This is the Celtic character set, covering Old Irish, Manx, Gaelic, Welsh, Cornish, and Breton.
Latin-9 is similar to the widely used Latin-1 but replaces some less common symbols with the Euro sign and French and Finnish letters that were missing in Latin-1.
This set covers many Southeast European languages, and most importantly supports Romanian more completely than Latin-2.
KOI8-R is a non-ISO character set popular in Russia before Unicode. The lower half is ASCII; the upper is a Cyrillic character set somewhat better designed than ISO 8859-5. KOI8-U, based on KOI8-R, has better support for Ukrainian. Neither of these sets are ISO-2022 compatible, unlike the ISO 8859 series.
Console support for KOI8-R is available under Linux through user-mode utilities that modify keyboard bindings and the EGA graphics table, and employ the "user mapping" font table in the console driver.
GB 2312 is a mainland Chinese national standard character set used to express simplified Chinese. Just like JIS X 0208, characters are mapped into a 94x94 two-byte matrix used to construct EUC-CN. EUC-CN is the most important encoding for Linux and includes ASCII and GB 2312. Note that EUC-CN is often called as GB, GB 2312, or CN-GB.
Big5 was a popular character set in Taiwan to express traditional Chinese. (Big5 is both a character set and an encoding.) It is a superset of ASCII. Non-ASCII characters are expressed in two bytes. Bytes 0xa1-0xfe are used as leading bytes for two-byte characters. Big5 and its extension were widely used in Taiwan and Hong Kong. It is not ISO 2022 compliant.
JIS X 0208 is a Japanese national standard character set. Though there are some more Japanese national standard character sets (like JIS X 0201, JIS X 0212, and JIS X 0213), this is the most important one. Characters are mapped into a 94x94 two-byte matrix, whose each byte is in the range 0x21-0x7e. Note that JIS X 0208 is a character set, not an encoding. This means that JIS X 0208 itself is not used for expressing text data. JIS X 0208 is used as a component to construct encodings such as EUC-JP, Shift_JIS, and ISO-2022-JP. EUC-JP is the most important encoding for Linux and includes ASCII and JIS X 0208. In EUC-JP, JIS X 0208 characters are expressed in two bytes, each of which is the JIS X 0208 code plus 0x80.
KS X 1001 is a Korean national standard character set. Just as JIS X 0208, characters are mapped into a 94x94 two-byte matrix. KS X 1001 is used like JIS X 0208, as a component to construct encodings such as EUC-KR, Johab, and ISO-2022-KR. EUC-KR is the most important encoding for Linux and includes ASCII and KS X 1001. KS C 5601 is an older name for KS X 1001.
The ISO 2022 and 4873 standards describe a font-control model based on VT100 practice. This model is (partially) supported by the Linux kernel and by xterm(1). Several ISO 2022-based character encodings have been defined, especially for Japanese.
There are 4 graphic character sets, called G0, G1, G2, and G3, and one of them is the current character set for codes with high bit zero (initially G0), and one of them is the current character set for codes with high bit one (initially G1). Each graphic character set has 94 or 96 characters, and is essentially a 7-bit character set. It uses codes either 040-0177 (041-0176) or 0240-0377 (0241-0376). G0 always has size 94 and uses codes 041-0176.
Switching between character sets is done using the shift
^N (SO or LS1),
^O (SI or LS0), ESC n (LS2),
ESC o (LS3), ESC N (SS2), ESC O (SS3), ESC ~ (LS1R), ESC }
(LS2R), ESC | (LS3R). The function LS
n makes character set
n the current
one for codes with high bit zero. The function LS
nR makes character set
n the current
one for codes with high bit one. The function SS
n makes character set
n=2 or 3) the current one
for the next character only (regardless of the value of its
high order bit).
A 94-character set is designated as G
n character set by an
escape sequence ESC ( xx (for G0), ESC ) xx (for G1), ESC *
xx (for G2), ESC + xx (for G3), where xx is a symbol or a
pair of symbols found in the ISO 2375 International
Register of Coded Character Sets. For example, ESC ( @
selects the ISO 646 character set as G0, ESC ( A selects
the UK standard character set (with pound instead of number
sign), ESC ( B selects ASCII (with dollar instead of
currency sign), ESC ( M selects a character set for African
languages, ESC ( ! A selects the Cuban character set, and
A 96-character set is designated as G
n character set by an
escape sequence ESC − xx (for G1), ESC . xx (for G2)
or ESC / xx (for G3). For example, ESC − G selects
the Hebrew alphabet as G1.
A multibyte character set is designated as G
n character set by an
escape sequence ESC $ xx or ESC $ ( xx (for G0), ESC $ ) xx
(for G1), ESC $ * xx (for G2), ESC $ + xx (for G3). For
example, ESC $ ( C selects the Korean character set for G0.
The Japanese character set selected by ESC $ B has a more
recent version selected by ESC & @ ESC $ B.
ISO 4873 stipulates a narrower use of character sets,
where G0 is fixed (always ASCII), so that G1, G2 and G3 can
be invoked only for codes with the high order bit set. In
^O are not used anymore, ESC (
xx can be used only with xx=B, and ESC ) xx, ESC * xx, ESC
+ xx are equivalent to ESC − xx, ESC . xx, ESC / xx,
TIS-620 is a Thai national standard character set and a superset of ASCII. In the same fashion as the ISO 8859 series, Thai characters are mapped into 0xa1-0xfe.
Unicode (ISO 10646) is a standard which aims to unambiguously represent every character in every human language. Unicode's structure permits 20.1 bits to encode every character. Since most computers don't include 20.1-bit integers, Unicode is usually encoded as 32-bit integers internally and either a series of 16-bit integers (UTF-16) (needing two 16-bit integers only when encoding certain rare characters) or a series of 8-bit bytes (UTF-8).
Linux represents Unicode using the 8-bit Unicode Transformation Format (UTF-8). UTF-8 is a variable length encoding of Unicode. It uses 1 byte to code 7 bits, 2 bytes for 11 bits, 3 bytes for 16 bits, 4 bytes for 21 bits, 5 bytes for 26 bits, 6 bytes for 31 bits.
Let 0,1,x stand for a zero, one, or arbitrary bit. A byte 0xxxxxxx stands for the Unicode 00000000 0xxxxxxx which codes the same symbol as the ASCII 0xxxxxxx. Thus, ASCII goes unchanged into UTF-8, and people using only ASCII do not notice any change: not in code, and not in file size.
A byte 110xxxxx is the start of a 2-byte code, and 110xxxxx 10yyyyyy is assembled into 00000xxx xxyyyyyy. A byte 1110xxxx is the start of a 3-byte code, and 1110xxxx 10yyyyyy 10zzzzzz is assembled into xxxxyyyy yyzzzzzz. (When UTF-8 is used to code the 31-bit ISO 10646 then this progression continues up to 6-byte codes.)
For most texts in ISO 8859 character sets, this means that the characters outside of ASCII are now coded with two bytes. This tends to expand ordinary text files by only one or two percent. For Russian or Greek texts, this expands ordinary text files by 100%, since text in those languages is mostly outside of ASCII. For Japanese users this means that the 16-bit codes now in common use will take three bytes. While there are algorithmic conversions from some character sets (especially ISO 8859-1) to Unicode, general conversion requires carrying around conversion tables, which can be quite large for 16-bit codes.
Note that UTF-8 is self-synchronizing: 10xxxxxx is a tail, any other byte is the head of a code. Note that the only way ASCII bytes occur in a UTF-8 stream, is as themselves. In particular, there are no embedded NULs ('\0') or '/'s that form part of some larger code.
Since ASCII, and, in particular, NUL and '/', are unchanged, the kernel does not notice that UTF-8 is being used. It does not care at all what the bytes it is handling stand for.
Rendering of Unicode data streams is typically handled through "subfont" tables which map a subset of Unicode to glyphs. Internally the kernel uses Unicode to describe the subfont loaded in video RAM. This means that in the Linux console in UTF-8 mode, one can use a character set with 512 different symbols. This is not enough for Japanese, Chinese, and Korean, but it is enough for most other purposes.
This page is part of release 4.07 of the Linux
man-pages project. A
description of the project, information about reporting bugs,
and the latest version of this page, can be found at
t -*- coding: UTF-8 -*-
Copyright (c) 1996 Eric S. Raymond <esrthyrsus.com>
and Copyright (c) Andries Brouwer <aebcwi.nl>
This is free documentation; you can redistribute it and/or
modify it under the terms of the GNU General Public License as
published by the Free Software Foundation; either version 2 of
the License, or (at your option) any later version.
This is combined from many sources, including notes by aeb and
research by esr. Portions derive from a writeup by Roman Czyborra.
Changes also by David Starner <dstarner98aasaa.ofe.org>.