newlib-cygwin/winsup/doc/setup-locale.xml

<?xml version="1.0" encoding='UTF-8'?>
<!DOCTYPE sect1 PUBLIC "-//OASIS//DTD DocBook V4.5//EN"
		"http://www.oasis-open.org/docbook/xml/4.5/docbookx.dtd">

<sect1 id="setup-locale"><title>Internationalization</title>

<sect2 id="setup-locale-ov"><title>Overview</title>

<para>
Internationalization support is controlled by the <envar>LANG</envar> and
<envar>LC_xxx</envar> environment variables.  You can set all of them
but Cygwin itself only honors the variables <envar>LC_ALL</envar>,
<envar>LC_CTYPE</envar>, and <envar>LANG</envar>, in this order, according
to the POSIX standard.  The content of these variables should follow the
POSIX standard for a locale specifier.  The correct form of a locale
specifier is</para>

<screen>
  language[[_TERRITORY][.charset][@modifier]]
</screen>

<para>"language" is a lowercase two character string per ISO 639-1, or,
if there is no ISO 639-1 code for the language (for instance, "Lower Sorbian"),
a three character string per ISO 639-3.</para>

<para>"TERRITORY" is an uppercase two character string per ISO 3166, charset is
one of a list of supported character sets.  The modifier doesn't matter
here (though some are recognized, see below).  If you're interested in the
exact description, you can find it in the online publication of the POSIX
manual pages on the homepage of the
<ulink url="http://www.opengroup.org/">Open Group</ulink>.</para>

<para>Typical locale specifiers are</para>

<screen>
  "de_CH"	   language = German, territory = Switzerland, default charset
  "fr_FR.UTF-8"    language = french, territory = France, charset = UTF-8
  "ko_KR.eucKR"    language = korean, territory = South Korea, charset = eucKR
  "syr_SY"         language = Syriac, territory = Syria, default charset
</screen>

<para>
If the locale specifier does not follow the above form, Cygwin checks
if the locale is one of the locale aliases defined in the file
<filename>/usr/share/locale/locale.alias</filename>.  If so, and if
the replacement localename is supported by the underlying Windows,
the locale is accepted, too.  So, given the default content of the
<filename>/usr/share/locale/locale.alias</filename> file, the below
examples would be valid locale specifiers as well.
</para>

<screen>
  "catalan"        defined as "ca_ES.ISO-8859-1" in locale.alias
  "japanese"       defined as "ja_JP.eucJP"      in locale.alias
  "turkish"        defined as "tr_TR.ISO-8859-9" in locale.alias
</screen>

<para>The file <filename>/usr/share/locale/locale.alias</filename> is
provided by the gettext package under Cygwin.</para>

<para>
At application startup, the application's locale is set to the default
"C" or "POSIX" locale.  This locale defaults to the ASCII character set
on the application level.  If you want to stick to the "C" locale and only
change to another charset, you can define this by setting one of the locale
environment variables to "C.charset".  For instance</para>

<screen>
  "C.ISO-8859-1"
</screen>

<note><para>The default locale in the absence of the aforementioned locale
environment variables is "C.UTF-8".</para></note>

<para>Windows uses the UTF-16 charset exclusively to store the names
of any object used by the Operating System.  This is especially important
with filenames.  Cygwin uses the setting of the locale environment variables
<envar>LC_ALL</envar>, <envar>LC_CTYPE</envar>, and <envar>LANG</envar>, to
determine how to convert Windows filenames from their UTF-16 representation
to the singlebyte or multibyte character set used by Cygwin.</para>

<para>
The setting of the locale environment variables at process startup
is effective for Cygwin's internal conversions to and from the Windows UTF-16
object names for the entire lifetime of the current process.  Changing
the environment variables to another value changes the way filenames are
converted in subsequently started child processes, but not within the same
process.</para>

<para>
However, even if one of the locale environment variables is set to
some other value than "C", this does <emphasis>only</emphasis> affect
how Cygwin itself converts filenames.  As the POSIX standard requires,
it's the application's responsibility to activate that locale for its
own purposes, typically by using the call</para>

<screen>
  setlocale (LC_ALL, "");
</screen>

<para>early in the application code.  Again, so that this doesn't get
lost:  If the application calls setlocale as above, and there is none
of the important locale variables set in the environment, the locale
is set to the default locale, which is "C.UTF-8".</para>

<para>But what about applications which are not locale-aware?  Per POSIX,
they are running in the "C" or "POSIX" locale, which implies the ASCII
charset.  The Cygwin DLL itself, however, will nevertheless use the locale
set in the environment (or the "C.UTF-8" default locale) for converting
filenames etc.</para>

<para>When the locale in the environment specifies an ASCII charset,
for example "C" or "en_US.ASCII", Cygwin will still use UTF-8
under the hood to translate filenames.  This allows for easier
interoperability with applications running in the default "C.UTF-8" locale.
</para>

<para>
The language and territory are used to fetch locale-dependent information
from Windows.  If the language and territory are not known to Windows, the
<function>setlocale</function> function fails.</para>

<para>The following modifiers are recognized.  Any other modifier is simply
ignored for now.</para>

<itemizedlist mark="bullet">

<listitem><para>
For locales which use the Euro (EUR) as currency, the modifier "@euro"
can be added to enforce usage of the ISO-8859-15 character set, which
includes a character for the "Euro" currency sign.
</para></listitem>

<listitem><para>
The default script used for all Serbian language locales (sr_BA, sr_ME, sr_RS,
and the deprecated sr_CS and sr_SP) is cyrillic.  With the "@latin" modifier
it gets switched to the latin script with the respective collation behaviour.
</para></listitem>

<listitem><para>
The default charset of the "be_BY" locale (Belarusian/Belarus) is CP1251.
With the "@latin" modifier it's UTF-8.
</para></listitem>

<listitem><para>
The default charset of the "tt_RU" locale (Tatar/Russia) is ISO-8859-5.
With the "@iqtelif" modifier it's UTF-8.
</para></listitem>

<listitem><para>
The default charset of the "uz_UZ" locale (Uzbek/Uzbekistan) is ISO-8859-1.
With the "@cyrillic" modifier it's UTF-8.
</para></listitem>

<listitem><para>
There's a class of characters in the Unicode character set, called the
"CJK Ambiguous Width" characters.  For these characters, the width
returned by the wcwidth/wcswidth functions is usually 1.  This can be a
problem with East-Asian languages, which historically use character sets
where these characters have a width of 2.  Therefore, wcwidth/wcswidth
return 2 as the width of these characters when an East-Asian charset such
as GBK or SJIS is selected, or when UTF-8 is selected and the language is
specified as "zh" (Chinese), "ja" (Japanese), or "ko" (Korean).  This is
not correct in all circumstances, hence the locale modifier "@cjknarrow"
can be used to force wcwidth/wcswidth to return 1 for the ambiguous width
characters.
</para></listitem>

<listitem><para>
For the same class of "CJK Ambiguous Width" characters, it may be
desirable to handle them as double-width even when a non-CJK language
setting is selected.  This supports e.g. certain graphic symbols used
by "Powerline" and provided by "Powerline fonts".  Some terminals have
options to enforce this width handling (xterm -cjk_width,
mintty -o Charwidth=ambig-wide, putty configuration) but that alone
makes character rendering and locale information inconsistent for those
characters.  The locale modifier "@cjkwide" supports consistent locale
response with this option; it forces wcwidth/wcswidth to return 2 for the
ambiguous width characters.
</para></listitem>

</itemizedlist>

</sect2>

<sect2 id="setup-locale-how"><title>How to set the locale</title>

<itemizedlist mark="bullet">

<listitem><para>
Assume that you've set one of the aforementioned environment variables to some
valid POSIX locale value, other than "C" and "POSIX".  Assume further that
you're living in Japan.  You might want to use the language code "ja" and the
territory "JP", thus setting, say, <envar>LANG</envar> to "ja_JP".  You didn't
set a character set, so what will Cygwin use now?  The default character set
is determined by the default Windows ANSI codepage for this language and
territory.  Cygwin uses a character set which is the typical Unix-equivalent
to the Windows ANSI codepage.  For instance:</para>

<screen>
  "en_US"		ISO-8859-1
  "el_GR"		ISO-8859-7
  "pl_PL"		ISO-8859-2
  "pl_PL@euro"		ISO-8859-15
  "ja_JP"		EUCJP
  "ko_KR"		EUCKR
  "te_IN"		UTF-8
</screen>
</listitem>

<listitem><para>
You don't want to use the default character set?  In that case you have to
specify the charset explicitly.  For instance, assume you're from Japan and
don't want to use the japanese default charset EUC-JP, but the Windows
default charset SJIS.  What you can do, for instance, is to set the
<envar>LANG</envar> variable in the <command>mintty</command> Cygwin Terminal
in the "Text" section of its "Options" dialog.  If you're starting your
Cygwin session via a batch file or a shortcut to a batch file, you can also
just set LANG there:</para>

<screen>
  @echo off

  C:
  chdir C:\cygwin\bin
  set LANG=ja_JP.SJIS
  bash --login -i
</screen>

<note><para>For a list of locales supported by your Windows machine, use the new
<command>locale -a</command> command, which is part of the Cygwin package.
For a description see <xref linkend="locale"></xref></para></note>

<note><para>For a list of supported character sets, see
<xref linkend="setup-locale-charsetlist"></xref>
</para></note>
</listitem>

<listitem><para>
Last, but not least, most singlebyte or doublebyte charsets have a big
disadvantage.  Windows filesystems use the Unicode character set in the
UTF-16 encoding to store filename information.  Not all characters
from the Unicode character set are available in a singlebyte or doublebyte
charset.  While Cygwin has a workaround to access files with unusual
characters (see <xref linkend="pathnames-unusual"></xref>), a better
workaround is to use always the UTF-8 character set.</para>

<para><emphasis>UTF-8 is the only multibyte character set which can represent
every Unicode character.</emphasis></para>

<screen>
  set LANG=es_MX.UTF-8
</screen>

<para>For a description of the Unicode standard, see the homepage of the
<ulink url="http://www.unicode.org/">Unicode Consortium</ulink>.
</para></listitem>

</itemizedlist>

</sect2>

<sect2 id="setup-locale-console"><title>The Windows Console character set</title>

<para>Sometimes the Windows console is used to run Cygwin applications.
While terminal emulations like the Cygwin Terminal <command>mintty</command>
or <command>xterm</command> have a distinct way to set the character set
used for in- and output, the Windows console hasn't such a way, since it's
not an application in its own right.</para>

<para>This problem is solved in Cygwin as follows.  When a Cygwin
process is started in a Windows console (either explicitly from cmd.exe,
or implicitly by, for instance, running the
<filename>C:\cygwin\Cygwin.bat</filename> batch file), the Console character
set is determined by the setting of the aforementioned
internationalization environment variables, the same way as described in
<xref linkend="setup-locale-how"></xref>.  </para>

<para>What is that good for?  Why not switch the console character set with
the applications requirements?  After all, the application knows if it uses
localization or not.  However, what if a non-localized application calls
a remote application which itself is localized?  This can happen with
<command>ssh</command> or <command>rlogin</command>.  Both commands don't
have and don't need localization and they never call
<function>setlocale</function>.  Setting one of the internationalization
environment variable to the same charset as the remote machine before
starting <command>ssh</command> or <command>rlogin</command> fixes that
problem.</para>

</sect2>

<sect2 id="setup-locale-problems"><title>Potential Problems when using Locales</title>

<para>
You can set the above internationalization variables not only when
starting the first Cygwin process, but also in your Cygwin shell on the
fly, even switch to yet another character set, and yet another.  In bash
for instance:</para>

<screen>
  <prompt>bash$</prompt> export LC_CTYPE="nl_BE.UTF-8"
</screen>

<para>However, here's a problem.  At the start of the first Cygwin process
in a session, the Windows environment is converted from UTF-16 to UTF-8.
The environment is another of the system objects stored in UTF-16 in
Windows.</para>

<para>As long as the environment only contains ASCII characters, this is
no problem at all.  But if it contains native characters, and you're planning
to use, say, GBK, the environment will result in invalid characters in
the GBK charset.  This would be especially a problem in variables like
<envar>PATH</envar>.  To circumvent the worst problems, Cygwin converts
the <envar>PATH</envar> environment variable to the charset set in the
environment, if it's different from the UTF-8 charset.</para>

<note><para>Per POSIX, the name of an environment variable should only
consist of valid ASCII characters, and only of uppercase letters, digits, and
the underscore for maximum portability.</para></note>

<para>Very old symbolic links may pose a problem when switching charsets on
the fly.  A symbolic link contains the filename of the target file the
symlink points to.  When a symlink had been created with versions of Cygwin
prior to Cygwin 1.7, the current ANSI or OEM character set had been used to
store the target filename, dependent on the old <envar>CYGWIN</envar>
environment variable setting <envar>codepage</envar> (see <xref
linkend="cygwinenv-removed-options"></xref>.  If the target filename
contains non-ASCII characters and you use another character set than
your default ANSI/OEM charset, the target filename of the symlink is now
potentially an invalid character sequence in the new character set.
This behaviour is not different from the behaviour in other Operating
Systems.  Recreate the symlink if that happens to you.</para>

</sect2>

<sect2 id="setup-locale-charsetlist"><title>List of supported character sets</title>

<para>Last but not least, here's the list of currently supported character
sets.  The left-hand expression is the name of the charset, as you would use
it in the internationalization environment variables as outlined above.
Note that charset specifiers are case-insensitive.  <literal>EUCJP</literal>
is equivalent to <literal>eucJP</literal> or <literal>eUcJp</literal>.
Writing the charset in the exact case as given in the list below is a
good convention, though.
</para>

<para>The right-hand side is the number of the equivalent Windows
codepage as well as the Windows name of the codepage.  They are only
noted here for reference.  Don't try to use the bare codepage number or
the Windows name of the codepage as charset in locale specifiers, unless
they happen to be identical with the left-hand side.  Especially in case
of the "CPxxx" style charsets, always use them with the trailing "CP".</para>

<para>This works:</para>

<screen>
  set LC_ALL=en_US.CP437
</screen>

<para>This does <emphasis>not</emphasis> work:</para>

<screen>
  set LC_ALL=en_US.437
</screen>

<para>You can find a full list of Windows codepages on the Microsoft MSDN page
<ulink url="http://msdn.microsoft.com/en-us/library/dd317756(VS.85).aspx">Code Page Identifiers</ulink>.</para>

<screen>
    Charset               Codepage
    -------------------   -------------------------------------------
    ASCII                 20127 (US_ASCII)

    CP437                   437 (OEM United States)
    CP720                   720 (DOS Arabic)
    CP737                   737 (OEM Greek)
    CP775                   775 (OEM Baltic)
    CP850                   850 (OEM Latin 1, Western European)
    CP852                   852 (OEM Latin 2, Central European)
    CP855                   855 (OEM Cyrillic)
    CP857                   857 (OEM Turkish)
    CP858                   858 (OEM Latin 1 + Euro Symbol)
    CP862                   862 (OEM Hebrew)
    CP866                   866 (OEM Russian)
    CP874                   874 (ANSI/OEM Thai)
    CP932		    932 (Shift_JIS, not exactly identical to SJIS)
    CP1125                 1125 (OEM Ukraine)
    CP1250                 1250 (ANSI Central European)
    CP1251                 1251 (ANSI Cyrillic)
    CP1252                 1252 (ANSI Latin 1, Western European)
    CP1253                 1253 (ANSI Greek)
    CP1254                 1254 (ANSI Turkish)
    CP1255                 1255 (ANSI Hebrew)
    CP1256                 1256 (ANSI Arabic)
    CP1257                 1257 (ANSI Baltic)
    CP1258                 1258 (ANSI/OEM Vietnamese)

    ISO-8859-1            28591 (ISO-8859-1)
    ISO-8859-2            28592 (ISO-8859-2)
    ISO-8859-3            28593 (ISO-8859-3)
    ISO-8859-4            28594 (ISO-8859-4)
    ISO-8859-5            28595 (ISO-8859-5)
    ISO-8859-6            28596 (ISO-8859-6)
    ISO-8859-7            28597 (ISO-8859-7)
    ISO-8859-8            28598 (ISO-8859-8)
    ISO-8859-9            28599 (ISO-8859-9)
    ISO-8859-10             -   (not available)
    ISO-8859-11             -   (not available)
    ISO-8859-13           28603 (ISO-8859-13)
    ISO-8859-14             -   (not available)
    ISO-8859-15           28605 (ISO-8859-15)
    ISO-8859-16             -   (not available)

    Big5                    950 (ANSI/OEM Traditional Chinese)
    EUCCN or euc-CN         936 (ANSI/OEM Simplified Chinese)
    EUCJP or euc-JP       20932 (EUC Japanese)
    EUCKR or euc-KR         949 (EUC Korean)
    GB2312                  936 (ANSI/OEM Simplified Chinese)
    GBK                     936 (ANSI/OEM Simplified Chinese)
    GEORGIAN-PS             -   (not available)
    KOI8-R                20866 (KOI8-R Russian Cyrillic)
    KOI8-U                21866 (KOI8-U Ukrainian Cyrillic)
    PT154                   -   (not available)
    SJIS                    -   (not available, almost, but not exactly CP932)
    TIS620 or TIS-620       874 (ANSI/OEM Thai)

    UTF-8 or utf8         65001 (UTF-8)
</screen>

</sect2>

</sect1>