mirror of
https://sourceware.org/git/glibc.git
synced 2024-11-27 15:30:07 +00:00
f982160ead
A handful of regexes were allowing +1 for yesexpr and -0 for noexpr, and it's the i18n definition. Standardize all locales by allowing these language-independent values in them. Example change for en_US goes from ^[yY] to ^[+1yY], and from ^[nN] to ^[-0nN].
1182 lines
41 KiB
Plaintext
1182 lines
41 KiB
Plaintext
comment_char %
|
|
escape_char /
|
|
|
|
% This file is part of the GNU C Library and contains locale data.
|
|
% The Free Software Foundation does not claim any copyright interest
|
|
% in the locale data contained in this file. The foregoing does not
|
|
% affect the license of the GNU C Library as a whole. It does not
|
|
% exempt you from the conditions of the license if your use would
|
|
% otherwise be governed by that license.
|
|
|
|
% Ukrainian Language Locale for Ukraine
|
|
% Source:
|
|
% Contact: Volodymyr M. Lisivka
|
|
% Email: v_lisivka@users.sourceforge.net
|
|
% Contact: Max Kutny
|
|
% Email: mkutny@gmail.com
|
|
% Language: uk
|
|
% Territory: UA
|
|
% Revision: 2.1.12
|
|
% Date: 2006-05-20
|
|
% Application: general
|
|
% Users: general
|
|
% Charset: UTF-8
|
|
%
|
|
% Keywords in this file are treated according to ISO/IEC 14652
|
|
% http://anubis.dkuug.dk/jtc1/sc22/wg20/docs/14652fcd.txt
|
|
%
|
|
% Minor fixes and overall cleanup by Max Kutny <mkutny@gmail.com>
|
|
%
|
|
|
|
|
|
LC_IDENTIFICATION
|
|
title "Ukrainian Language Locale for Ukraine"
|
|
source ""
|
|
address ""
|
|
contact "GNU libc maintainers"
|
|
email "bug-glibc-locales@gnu.org"
|
|
tel ""
|
|
fax ""
|
|
language "Ukrainian"
|
|
territory "Ukraine"
|
|
revision "2.1.12"
|
|
date "2006-05-20"
|
|
|
|
category "i18n:2012";LC_IDENTIFICATION
|
|
category "i18n:2012";LC_CTYPE
|
|
category "i18n:2012";LC_COLLATE
|
|
category "i18n:2012";LC_TIME
|
|
category "i18n:2012";LC_NUMERIC
|
|
category "i18n:2012";LC_MONETARY
|
|
category "i18n:2012";LC_MESSAGES
|
|
category "i18n:2012";LC_PAPER
|
|
category "i18n:2012";LC_NAME
|
|
category "i18n:2012";LC_ADDRESS
|
|
category "i18n:2012";LC_TELEPHONE
|
|
category "i18n:2012";LC_MEASUREMENT
|
|
END LC_IDENTIFICATION
|
|
|
|
|
|
LC_CTYPE
|
|
|
|
copy "<U0069><U0031><U0038><U006E>"
|
|
|
|
translit_start
|
|
|
|
% http://www.rada.gov.ua/translit.htm
|
|
% UKRAINIAN-ENGLISH TRANSLITERATION TABLE
|
|
%
|
|
% On 19 April 1996, an official Ukrainian-English transliteration system
|
|
% was adopted by the Ukrainian Legal Terminology Commission (Decision
|
|
% No 9).
|
|
%
|
|
% * Use of the approved system is not mandatory for the transliteration
|
|
% of foreign names into Ukrainian.
|
|
% * Transliteration should be made directly between Ukrainian and English
|
|
% without the use of any intermediary languages.
|
|
% * Decision No9, in accordance with the Legal Terminology Commission's
|
|
% express authority, is binding only for the transliteration of Ukrainian
|
|
% names in English in legislative and official acts.
|
|
% * For brevity's sake, the system routinely allows for names such as the
|
|
% city of 'Zaporizhzhia' to be given as 'Zaporizhia,' 'L`viv' as 'Lviv',
|
|
% etc. Also included is a short list of official spellings for miscellaneous
|
|
% terms: 'Ukraine' (no use of the article 'the'), 'Crimea' (as opposed to
|
|
% 'Krym'), 'Black Sea,' and 'Sea of Azov'. In certain cases, 'traditional'
|
|
% forms may be shown in parentheses after the official form:
|
|
% 'Dnipro (Dnieper).'
|
|
% * In addition, apostrophe marks and softening marks may be omitted upon
|
|
% transliteration into English.
|
|
%
|
|
% Ukrainian English Note Example
|
|
% letter letter
|
|
%
|
|
|
|
% <U0410> <U0410> - <U0410><U043B><U0443><U0448><U0442><U0430> - Alushta
|
|
<U0410> "<U0041>"
|
|
<U0430> "<U0061>"
|
|
|
|
% <U0411> B - <U0411><U043E><U0440><U0449><U0430><U0433><U0456><U0432><U043A><U0430> - Borschahivka
|
|
<U0411> "<U0042>"
|
|
<U0431> "<U0062>"
|
|
|
|
% <U0412> V - <U0412><U0438><U0448><U0433><U043E><U0440><U043E><U0434> - Vyshhorod
|
|
<U0412> "<U0056>"
|
|
<U0432> "<U0076>"
|
|
|
|
% <U0413> H, gh <U041D>-in most cases, gh - when recreating the combination "<U0437><U0433>" <U0413><U0430><U0434><U044F><U0447> - Hadiach; <U0417><U0433><U043E><U0440><U0430><U043D><U0438> - Zghorany
|
|
<U0413> "<U0048>"
|
|
<U0433> "<U0068>"
|
|
<U0417><U0413> "<U005A><U0047><U0048>"
|
|
<U0417><U0433> "<U005A><U0067><U0068>"
|
|
<U0437><U0413> "<U007A><U0047><U0048>"
|
|
<U0437><U0433> "<U007A><U0067><U0068>"
|
|
|
|
% <U0490> G - <U0490><U0430><U043B><U0430><U0491><U0430><U043D> - Galagan
|
|
<U0490> "<U0047>"
|
|
<U0491> "<U0067>"
|
|
|
|
% <U0414> D - <U0414><U043E><U043D> - Don
|
|
<U0414> "<U0044>"
|
|
<U0434> "<U0064>"
|
|
|
|
% <U0415> E - <U0420><U0456><U0432><U043D><U0435> - Rivne
|
|
<U0415> "<U0045>"
|
|
<U0435> "<U0065>"
|
|
|
|
% <U0404> Ye, ie Ye - at the beginning of words, ie - in other positions <U0404><U043D><U0430><U043A><U0456><U0454><U0432><U0435> - Yenakiieve; <U041D><U0430><U0454><U043D><U043A><U043E> - Naienko
|
|
<U0404> "<U0059><U0045>"
|
|
<U0454> "<U0079><U0065>"
|
|
% TODO: 'ie' if after any letter
|
|
|
|
% <U0416> Zh - <U0416><U0438><U0442><U043E><U043C><U0438><U0440> - Zhytomyr
|
|
<U0416> "<U005A><U0048>"
|
|
<U0436> "<U007A><U0068>"
|
|
|
|
% <U0417> Z - <U0417><U0430><U043A><U0430><U0440><U043F><U0430><U0442><U0442><U044F> - Zakarpattia
|
|
<U0417> "<U005A>"
|
|
<U0437> "<U007A>"
|
|
|
|
% <U0418> Y - <U041C><U0435><U0434><U0432><U0438><U043D> - Medvyn
|
|
<U0418> "<U0059>"
|
|
<U0438> "<U0079>"
|
|
|
|
% <U0406> I - <U0406><U0440><U0448><U0430><U0432><U0430> - Irshava
|
|
<U0406> "<U0049>"
|
|
<U0456> "<U0069>"
|
|
|
|
% <U0407> Yi, I Yi - at the beginning of words, i - in other positions <U0407><U0436><U0430><U043A><U0435><U0432><U0438><U0447> - Yizhakevych; <U041A><U0430><U0434><U0456><U0457><U0432><U043A><U0430> - Kadiivka
|
|
<U0407> "<U0059><U0049>"
|
|
<U0457> "<U0079><U0069>"
|
|
|
|
% TODO: 'i' if after any letter
|
|
% <U0419> Y, i Y - at the beginning of words, i - in other positions <U0419><U043E><U0441><U0438><U043F><U0456><U0432><U043A><U0430> - Yosypivka; <U0421><U0442><U0440><U0438><U0439> - Stryi
|
|
<U0419> "<U0059>"
|
|
<U0439> "<U0079>"
|
|
% TODO: 'i' if after any letter
|
|
|
|
% <U041A> K - <U041A><U0438><U0457><U0432> - Kyiv
|
|
<U041A> "<U004B>"
|
|
<U043A> "<U006B>"
|
|
|
|
% <U041B> L - <U041B><U0435><U0431><U0435><U0434><U0438><U043D> - Lebedyn
|
|
<U041B> "<U004C>"
|
|
<U043B> "<U006C>"
|
|
|
|
% <U041C> M - <U041C><U0438><U043A><U043E><U043B><U0430><U0457><U0432> - Mykolaiv
|
|
<U041C> "<U004D>"
|
|
<U043C> "<U006D>"
|
|
|
|
% <U041D> N - <U041D><U0456><U0436><U0438><U043D> - Nizhin
|
|
<U041D> "<U004E>"
|
|
<U043D> "<U006E>"
|
|
|
|
% <U041E> O - <U041E><U0434><U0435><U0441><U0430> - Odesa
|
|
<U041E> "<U004F>"
|
|
<U043E> "<U006F>"
|
|
|
|
% <U041F> P - <U041F><U043E><U043B><U0442><U0430><U0432><U0430> - Poltava
|
|
<U041F> "<U0050>"
|
|
<U043F> "<U0070>"
|
|
|
|
% <U0420> R - <U0420><U043E><U043C><U043D><U0438> - Romny
|
|
<U0420> "<U0052>"
|
|
<U0440> "<U0072>"
|
|
|
|
% <U0421> S - <U0421><U0443><U043C><U0438> - Sumy
|
|
<U0421> "<U0053>"
|
|
<U0441> "<U0073>"
|
|
|
|
% <U0422> T - <U0422><U0435><U0442><U0435><U0440><U0456><U0432> - Teteriv
|
|
<U0422> "<U0054>"
|
|
<U0442> "<U0074>"
|
|
|
|
% <U0423> U - <U0423><U0436><U0433><U043E><U0440><U043E><U0434> - Uzhhorod
|
|
<U0423> "<U0055>"
|
|
<U0443> "<U0075>"
|
|
|
|
% <U0424> F - <U0424><U0430><U0441><U0442><U0456><U0432> - Fastiv
|
|
<U0424> "<U0046>"
|
|
<U0444> "<U0066>"
|
|
|
|
% <U0425> Kh - <U0425><U0430><U0440><U043A><U0456><U0432> - Kharkiv
|
|
<U0425> "<U004B><U0048>"
|
|
<U0445> "<U006B><U0068>"
|
|
|
|
% <U0426> Ts - <U0411><U0456><U043B><U0430> <U0426><U0435><U0440><U043A><U0432><U0430> - Bila Tserkva
|
|
<U0426> "<U0054><U0053>"
|
|
<U0446> "<U0074><U0073>"
|
|
|
|
% <U0427> Ch - <U0427><U0435><U0440><U043D><U0456><U0432><U0446><U0456> - Chernivtsi
|
|
<U0427> "<U0043><U0048>"
|
|
<U0447> "<U0063><U0068>"
|
|
|
|
% <U0428> Sh - <U0428><U043E><U0441><U0442><U043A><U0430> - Shostka
|
|
<U0428> "<U0053><U0048>"
|
|
<U0448> "<U0073><U0068>"
|
|
|
|
% <U0429> Sch - <U0413><U043E><U0449><U0430> - Hoscha
|
|
<U0429> "<U0053><U0043><U0048>"
|
|
<U0449> "<U0073><U0063><U0068>"
|
|
|
|
% <U042C> ` (see commentary) <U0420><U0443><U0441><U044C> - Rus`; <U041B><U044C><U0432><U0456><U0432> - L`viv
|
|
<U042C> "<U0060>"
|
|
<U044C> "<U0060>"
|
|
|
|
% <U042E> Yu, iu Yu - at the beginning of words, iu - in other positions <U042E><U0440><U0456><U0439> - Yurii;<U041A><U0440><U044E><U043A><U0456><U0432><U043A><U0430> - Krukivka
|
|
<U042E> "<U0059><U0055>"
|
|
<U044E> "<U0079><U0075>"
|
|
% TODO: 'iu' if after any letter
|
|
|
|
% <U042F> Ya, ia Ya - at the beginning of words, ia - in other positions <U042F><U0433><U043E><U0442><U0438><U043D> - Yahotyn;I<U0447><U043D><U044F> - Ichnia
|
|
<U042F> "<U0059><U0041>"
|
|
<U044F> "<U0079><U0061>"
|
|
|
|
% TODO: 'ia' if after any letter
|
|
% ' ' (see commentary) <U0417><U043D><U0430><U043C>'<U044F><U043D><U043A><U0430> - Znamianka
|
|
|
|
|
|
% Specials:
|
|
|
|
% NUMERO SIGN
|
|
<U2116> "<U004E><U006F>"
|
|
|
|
% LEFT-POINTING DOUBLE ANGLE QUOTATION MARK
|
|
<U00AB> "<U003C><U003C>"
|
|
|
|
% RIGHT-POINTING DOUBLE ANGLE QUOTATION MARK
|
|
<U00BB> "<U003E><U003E>"
|
|
|
|
translit_end
|
|
|
|
END LC_CTYPE
|
|
|
|
|
|
|
|
LC_COLLATE
|
|
% Ukrainian Alpahabet (1881-1933) (old and correct)
|
|
%
|
|
% <U0410><U0430> <U0411><U0431> <U0412><U0432> <U0413><U0433>
|
|
% <U0490><U0491> <U0414><U0434> <U0415><U0435> <U0404><U0454>
|
|
% <U0416><U0436> <U0417><U0437> <U0418><U0438> <U0406><U0456>
|
|
% <U0407><U0457> <U0419><U0439> <U041A><U043A> <U041B><U043B>
|
|
% <U041C><U043C> <U041D><U043D> <U041E><U043E> <U041F><U043F>
|
|
% <U0420><U0440> <U0421><U0441> <U0422><U0442> <U0423><U0443>
|
|
% <U0424><U0444> <U0425><U0445> <U0426><U0446> <U0427><U0447>
|
|
% <U0428><U0448> <U0429><U0449> <U042E><U044E> <U042F><U044F>
|
|
% <U042C><U044C>
|
|
%
|
|
%
|
|
% Ukrainian Alpahabet (1933-1991) (old)
|
|
%
|
|
% <U0410><U0430> <U0411><U0431> <U0412><U0432> <U0413><U0433>
|
|
% <U0414><U0434> <U0415><U0435> <U0404><U0454> <U0416><U0436>
|
|
% <U0417><U0437> <U0418><U0438> <U0406><U0456> <U0407><U0457>
|
|
% <U0419><U0439> <U041A><U043A> <U041B><U043B> <U041C><U043C>
|
|
% <U041D><U043D> <U041E><U043E> <U041F><U043F> <U0420><U0440>
|
|
% <U0421><U0441> <U0422><U0442> <U0423><U0443> <U0424><U0444>
|
|
% <U0425><U0445> <U0426><U0446> <U0427><U0447> <U0428><U0448>
|
|
% <U0429><U0449> <U042E><U044E> <U042F><U044F> <U042C><U044C>
|
|
%
|
|
% Note:
|
|
% Ukrainian letter GHE_WITH_UPTURN (<U0490>) was removed from Ukrainian
|
|
% alphabet by Stalin in 1933 and was returned back in 1991 when
|
|
% Ukraine became independent from Soviet Union.
|
|
%
|
|
% Ukrainian Alphabet (1991) (current but not correct)
|
|
%
|
|
% <U0410><U0430> <U0411><U0431> <U0412><U0432> <U0413><U0433>
|
|
% <U0490><U0491> <U0414><U0434> <U0415><U0435> <U0404><U0454>
|
|
% <U0416><U0436> <U0417><U0437> <U0418><U0438> <U0406><U0456>
|
|
% <U0407><U0457> <U0419><U0439> <U041A><U043A> <U041B><U043B>
|
|
% <U041C><U043C> <U041D><U043D> <U041E><U043E> <U041F><U043F>
|
|
% <U0420><U0440> <U0421><U0441> <U0422><U0442> <U0423><U0443>
|
|
% <U0424><U0444> <U0425><U0445> <U0426><U0446> <U0427><U0447>
|
|
% <U0428><U0448> <U0429><U0449> <U042C><U044C> <U042E><U044E>
|
|
% <U042F><U044F>
|
|
%
|
|
% Note:
|
|
% Soft sign (<U042C>) is not considered to be a letter and therefore should have been
|
|
% placed at the end of the table. Unfortunately this letter was reordered in
|
|
% Ukrainian alpabet right before Ukraine got independency (1990-1991) by Soviet
|
|
% academic Ivanenko who tried to make MS DOS code pages compatible between
|
|
% Russian, Ukrainian and Belarusian languages.
|
|
%
|
|
% See "<U0050><U0072><U006F><U0062><U006C><U0065><U006D><U0079><U0020><U0075><U006B><U0072><U0061><U0069><U006E><U0069><U007A><U0061><U0074><U0063><U0069><U0069><U0020><U006B><U006F><U006D><U0070><U0075><U0074><U0065><U0072><U0069><U0076>" N2 (UDK 681.3.06), p. 16
|
|
% Ivanenko L. M. "Na shliakhu pobudovy <<idealnoho>> standartu abo
|
|
% synii ptakh Neokyrylytci".
|
|
|
|
copy "<U0069><U0073><U006F><U0031><U0034><U0036><U0035><U0031><U005F><U0074><U0031>"
|
|
|
|
% Ukrainian ghe is missing in iso14651_t1
|
|
collating-symbol <UKR-GHE>
|
|
|
|
% Apostrophe must be ignored during sorting because it's just a sign, not a
|
|
% real letter.
|
|
% ( "<U006E><U0060>"=="<U006E>", "<U0027><U0079><U0061>"=="<U0079><U0061>", etc. )
|
|
%
|
|
% Apostrophe already ignored by iso14651_t1.
|
|
%
|
|
% In the official alphabet the soft sign is a letter and has a hard position in
|
|
% the order.
|
|
|
|
|
|
collating-symbol <CAP-MIN>
|
|
collating-symbol <MIN-CAP>
|
|
|
|
% Taken from bg_BG 2.0.1:
|
|
|
|
% We have made the following changes to the basic collation scheme in
|
|
% the file iso14651_t1:
|
|
% 1. The Cyrillic script is first in the order.
|
|
% 2. The non-Bulgarian Cyrillic letters are sorted according to
|
|
% their transliteration with Bulgarian Cyrillic letters.
|
|
|
|
% Local changes:
|
|
% Added <UKR-GHE> collation symbol at proper position.
|
|
% Reordering of <UKR-I>,<UKR-YI> to wrong positions (according to Ukrainian
|
|
% alhabet) was removed.
|
|
|
|
reorder-after <9>
|
|
<CYR-A>
|
|
<CYR-BE>
|
|
<CYR-VE>
|
|
<CYR-GHE>
|
|
<UKR-GHE>
|
|
<CYR-DE>
|
|
<CYR-GZHE>
|
|
<CYR-DJE>
|
|
<CYR-IE>
|
|
<UKR-IE>
|
|
<CYR-IO>
|
|
<CYR-ZHE>
|
|
<CYR-ZE>
|
|
<CYR-DZE>
|
|
<CYR-I>
|
|
<UKR-I>
|
|
<UKR-YI>
|
|
<CYR-IBRE>
|
|
<CYR-JE>
|
|
<CYR-KA>
|
|
<CYR-EL>
|
|
<CYR-LJE>
|
|
<CYR-EM>
|
|
<CYR-EN>
|
|
<CYR-NJE>
|
|
<CYR-O>
|
|
<CYR-PE>
|
|
<CYR-ER>
|
|
<CYR-ES>
|
|
<CYR-TE>
|
|
<CYR-KJE>
|
|
<CYR-TSHE>
|
|
<CYR-OU>
|
|
<CYR-OUBRE>
|
|
<CYR-EF>
|
|
<CYR-HA>
|
|
<CYR-TSE>
|
|
<CYR-CHE>
|
|
<CYR-DCHE>
|
|
<CYR-SHA>
|
|
<CYR-SHTSHA>
|
|
<CYR-SIGDUR>
|
|
<CYR-YEROU>
|
|
<CYR-SIGMOUIL>
|
|
<CYR-E>
|
|
<CYR-YOU>
|
|
<CYR-YA>
|
|
|
|
% Placing <UKR-GHE> to proper position.
|
|
reorder-after <U0413>
|
|
<U0490> <UKR-GHE>;<BAS>;<CAP>;IGNORE
|
|
<U0403> <CYR-GHE>;<CYR-GZHE>;<CAP>;IGNORE % Mac. gje
|
|
reorder-after <U0433>
|
|
<U0491> <UKR-GHE>;<BAS>;<MIN>;IGNORE
|
|
<U0453> <CYR-GHE>;<CYR-GZHE>;<MIN>;IGNORE % Mac. gje
|
|
|
|
reorder-after <U0414>
|
|
<U0402> "<U003C><U0043><U0059><U0052><U002D><U0044><U0045><U003E><U003C><U0043><U0059><U0052><U002D><U005A><U0048><U0045><U003E>";"<U003C><U004C><U0049><U0047><U003E><U003C><U004C><U0049><U0047><U003E>";"<U003C><U0043><U0041><U0050><U003E><U003C><U0043><U0041><U0050><U003E>";IGNORE % CYR-DJE
|
|
<U040F> "<U003C><U0043><U0059><U0052><U002D><U0044><U0045><U003E><U003C><U0043><U0059><U0052><U002D><U005A><U0048><U0045><U003E>";"<U003C><U0043><U0059><U0052><U002D><U0044><U0043><U0048><U0045><U003E><U003C><U004C><U0049><U0047><U003E>";"<U003C><U0043><U0041><U0050><U003E><U003C><U0043><U0041><U0050><U003E>";IGNORE % CYR-DCHE
|
|
<U0405> "<U003C><U0043><U0059><U0052><U002D><U0044><U0045><U003E><U003C><U0043><U0059><U0052><U002D><U005A><U0045><U003E>";"<U003C><U004C><U0049><U0047><U003E><U003C><U004C><U0049><U0047><U003E>";"<U003C><U0043><U0041><U0050><U003E><U003C><U0043><U0041><U0050><U003E>";IGNORE % CYR-DZE
|
|
reorder-after <U0434>
|
|
<U0452> "<U003C><U0043><U0059><U0052><U002D><U0044><U0045><U003E><U003C><U0043><U0059><U0052><U002D><U005A><U0048><U0045><U003E>";"<U003C><U004C><U0049><U0047><U003E><U003C><U004C><U0049><U0047><U003E>";"<U003C><U004D><U0049><U004E><U003E><U003C><U004D><U0049><U004E><U003E>";IGNORE % CYR-DJE
|
|
<U045F> "<U003C><U0043><U0059><U0052><U002D><U0044><U0045><U003E><U003C><U0043><U0059><U0052><U002D><U005A><U0048><U0045><U003E>";"<U003C><U0043><U0059><U0052><U002D><U0044><U0043><U0048><U0045><U003E><U003C><U004C><U0049><U0047><U003E>";"<U003C><U004D><U0049><U004E><U003E><U003C><U004D><U0049><U004E><U003E>";IGNORE % CYR-DCHE
|
|
<U0455> "<U003C><U0043><U0059><U0052><U002D><U0044><U0045><U003E><U003C><U0043><U0059><U0052><U002D><U005A><U0045><U003E>";"<U003C><U004C><U0049><U0047><U003E><U003C><U004C><U0049><U0047><U003E>";"<U003C><U004D><U0049><U004E><U003E><U003C><U004D><U0049><U004E><U003E>";IGNORE % CYR-DZE
|
|
|
|
reorder-after <U0435>
|
|
<U0451> <CYR-IE>;<CYR-IO>;<MIN>;IGNORE
|
|
<U044D> <CYR-IE>;<CYR-E>;<MIN>;IGNORE
|
|
reorder-after <U0415>
|
|
<U0401> <CYR-IE>;<CYR-IO>;<CAP>;IGNORE
|
|
<U042D> <CYR-IE>;<CYR-E>;<CAP>;IGNORE
|
|
|
|
reorder-after <U041A>
|
|
<U040C> <CYR-KA>;<CYR-KJE>;<CAP>;IGNORE
|
|
reorder-after <U043A>
|
|
<U045C> <CYR-KA>;<CYR-KJE>;<MIN>;IGNORE
|
|
|
|
reorder-after <U041D>
|
|
<U040A> "<U003C><U0043><U0059><U0052><U002D><U0045><U004E><U003E><U003C><U0043><U0059><U0052><U002D><U0053><U0049><U0047><U004D><U004F><U0055><U0049><U004C><U003E>";"<U003C><U004C><U0049><U0047><U003E><U003C><U004C><U0049><U0047><U003E>";"<U003C><U0043><U0041><U0050><U003E><U003C><U0043><U0041><U0050><U003E>";IGNORE % CYR-NJE
|
|
reorder-after <U043D>
|
|
<U045A> "<U003C><U0043><U0059><U0052><U002D><U0045><U004E><U003E><U003C><U0043><U0059><U0052><U002D><U0053><U0049><U0047><U004D><U004F><U0055><U0049><U004C><U003E>";"<U003C><U004C><U0049><U0047><U003E><U003C><U004C><U0049><U0047><U003E>";"<U003C><U004D><U0049><U004E><U003E><U003C><U004D><U0049><U004E><U003E>";IGNORE % CYR-NJE
|
|
|
|
reorder-after <U0427>
|
|
<U040B> <CYR-CHE>;<CYR-TSHE>;<CAP>;IGNORE
|
|
reorder-after <U0447>
|
|
<U045B> <CYR-CHE>;<CYR-TSHE>;<MIN>;IGNORE
|
|
|
|
reorder-after <U041B>
|
|
<U0409> "<U003C><U0043><U0059><U0052><U002D><U0045><U004C><U003E><U003C><U0043><U0059><U0052><U002D><U0053><U0049><U0047><U004D><U004F><U0055><U0049><U004C><U003E>";"<U003C><U004C><U0049><U0047><U003E><U003C><U004C><U0049><U0047><U003E>";"<U003C><U0043><U0041><U0050><U003E><U003C><U0043><U0041><U0050><U003E>";IGNORE % CYR-LJE
|
|
reorder-after <U043B>
|
|
<U0459> "<U003C><U0043><U0059><U0052><U002D><U0045><U004C><U003E><U003C><U0043><U0059><U0052><U002D><U0053><U0049><U0047><U004D><U004F><U0055><U0049><U004C><U003E>";"<U003C><U004C><U0049><U0047><U003E><U003C><U004C><U0049><U0047><U003E>";"<U003C><U004D><U0049><U004E><U003E><U003C><U004D><U0049><U004E><U003E>";IGNORE % CYR-LJE
|
|
|
|
reorder-after <U0423>
|
|
<U040E> <CYR-OU>;<CYR-OUBRE>;<CAP>;IGNORE
|
|
reorder-after <U0443>
|
|
<U045E> <CYR-OU>;<CYR-OUBRE>;<MIN>;IGNORE
|
|
|
|
% Capital letters go before small letters. ("<U0041>"<"<U0061>")
|
|
reorder-after <CAP>
|
|
<CAP-MIN>
|
|
<MIN-CAP>
|
|
<MIN>
|
|
|
|
|
|
% Reorder letters with soft sign
|
|
% [<U0432><U0434><U0437><U043B><U043D><U0440><U0441><U0442><U0446>] + <U044C>
|
|
|
|
reorder-after <CYR-VE>
|
|
<V+SS>
|
|
reorder-after <U0412>
|
|
<V-SS> <CYR-VE>;<BAS>;<CAP>;<V+SS>
|
|
<V-ss> <CYR-VE>;<BAS>;<CAP-MIN>;<V+SS>
|
|
reorder-after <U0432>
|
|
<v-SS> <CYR-VE>;<BAS>;<MIN-CAP>;<V+SS>
|
|
<v-ss> <CYR-VE>;<BAS>;<MIN>;<V+SS>
|
|
|
|
reorder-after <CYR-DE>
|
|
<D+SS>
|
|
reorder-after <U0414>
|
|
<D-SS> <CYR-DE>;<BAS>;<CAP>;<D+SS>
|
|
<D-ss> <CYR-DE>;<BAS>;<CAP-MIN>;<D+SS>
|
|
reorder-after <U0434>
|
|
<d-SS> <CYR-DE>;<BAS>;<MIN-CAP>;<D+SS>
|
|
<d-ss> <CYR-DE>;<BAS>;<MIN>;<D+SS>
|
|
|
|
reorder-after <CYR-ZE>
|
|
<Z+SS>
|
|
reorder-after <U0417>
|
|
<Z-SS> <CYR-ZE>;<BAS>;<CAP>;<Z+SS>
|
|
<Z-ss> <CYR-ZE>;<BAS>;<CAP-MIN>;<Z+SS>
|
|
reorder-after <U0437>
|
|
<z-SS> <CYR-ZE>;<BAS>;<MIN-CAP>;<Z+SS>
|
|
<z-ss> <CYR-ZE>;<BAS>;<MIN>;<Z+SS>
|
|
|
|
reorder-after <CYR-EL>
|
|
<L+SS>
|
|
reorder-after <U041B>
|
|
<L-SS> <CYR-EL>;<BAS>;<CAP>;<L+SS>
|
|
<L-ss> <CYR-EL>;<BAS>;<CAP-MIN>;<L+SS>
|
|
reorder-after <U043B>
|
|
<l-SS> <CYR-EL>;<BAS>;<MIN-CAP>;<L+SS>
|
|
<l-ss> <CYR-EL>;<BAS>;<MIN>;<L+SS>
|
|
|
|
reorder-after <CYR-EN>
|
|
<N+SS>
|
|
reorder-after <U041D>
|
|
<N-SS> <CYR-EN>;<BAS>;<CAP>;<N+SS>
|
|
<N-ss> <CYR-EN>;<BAS>;<CAP-MIN>;<N+SS>
|
|
reorder-after <U043D>
|
|
<n-SS> <CYR-EN>;<BAS>;<MIN-CAP>;<N+SS>
|
|
<n-ss> <CYR-EN>;<BAS>;<MIN>;<N+SS>
|
|
|
|
reorder-after <CYR-ER>
|
|
<R+SS>
|
|
reorder-after <U0420>
|
|
<R-SS> <CYR-ER>;<BAS>;<CAP>;<R+SS>
|
|
<R-ss> <CYR-ER>;<BAS>;<CAP-MIN>;<R+SS>
|
|
reorder-after <U0440>
|
|
<r-SS> <CYR-ER>;<BAS>;<MIN-CAP>;<R+SS>
|
|
<r-ss> <CYR-ER>;<BAS>;<MIN>;<R+SS>
|
|
|
|
reorder-after <CYR-ES>
|
|
<S+SS>
|
|
reorder-after <U0421>
|
|
<S-SS> <CYR-ES>;<BAS>;<CAP>;<S+SS>
|
|
<S-ss> <CYR-ES>;<BAS>;<CAP-MIN>;<S+SS>
|
|
reorder-after <U0441>
|
|
<s-SS> <CYR-ES>;<BAS>;<MIN-CAP>;<S+SS>
|
|
<s-ss> <CYR-ES>;<BAS>;<MIN>;<S+SS>
|
|
|
|
reorder-after <CYR-TE>
|
|
<T+SS>
|
|
reorder-after <U0422>
|
|
<T-SS> <CYR-TE>;<BAS>;<CAP>;<T+SS>
|
|
<T-ss> <CYR-TE>;<BAS>;<CAP-MIN>;<T+SS>
|
|
reorder-after <U0442>
|
|
<t-SS> <CYR-TE>;<BAS>;<MIN-CAP>;<T+SS>
|
|
<t-ss> <CYR-TE>;<BAS>;<MIN>;<T+SS>
|
|
|
|
reorder-after <CYR-TSE>
|
|
<TSE+SS>
|
|
reorder-after <U0426>
|
|
<TS-SS> <CYR-TSE>;<BAS>;<CAP>;<TSE+SS>
|
|
<TS-ss> <CYR-TSE>;<BAS>;<CAP-MIN>;<TSE+SS>
|
|
reorder-after <U0446>
|
|
<ts-SS> <CYR-TSE>;<BAS>;<MIN-CAP>;<TSE+SS>
|
|
<ts-ss> <CYR-TSE>;<BAS>;<MIN>;<TSE+SS>
|
|
|
|
% To get back to correct but still unofficial pre-1991 alphabet uncomment the
|
|
% following lines that move soft sign (<CYR-YEROU>) to the end of the alphabet.
|
|
%
|
|
% reorder-after <CYR-YA>
|
|
% <CYR-YEROU>
|
|
% reorder-after <U042F>
|
|
% <U042C> <CYR-YEROU>;<BAS>;<CAP>;IGNORE
|
|
% reorder-after <U044F>
|
|
% <U044C> <CYR-YEROU>;<BAS>;<MIN>;IGNORE
|
|
|
|
reorder-end
|
|
|
|
|
|
END LC_COLLATE
|
|
|
|
|
|
|
|
LC_MESSAGES
|
|
|
|
|
|
% WARNING: symbols "<U0059>" and "<U0043><U0079><U0072><U0069><U006C><U006C><U0069><U0063><U005F><U004E>" are placed at one key in Ukrainian keyboard
|
|
% layout although the meaning is semantically opposite ("<U0059>" stands for "<U0059><U0065><U0073>"
|
|
% while "<U0043><U0079><U0072><U0069><U006C><U006C><U0069><U0063><U005F><U004E>" stands for "<U004E><U006F>").
|
|
%
|
|
% This problem may be dealt with by requiring at least two letters to be
|
|
% entered in Ukrainian language.
|
|
|
|
% A regular expression that describes possible yes-responses
|
|
% "<U0079>", "<U0059>", "<U002B>", "<U0074><U0061>" or "<U0074><U0061><U006B>" in Cyrillic
|
|
yesexpr "<U005E><U0028><U005B><U002B><U0031><U0059><U0079><U005D><U007C><U005B><U0422><U0442><U005D><U005B><U0410><U0430><U005D><U005B><U041A><U043A><U005D><U003F><U0029><U0024>"
|
|
|
|
% WARNING: symbols "<U004E>" and "<U0043><U0079><U0072><U0069><U006C><U006C><U0069><U0063><U005F><U0054>" are placed at one key in Ukrainian
|
|
% keyboard layout although the meaning is semantically opposite ("<U004E>" stands for
|
|
% "<U004E><U006F>" while "<U0043><U0079><U0072><U0069><U006C><U006C><U0069><U0063><U005F><U0054>" stands for "<U0059><U0065><U0073>").
|
|
%
|
|
% We may require two Ukrainian letters as input to deal with this potential
|
|
% problem.
|
|
|
|
% A regular expression that describes possible no-responses.
|
|
% "<U006E>", "<U004E>", "<U002D>" or word "<U006E><U0069>" in Cyrillic
|
|
noexpr "<U005E><U0028><U005B><U002D><U0030><U004E><U006E><U005D><U007C><U005B><U041D><U043D><U005D><U005B><U0406><U0456><U005D><U0029><U0024>"
|
|
% "<U006E>" "<U004E>" "<U002D>" or word "<U006E><U0069>" in Cyrillic
|
|
|
|
|
|
% Output string for ``yes'' ("<U0074><U0061><U006B>").
|
|
yesstr "<U0442><U0430><U043A>"
|
|
|
|
% Output string for ``no'' ("<U006E><U0069>").
|
|
nostr "<U043D><U0456>"
|
|
|
|
END LC_MESSAGES
|
|
|
|
|
|
|
|
LC_MONETARY
|
|
|
|
% Examples:
|
|
%
|
|
% 100,00 hr
|
|
% -100,00 hr
|
|
% UAH 100,00
|
|
% UAH -100,00
|
|
%
|
|
% 0,01 hr
|
|
% 0,10 hr
|
|
% 1,00 hr
|
|
% 10,00 hr
|
|
% 100,00 hr
|
|
% 1000,00 hr (EXCEPTION)
|
|
% 10 000,00 hr
|
|
% 100 000,00 hr
|
|
% 1 000 000,00 hr
|
|
% ...
|
|
%
|
|
% Note:
|
|
% There is potential conflict between "<U0068><U0072><U0079><U0076><U006E><U0079><U0061>" ("<U0068><U0072>", "<U0068><U0072><U0076><U002E>", money),
|
|
% "<U0068><U0072><U0061><U006D>" ("<U0068>","<U0068><U0072><U002E>", weight) and "<U0068><U006F><U0064><U0079><U006E><U0061>" ("<U0068><U006F><U0064>","<U0068><U006F><U0064><U002E>", time),
|
|
% because some people uses "<U0068>" for "<U0068><U006F><U0064><U0079><U006E><U0079>", "<U0068><U0072>" for "<U0068><U0072><U0061><U006D><U0079>", etc.
|
|
%
|
|
% 200h - 200 hram (weight)
|
|
% 200 hr. - 200 hram (weight)
|
|
% 200hd - 200 hodyn (time)
|
|
% 200 hod. - 200 hodyn (time)
|
|
% 200hr - 200 hryven (money)
|
|
% 200 hrv. - 200 hryven (money)
|
|
|
|
% the local currency symbol
|
|
currency_symbol "<U0433><U0440><U043D><U002E>" % hr (hryvnya)
|
|
|
|
% This must be a 4-character string containing the international currency
|
|
% symbol as defined by the ISO 4217 standard (three characters) followed
|
|
% by a separator.
|
|
%
|
|
% DSTU ISO 3166
|
|
% Code Name Name in English language
|
|
% UAH <U0433><U0440><U0438><U0432><U043D><U044F> hryvnya
|
|
int_curr_symbol "<U0055><U0041><U0048><U0020>"
|
|
|
|
% This string is used as the decimal delimiter when formatting monetary
|
|
% quantities.
|
|
% see LC_NUMERIC:decimal_point
|
|
mon_decimal_point "<U002C>"
|
|
|
|
% This string is used as a thousands separator when formatting monetary
|
|
% quantities.
|
|
% see LC_NUMERIC:thousands separator
|
|
mon_thousands_sep "<U00A0>" % <NBSP> (0xA0)
|
|
|
|
% Define the size of each group of digits in formatted monetary
|
|
% quantities. The operand is a sequence of integers separated by
|
|
% semicolons. Each integer specifies the number of digits in each
|
|
% group, with the initial integer defining the size of the group
|
|
% immediately preceding the decimal delimiter, and the following
|
|
% integers defining the preceding groups. If the last integer is
|
|
% not -1, then the size of the previous group (if any) shall be
|
|
% repeatedly used for the remainder of the digits. If the last
|
|
% integer is -1, then no further grouping shall be performed.
|
|
%
|
|
% FIXME:
|
|
% Note difference between "<U0031><U0032><U0033><U0020><U0034><U0035><U0036><U0020><U0037><U0038><U0039><U002C><U0030><U0031>" and "<U0031><U0032><U0033><U0034><U002C><U0035><U0036>".
|
|
% Unfortunately, it can't be implemented with current (2.3.6) glibc.
|
|
mon_grouping 3;3
|
|
|
|
% String that is used to indicate a positive sign for monetary
|
|
% quantities
|
|
positive_sign ""
|
|
|
|
% String that is used to indicate a negative sign for monetary
|
|
% quantities
|
|
negative_sign "<U002D>"
|
|
|
|
% An integer representing the number of fractional digits (those to the right of
|
|
% the decimal delimiter) to be written in a formatted monetary quantity using
|
|
% int_curr_symbol.
|
|
int_frac_digits 2
|
|
|
|
% An integer representing the number of fractional digits (those to the right of
|
|
% the decimal delimiter) to be written in a formatted monetary quantity using
|
|
% currency_symbol.
|
|
frac_digits 2
|
|
|
|
% 1 - the currency_symbol or int_curr_symbol should precede the
|
|
% formatted monetary quantity.
|
|
% 0 - the symbol succeeds the value.
|
|
p_cs_precedes 0
|
|
int_p_cs_precedes 1
|
|
|
|
% 0 - the currency_symbol succeeds the negative value.
|
|
% 1 - the currency_symbol precedes the negative value.
|
|
n_cs_precedes 0
|
|
int_n_cs_precedes 1
|
|
|
|
% 0 - means that no space should be printed between the currency_symbol
|
|
% and the value for a monetary quantity with a non-negative value.
|
|
% 1 - means that a space should be printed between the currency_symbol
|
|
% and the value.
|
|
% 2 - means that a space should be printed between the symbol and the
|
|
% sign string, if adjacent.
|
|
p_sep_by_space 2
|
|
int_p_sep_by_space 2
|
|
|
|
% 0 - no space separates the currency_symbol or int_curr_symbol from the
|
|
% value for a negative monetary quantity.
|
|
% 1 - a space separates the symbol from the value.
|
|
% 2 - a space separates the symbol and the sign string, if adjacent.
|
|
n_sep_by_space 1
|
|
int_n_sep_by_space 1
|
|
|
|
% An integer set to a value indicating the positioning of the
|
|
% positive_sign for a monetary quantity with a non-negative value.
|
|
%
|
|
% 0 - Parentheses enclose the quantity and the currency_symbol or
|
|
% int_curr_symbol.
|
|
% 1 - The sign string precedes the quantity and the currency_symbol or the
|
|
% int_curr_symbol.
|
|
% 2 - The sign string succeeds the quantity and the currency_symbol or the
|
|
% int_curr_symbol.
|
|
% 3 - The sign string precedes the currency_symbol or the int_curr_symbol.
|
|
% 4 - The sign string succeeds the currency_symbol or the int_curr_symbol.
|
|
p_sign_posn 1
|
|
int_p_sign_posn 4
|
|
|
|
% An integer set to a value indicating the positioning of the
|
|
% negative_sign for a monetary quantity with a negative value.
|
|
%
|
|
% 0 - Parentheses enclose the quantity and the currency_symbol or
|
|
% int_curr_symbol.
|
|
% 1 - The sign string precedes the quantity and the currency_symbol or the
|
|
% int_curr_symbol.
|
|
% 2 - The sign string succeeds the quantity and the currency_symbol or the
|
|
% int_curr_symbol.
|
|
% 3 - The sign string precedes the currency_symbol or the int_curr_symbol.
|
|
% 4 - The sign string succeeds the currency_symbol or the int_curr_symbol.
|
|
n_sign_posn 1
|
|
int_n_sign_posn 4
|
|
|
|
END LC_MONETARY
|
|
|
|
|
|
|
|
LC_NUMERIC
|
|
|
|
% Examples:
|
|
%
|
|
% 1
|
|
% 10
|
|
% 100
|
|
% 1000 (EXCEPTION)
|
|
% 10 000
|
|
% 100 000
|
|
% 1 000 000
|
|
%
|
|
% 1,0
|
|
% 0,1
|
|
% 0,01
|
|
% 0,001
|
|
% 0,0001
|
|
% 0,00001
|
|
%
|
|
% 1
|
|
% -1
|
|
%
|
|
% List with numbers - 1,0; 2,0; 3,0; 4,0.
|
|
%
|
|
% Note:
|
|
% We use ';' as separator between numbers.
|
|
|
|
% The string that will be used as the decimal delimiter when formatting
|
|
% numeric quantities.
|
|
decimal_point "<U002C>" % see LC_MONETARY:mon_decimal_pint
|
|
% Comma cause lot of problems - changed to dot.
|
|
% decimal_point "<U002E>" % see LC_MONETARY:mon_decimal_pint
|
|
|
|
% The string that will be used as a group separator when formatting
|
|
% numeric quantities.
|
|
% see LC_MONETARY:mon_thousands_sep
|
|
thousands_sep "<U00A0>" % <NBSP> (0xA0)
|
|
|
|
% A string that describes the formatting of numeric quantities.
|
|
%
|
|
% See mon_grouping
|
|
%
|
|
% FIXME:
|
|
% Note difference between "<U0031><U0032><U0033><U0020><U0034><U0035><U0036><U0020><U0037><U0038><U0039><U002C><U0030><U0031>" and "<U0031><U0032><U0033><U0034><U002C><U0035><U0036>".
|
|
% Unfortunately, it can't be implemented with current glibc.
|
|
grouping 3;3
|
|
|
|
END LC_NUMERIC
|
|
|
|
|
|
|
|
LC_TIME
|
|
|
|
% Examples of date:
|
|
%
|
|
% sereda, chotyrnadtcyate travnya dvitysyachi tret`oho roku n.e.
|
|
% sereda, chotyrnadtcyate travnya dvitysyachi tret`oho roku
|
|
% sereda, 14-e travnya 2003-ho roku n.e.
|
|
% sereda, 1-e travnya 2003-ho roku
|
|
% sereda, 1 travnya 2003
|
|
% sereda, 01-tra-2003
|
|
% sr, 01-tra-03
|
|
% sr, 01-tra-2003
|
|
% 01.05.03 (Recomended by DSTU)
|
|
% 01.05.2003 (Allowed but not recomended by DSTU)
|
|
% 2003.05.01 (Allowed but not recomended by DSTU)
|
|
% 01/tra/03 (Deprecated, but still commonly used format)
|
|
% 1-e travnya (not "<U0031><U0020><U0074><U0072><U0061><U0076><U006E><U0079><U0061>" or "<U0031><U0020><U0074><U0072><U0061><U0076><U0065><U006E><U0060>" or "<U0031><U002D><U0074><U0072><U0061>")
|
|
%
|
|
% Notes:
|
|
% month after day, year after month
|
|
% all month/weekday names and abbreviations in lower case
|
|
% month name in date must be used in another form, than in calendar
|
|
% day name may be used in another form in another context
|
|
% (sereda, v seredu, tciyeyi seredy, etc.)
|
|
|
|
% Examples of time:
|
|
% 23:59:59 +0200
|
|
% 00:00:00 +0200
|
|
% 00:00
|
|
|
|
week 7;19971130;1
|
|
|
|
% A list of abbreviated weekday names. (%a)
|
|
% Note:
|
|
% Never use three letters for day abbreviations because of conflict between
|
|
% SEReda (day name) and
|
|
% SERpen` (month name).
|
|
abday /
|
|
"<U043D><U0434>"; %nd /
|
|
"<U043F><U043D>"; %pn /
|
|
"<U0432><U0442>"; %vt /
|
|
"<U0441><U0440>"; %sr /
|
|
"<U0447><U0442>"; %cht /
|
|
"<U043F><U0442>"; %pt /
|
|
"<U0441><U0431>" %sb
|
|
|
|
% A list of weekday names starting with first day of week as defined by <week> keyword. (%A)
|
|
day /
|
|
"<U043D><U0435><U0434><U0456><U043B><U044F>"; %nedilya /
|
|
"<U043F><U043E><U043D><U0435><U0434><U0456><U043B><U043E><U043A>"; %ponedilok /
|
|
"<U0432><U0456><U0432><U0442><U043E><U0440><U043E><U043A>"; %vivtorok /
|
|
"<U0441><U0435><U0440><U0435><U0434><U0430>"; %sereda /
|
|
"<U0447><U0435><U0442><U0432><U0435><U0440>"; %chetver /
|
|
"<U043F><U0027><U044F><U0442><U043D><U0438><U0446><U044F>"; %pjatnycya /
|
|
"<U0441><U0443><U0431><U043E><U0442><U0430>" %subota
|
|
|
|
% A list of abbreviated month names. (%b)
|
|
abmon /
|
|
"<U0441><U0456><U0447>"; %sich /
|
|
"<U043B><U044E><U0442>"; %lyut /
|
|
"<U0431><U0435><U0440>"; %ber /
|
|
"<U043A><U0432><U0456>"; %kvi /
|
|
"<U0442><U0440><U0430>"; %tra /
|
|
"<U0447><U0435><U0440>"; %cher /
|
|
"<U043B><U0438><U043F>"; %lyp /
|
|
"<U0441><U0435><U0440>"; %ser /
|
|
"<U0432><U0435><U0440>"; %ver /
|
|
"<U0436><U043E><U0432>"; %zhov /
|
|
"<U043B><U0438><U0441>"; %lys /
|
|
"<U0433><U0440><U0443>" %hru
|
|
|
|
% A list of month names in proper form for calendar, see alt_digits. (%B)
|
|
mon /
|
|
"<U0441><U0456><U0447><U0435><U043D><U044C>"; %sichen` /
|
|
"<U043B><U044E><U0442><U0438><U0439>"; %lyutyj /
|
|
"<U0431><U0435><U0440><U0435><U0437><U0435><U043D><U044C>"; %berezen` /
|
|
"<U043A><U0432><U0456><U0442><U0435><U043D><U044C>"; %kviten` /
|
|
"<U0442><U0440><U0430><U0432><U0435><U043D><U044C>"; %traven` /
|
|
"<U0447><U0435><U0440><U0432><U0435><U043D><U044C>"; %cherven` /
|
|
"<U043B><U0438><U043F><U0435><U043D><U044C>"; %lypen` /
|
|
"<U0441><U0435><U0440><U043F><U0435><U043D><U044C>"; %serpen` /
|
|
"<U0432><U0435><U0440><U0435><U0441><U0435><U043D><U044C>"; %veresen` /
|
|
"<U0436><U043E><U0432><U0442><U0435><U043D><U044C>"; %zhovten` /
|
|
"<U043B><U0438><U0441><U0442><U043E><U043F><U0430><U0434>"; %lystopad /
|
|
"<U0433><U0440><U0443><U0434><U0435><U043D><U044C>" %hruden`
|
|
|
|
% Initially alt_digits was supposed to hold alternative symbols for _digits_,
|
|
% corresponding to %O modified conversion specification.
|
|
% Although in Ukrainian language alternate _names_ are used instead of digits.
|
|
% We'll use this keyword to present a list of month names in proper form for
|
|
% date, see mon. (%Om)
|
|
%
|
|
% This hack is dedicated for months it won't work for other %O* modifiers
|
|
% (weeks, days etc).
|
|
%
|
|
alt_digits "<U0030>"; % digits are starting from zero /
|
|
"<U0441><U0456><U0447><U043D><U044F>"; % sichnya /
|
|
"<U043B><U044E><U0442><U043E><U0433><U043E>"; % lyutoho /
|
|
"<U0431><U0435><U0440><U0435><U0437><U043D><U044F>"; % bereznya /
|
|
"<U043A><U0432><U0456><U0442><U043D><U044F>"; % kvitnya /
|
|
"<U0442><U0440><U0430><U0432><U043D><U044F>"; % travnya /
|
|
"<U0447><U0435><U0440><U0432><U043D><U044F>"; % chervnya /
|
|
"<U043B><U0438><U043F><U043D><U044F>"; % lypnya /
|
|
"<U0441><U0435><U0440><U043F><U043D><U044F>"; % serpnya /
|
|
"<U0432><U0435><U0440><U0435><U0441><U043D><U044F>"; % veresnya /
|
|
"<U0436><U043E><U0432><U0442><U043D><U044F>"; % zhovtnya /
|
|
"<U043B><U0438><U0441><U0442><U043E><U043F><U0430><U0434><U0430>"; % lystopada /
|
|
"<U0433><U0440><U0443><U0434><U043D><U044F>" % hrudnya
|
|
|
|
% Appropriate date representation for date(1).
|
|
date_fmt "<U0025><U0041><U002C><U0020><U0025><U002D><U0064><U0020><U0025><U004F><U006D><U0020><U0025><U0059><U0020><U0025><U0058><U0020><U0025><U007A>"
|
|
|
|
% The appropriate date and time format. (%c)
|
|
d_t_fmt "<U0025><U0061><U002C><U0020><U0025><U0064><U002D><U0025><U0062><U002D><U0025><U0059><U0020><U0025><U0058><U0020><U0025><U007A>"
|
|
|
|
% The appropriate date format. (%x)
|
|
d_fmt "<U0025><U0064><U002E><U0025><U006D><U002E><U0025><U0079>"
|
|
|
|
% The appropriate time format. (%X)
|
|
t_fmt "<U0025><U0054>"
|
|
|
|
% The appropriate time format when using 12h clock format. (%r)
|
|
% If the string is empty the 12-hour format is not supported by in the FDCC-set.
|
|
t_fmt_ampm ""
|
|
|
|
%The appropriate representation of the am and pm strings. (%p)
|
|
%am_pm "<U0434><U043E>";"<U043F><U043E>" % "<U0064><U006F>" ; "<U0070><U006F>" - do obidu/po obidi
|
|
% Empty strings are used to force 24h time format.
|
|
am_pm "";""
|
|
|
|
% Define the first day of the week to be displayed in a calendar.
|
|
% This weekday is relative to the date defined in the <week> keyword.
|
|
first_weekday 2
|
|
|
|
% Example:
|
|
%
|
|
% traven` cherven`
|
|
% pn 5 12 19 26 2 9 16 23 30
|
|
% vt 6 13 20 27 3 10 17 24
|
|
% sr 7 14 21 28 4 11 18 25
|
|
% ct +1+ 8 15 22 29 5 12 19 26
|
|
% pt +2++9+16 23 30 6 13 20 27
|
|
% *sb* *3*10*17*24*31* *7*14*21+28+
|
|
% +nd+ +4+11+18+25+ +1+ +8+15+22+29+
|
|
%
|
|
% + - means red color
|
|
% * - means bold font
|
|
%
|
|
% Direction of the display of calendar dates:
|
|
% 1 - left-right from top
|
|
% 2 - top-down from left
|
|
% 3 - right-left from top
|
|
cal_direction 2
|
|
|
|
END LC_TIME
|
|
|
|
|
|
|
|
LC_PAPER
|
|
copy "i18n"
|
|
END LC_PAPER
|
|
|
|
|
|
|
|
LC_TELEPHONE
|
|
|
|
% Format of a telephone number for international use.
|
|
%
|
|
% a area code without nationwide prefix (prefix is often <0>).
|
|
% A area code including nationwide prefix (prefix is often <0>).
|
|
% l local number (within area code).
|
|
% e extension (to local number)
|
|
% c country code
|
|
% C alternate carrier service code used for dialling abroad
|
|
% t Insert a <space> if the previous descriptor's value was not an empty
|
|
% string; otherwise ignore.
|
|
%
|
|
% Example: +380 44 4908888
|
|
%
|
|
tel_int_fmt "<U002B><U0025><U0063><U0020><U0025><U0061><U0020><U0025><U006C>"
|
|
|
|
% Format of a telephone number for domestic use.
|
|
% See tel_int_fmt.
|
|
%
|
|
% Example: (044) 4908888
|
|
%
|
|
tel_dom_fmt "<U0028><U0025><U0041><U0029><U0020><U0025><U006C>"
|
|
|
|
% Prefix to call international phone numbers.
|
|
int_select "<U0038><U007E><U0031><U0030>"
|
|
|
|
% Prefix used from other countries to dial Ukraine. (%c)
|
|
int_prefix "<U0033><U0038><U0030>"
|
|
|
|
END LC_TELEPHONE
|
|
|
|
|
|
|
|
LC_MEASUREMENT
|
|
copy "i18n"
|
|
END LC_MEASUREMENT
|
|
|
|
|
|
|
|
LC_NAME
|
|
% Examples:
|
|
% Full name: pan Volodymyr Mykolayovych Lisivka
|
|
% Means:
|
|
% pan - mister
|
|
% Volodymyr - the man name (be Lord of the World :) )
|
|
% Mykolayovych - middle name (son of Mykola)
|
|
% Lisivka - family name
|
|
%
|
|
% For documents: Lisivka V. M.
|
|
% Regular name: pan Volodymyr Lisivka
|
|
% Bossname: Volodymyr Mykolayovych
|
|
% For friends: Volodya
|
|
% For family: Vovka (for brothers), Vova(for mother)
|
|
%
|
|
% Begining of official letters:
|
|
% Shanovnyj(-na) Volodymyre Lisivka, ...
|
|
%
|
|
% A salutation to not well known person:
|
|
% pane Lisivka
|
|
%
|
|
% A salutation to boss:
|
|
% Volodymyre Mykolayovychu
|
|
%
|
|
% A salutation to well known person (but not a friend):
|
|
% pane Volodymyre
|
|
%
|
|
% FIXME:
|
|
% to make proper salutation, we must change person name and middle name
|
|
% to proper forms (klychna forma), but this is imposible without
|
|
% ispell-like rules.
|
|
% Now I use regular form, like in Russian (sic!) langauge.
|
|
|
|
% Field descriptors for the <name_fmt> keyword:
|
|
% f - family name
|
|
% F - family name in uppercase
|
|
% g - first given name
|
|
% G - first given initial
|
|
% l - first given name with Latin letters
|
|
% o - other shorter name
|
|
% m - additional given name
|
|
% M - initials for additional given names
|
|
% p - profession
|
|
% s - salutation, such as "<U0044><U006F><U0063><U0074><U006F><U0072>"
|
|
% S - abbreviated salutation, such as "<U004D><U0072><U002E>" or "<U0044><U0072><U002E>"
|
|
% d - salutation, using the FDCC-sets conventions
|
|
% 1 for the name_gen
|
|
% 2 for name_mr
|
|
% 3 for name_mrs
|
|
% 4 for name_miss
|
|
% 5 for name_ms
|
|
% t - if the preceding field descriptor resulted in an empty string,
|
|
% then the empty string, else a <space>.
|
|
name_fmt "<U0025><U0064><U0025><U0074><U0025><U0067><U0025><U0074><U0025><U006D><U0025><U0074><U0025><U0066>" % pan Volodymyr Mykolayovych Lisivka
|
|
|
|
% A salutation valid for all persons
|
|
name_gen "<U0448><U0430><U043D><U043E><U0432><U043D><U0438><U0439><U0028><U002D><U043D><U0430><U0029>" % shanovnyj(-na)
|
|
|
|
% A salutation valid for males
|
|
name_mr "<U043F><U0430><U043D>" % pane
|
|
|
|
% A salutation valid for all females
|
|
name_ms "<U043F><U0430><U043D><U0456>" % pani
|
|
|
|
% A salutation valid for married females
|
|
name_mrs "<U043F><U0430><U043D><U0456>" % pani
|
|
|
|
% A salutation valid for unmarried females
|
|
name_miss "<U043F><U0430><U043D><U043D><U0430>" % panna
|
|
|
|
END LC_NAME
|
|
|
|
|
|
|
|
LC_ADDRESS
|
|
|
|
% Example of address in a city:
|
|
%
|
|
% Volodymyru Lisivtci
|
|
% vul. Pul`uya 9, kv. 21
|
|
% m. L`viv
|
|
% 79005
|
|
%
|
|
% Means:
|
|
% Volodymyru Lisivtci - to what person
|
|
% vul. Pul`uya 9, kv. 21 - street and building, apartment number (or firm name)
|
|
% m. L`viv - city name (m. - misto)
|
|
% 79005 - ZIP code
|
|
%
|
|
% Example of firm address in a city:
|
|
%
|
|
% Volodymyru Lisivtci, firma "<U004D><U0069><U0073><U0074><U0065><U0072><U0069><U0079><U0061>"
|
|
% vul. Naukova 5a, 2-yj poverh, k. 239
|
|
% m. L`viv
|
|
% 79005
|
|
%
|
|
% Means:
|
|
% firma "<U004D><U0069><U0073><U0074><U0065><U0072><U0069><U0079><U0061>" - firm name
|
|
% 2-yj poverh - floor number (ground floor has number one)
|
|
%
|
|
% Example of address in a town:
|
|
%
|
|
% Volodymyru Lisivtci
|
|
% vul. Myru, kv. 19
|
|
% m. Zdolbuniv
|
|
% Rivnens`koji oblasti
|
|
% 00000
|
|
%
|
|
% Means:
|
|
% Rivnens`koji oblasti - in region of city Rivne (oblast`)
|
|
%
|
|
% Example of address in a village:
|
|
%
|
|
% Volodymyru Lisivtci
|
|
% vul. Hranychna 7
|
|
% s. Bohdashiv
|
|
% Zdolbunivs`koho rajonu Rivnens`koji oblasti
|
|
% 00000
|
|
%
|
|
% Means:
|
|
% vul. Hranychna 7 - street and building
|
|
% s. Bohdashiv - village name (s. - selo)
|
|
% Zdolbunivs`koho rajonu - in region of town Zdolbuniv (rajon)
|
|
|
|
|
|
% n Person's name, possibly constructed with the LC_NAME <name_fmt> keyword.
|
|
% a Care of person, or organization.
|
|
% f Firm name.
|
|
% d Department name.
|
|
% b Building name.
|
|
% s Street or block (eg. Japanese) name.
|
|
% h House number or designation.
|
|
% N Insert an <end-of-line> if the previous descriptor s value was not an empty
|
|
% string; otherwise ignore.
|
|
% t Insert a <space> if the previous descriptor s value was not an empty string;
|
|
% otherwise ignore.
|
|
% r Room number, door designation.
|
|
% e Floor number.
|
|
% C Country designation, from the <country_post> keyword.
|
|
% l Local township within town or city
|
|
% z Zip number, postal code.
|
|
% T Town, city.
|
|
% S State, province, or prefecture.
|
|
% c Country, as taken from data record.
|
|
%
|
|
% Each field descriptor may have an <R> after the <%> to specify that the
|
|
% information is taken from a Romanized version string of the
|
|
% entity.
|
|
%
|
|
% BUG: %n escape sequence from ISO/IEC 14652:2002 is not supported by glibc-2.3.2.
|
|
% BUG: %l escape sequence from ISO/IEC 14652:2002 is not supported by glibc-2.3.2.
|
|
%postal_fmt "<U0025><U0061><U0025><U0074><U0025><U006E><U0025><U0074><U0025><U0066><U0025><U0074><U0025><U0064><U0025><U004E><U0025><U0073><U0025><U0074><U0025><U0068><U0025><U0074><U002C><U0020><U0025><U0062><U0025><U0074><U0025><U0065><U0025><U0074><U0025><U0072><U0025><U004E><U0025><U006C><U0025><U0074><U0025><U0054><U0025><U0074><U0025><U0053><U0025><U004E><U0025><U007A><U0025><U004E><U0025><U0063>"
|
|
postal_fmt "<U0025><U0061><U0025><U0074><U0025><U0066><U0025><U0074><U0025><U0064><U0025><U004E><U0025><U0073><U0025><U0074><U0025><U0068><U0025><U0074><U002C><U0020><U0025><U0062><U0025><U0074><U0025><U0065><U0025><U0074><U0025><U0072><U0025><U004E><U0025><U0054><U0025><U0074><U0025><U0053><U0025><U004E><U0025><U007A><U0025><U004E><U0025><U0063>"
|
|
|
|
% Giving:
|
|
%
|
|
% Care of person Person's name Firm Department
|
|
% Street Number, Building Floor Room
|
|
% Town City State
|
|
% Zip
|
|
% Country
|
|
|
|
% Country name in the language of the current document.
|
|
country_name "<U0423><U043A><U0440><U0430><U0457><U043D><U0430>" % Ukrayina
|
|
|
|
% Language name in the language of the current document.
|
|
lang_name "<U0443><U043A><U0440><U0430><U0457><U043D><U0441><U044C><U043A><U0430>" % ukrayins`ka [mova]
|
|
|
|
% Abbreviation of the country, see CERT-MAILCODES. (%c)
|
|
country_post "<U0055><U0041>"
|
|
|
|
% Two-letter abbreviation of the country, see ISO 3166.
|
|
country_ab2 "<U0055><U0041>"
|
|
|
|
% Three-letter abbreviation of the country, see ISO 3166.
|
|
country_ab3 "<U0055><U004B><U0052>"
|
|
|
|
% Numeric country code, see ISO 3166.
|
|
country_num 804
|
|
|
|
% Code for country car number.
|
|
country_car "<U0055><U0041>"
|
|
|
|
% ISBN code, for books.
|
|
country_isbn 966
|
|
|
|
% Two-letter abbreviation of the language, see ISO 639.
|
|
lang_ab "<U0075><U006B>"
|
|
% lang_ab2 "<U0075><U006B>" % ISO/IEC 14652:2002
|
|
|
|
% Three-letter abbreviation of the language, see ISO 639-2.
|
|
lang_term "<U0075><U006B><U0072>"
|
|
% lang_ab3_term "<U0075><U006B><U0072>" % ISO/IEC 14652:2002
|
|
|
|
% Three-letter abbreviation of the language for library use, see ISO 639-2.
|
|
lang_lib "<U0075><U006B><U0072>"
|
|
% lang_ab3_lib "<U0075><U006B><U0072>" % ISO/IEC 14652:2002
|
|
|
|
END LC_ADDRESS
|