mirror of
https://git.savannah.gnu.org/git/emacs.git
synced 2024-12-26 10:49:33 +00:00
651 lines
21 KiB
EmacsLisp
651 lines
21 KiB
EmacsLisp
;;; european.el --- support for European languages -*- coding: iso-2022-7bit; -*-
|
||
|
||
;; Copyright (C) 1997-1998, 2000-2013 Free Software Foundation, Inc.
|
||
;; Copyright (C) 1995, 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004,
|
||
;; 2005, 2006, 2007, 2008, 2009, 2010, 2011
|
||
;; National Institute of Advanced Industrial Science and Technology (AIST)
|
||
;; Registration Number H14PRO021
|
||
;; Copyright (C) 2003
|
||
;; National Institute of Advanced Industrial Science and Technology (AIST)
|
||
;; Registration Number H13PRO009
|
||
|
||
;; Keywords: multilingual, European
|
||
|
||
;; This file is part of GNU Emacs.
|
||
|
||
;; GNU Emacs is free software: 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 3 of the License, or
|
||
;; (at your option) any later version.
|
||
|
||
;; GNU Emacs is distributed in the hope that it will be useful,
|
||
;; but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||
;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||
;; GNU General Public License for more details.
|
||
|
||
;; You should have received a copy of the GNU General Public License
|
||
;; along with GNU Emacs. If not, see <http://www.gnu.org/licenses/>.
|
||
|
||
;;; Commentary:
|
||
|
||
;; For European scripts, all the ISO Latin character sets are
|
||
;; supported, along with various others.
|
||
|
||
;;; Code:
|
||
|
||
;; Latin-1 (ISO-8859-1)
|
||
|
||
(set-language-info-alist
|
||
"Latin-1" '((charset iso-8859-1)
|
||
(coding-system iso-latin-1 iso-latin-9 windows-1252)
|
||
(coding-priority iso-latin-1)
|
||
(nonascii-translation . iso-8859-1)
|
||
(unibyte-display . iso-latin-1)
|
||
(input-method . "latin-1-prefix")
|
||
(sample-text
|
||
. "Hello, Hej, Tere, Hei, Bonjour, Gr$(D+d)N(B Gott, Ciao, $(D"B(BHola!")
|
||
(documentation . "\
|
||
This language environment is a generic one for the Latin-1 (ISO-8859-1)
|
||
character set which supports the following European languages:
|
||
Albanian, Basque, Breton, Catalan, Danish, Dutch, English, Faeroese,
|
||
Finnish, French (with restrictions -- see Latin-9), Frisian, Galician,
|
||
German, Greenlandic, Icelandic, Irish Gaelic (new orthography),
|
||
Italian, Latin, Luxemburgish, Norwegian, Portuguese, Rhaeto-Romanic,
|
||
Scottish Gaelic, Spanish, and Swedish.
|
||
We also have specific language environments for the following languages:
|
||
For Dutch, \"Dutch\".
|
||
For German, \"German\".
|
||
For French, \"French\".
|
||
For Italian, \"Italian\".
|
||
For Slovenian, \"Slovenian\".
|
||
For Spanish, \"Spanish\".
|
||
|
||
Latin-1 also covers several written languages outside Europe, including
|
||
Indonesian/Malay, Tagalog (Philippines), Swahili and Afrikaans."))
|
||
'("European"))
|
||
|
||
|
||
;; Latin-2 (ISO-8859-2)
|
||
|
||
(define-coding-system 'iso-latin-2
|
||
"ISO 2022 based 8-bit encoding for Latin-2 (MIME:ISO-8859-2)."
|
||
:coding-type 'charset
|
||
:mnemonic ?2
|
||
:charset-list '(iso-8859-2)
|
||
:mime-charset 'iso-8859-2)
|
||
|
||
(define-coding-system-alias 'iso-8859-2 'iso-latin-2)
|
||
(define-coding-system-alias 'latin-2 'iso-latin-2)
|
||
|
||
(set-language-info-alist
|
||
"Latin-2" '((charset iso-8859-2)
|
||
(coding-system iso-latin-2 windows-1250)
|
||
(coding-priority iso-latin-2)
|
||
(nonascii-translation . iso-8859-2)
|
||
(unibyte-display . iso-latin-2)
|
||
(input-method . "latin-2-prefix")
|
||
(documentation . "\
|
||
This language environment is a generic one for the Latin-2 (ISO-8859-2)
|
||
character set which supports the following languages:
|
||
Albanian, Czech, English, German, Hungarian, Polish, Romanian,
|
||
Serbo-Croatian or Croatian, Slovak, Slovene, Sorbian (upper and lower),
|
||
and Swedish.
|
||
We also have specific language environments for the following languages:
|
||
For Czech, \"Czech\".
|
||
For Croatian, \"Croatian\".
|
||
For Polish, \"Polish\".
|
||
For Romanian, \"Romanian\".
|
||
For Slovak, \"Slovak\"."))
|
||
'("European"))
|
||
|
||
|
||
;; Latin-3 (ISO-8859-3)
|
||
|
||
(define-coding-system 'iso-latin-3
|
||
"ISO 2022 based 8-bit encoding for Latin-3 (MIME:ISO-8859-3)."
|
||
:coding-type 'charset
|
||
:mnemonic ?3
|
||
:charset-list '(iso-8859-3)
|
||
:mime-charset 'iso-8859-3)
|
||
|
||
(define-coding-system-alias 'iso-8859-3 'iso-latin-3)
|
||
(define-coding-system-alias 'latin-3 'iso-latin-3)
|
||
|
||
(set-language-info-alist
|
||
"Latin-3" '((charset iso-8859-3)
|
||
(coding-system iso-latin-3)
|
||
(coding-priority iso-latin-3)
|
||
(nonascii-translation . iso-8859-3)
|
||
(unibyte-display . iso-latin-3)
|
||
(input-method . "latin-3-prefix")
|
||
(documentation . "\
|
||
These languages are supported with the Latin-3 (ISO-8859-3) character set:
|
||
Afrikaans, Catalan, Dutch, English, Esperanto, French, Galician,
|
||
German, Italian, Maltese, Spanish, and Turkish."))
|
||
'("European"))
|
||
|
||
|
||
;; Latin-4 (ISO-8859-4)
|
||
|
||
(define-coding-system 'iso-latin-4
|
||
"ISO 2022 based 8-bit encoding for Latin-4 (MIME:ISO-8859-4)."
|
||
:coding-type 'charset
|
||
:mnemonic ?4
|
||
:charset-list '(iso-8859-4)
|
||
:mime-charset 'iso-8859-4)
|
||
|
||
(define-coding-system-alias 'iso-8859-4 'iso-latin-4)
|
||
(define-coding-system-alias 'latin-4 'iso-latin-4)
|
||
|
||
(set-language-info-alist
|
||
"Latin-4" '((charset iso-8859-4)
|
||
(coding-system iso-8859-4)
|
||
(coding-priority iso-8859-4)
|
||
(nonascii-translation . iso-8859-4)
|
||
(unibyte-display . iso-8859-4)
|
||
(input-method . "latin-4-postfix")
|
||
(documentation . "\
|
||
These languages are supported with the Latin-4 (ISO-8859-4) character set:
|
||
Danish, English, Estonian, Finnish, German, Greenlandic, Latvian,
|
||
Lithuanian, Norwegian, and Sami."))
|
||
'("European"))
|
||
|
||
|
||
;; Latin-5 (ISO-8859-9)
|
||
|
||
(define-coding-system 'iso-latin-5
|
||
"ISO 2022 based 8-bit encoding for Latin-5 (MIME:ISO-8859-9)."
|
||
:coding-type 'charset
|
||
:mnemonic ?9
|
||
:charset-list '(iso-8859-9)
|
||
:mime-charset 'iso-8859-9)
|
||
|
||
(define-coding-system-alias 'iso-8859-9 'iso-latin-5)
|
||
(define-coding-system-alias 'latin-5 'iso-latin-5)
|
||
|
||
(set-language-info-alist
|
||
"Latin-5" '((charset iso-8859-9)
|
||
(coding-system iso-latin-5)
|
||
(coding-priority iso-latin-5)
|
||
(nonascii-translation . iso-8859-9)
|
||
(unibyte-display . iso-latin-5)
|
||
(input-method . "latin-5-postfix")
|
||
(documentation . "Support for Latin-5.\
|
||
See also the Turkish environment."))
|
||
'("European"))
|
||
|
||
|
||
;; Latin-6 (ISO-8859-10)
|
||
|
||
(define-coding-system 'iso-latin-6
|
||
"ISO 2022 based 8-bit encoding for Latin-6 (MIME:ISO-8859-10)."
|
||
:coding-type 'charset
|
||
:mnemonic ?9
|
||
:charset-list '(iso-8859-10)
|
||
:mime-charset 'iso-8859-10)
|
||
|
||
(define-coding-system-alias 'iso-8859-10 'iso-latin-6)
|
||
(define-coding-system-alias 'latin-6 'iso-latin-6)
|
||
|
||
(set-language-info-alist
|
||
"Latin-6" '((charset iso-8859-10)
|
||
(coding-system iso-latin-6)
|
||
(coding-priority iso-latin-6)
|
||
(nonascii-translation . iso-8859-10)
|
||
(unibyte-display . iso-latin-6)
|
||
;; Fixme: input method.
|
||
(documentation . "Support for generic Latin-6 (Northern European)."))
|
||
'("European"))
|
||
|
||
|
||
;; Latin-7 (ISO-8859-13)
|
||
|
||
(define-coding-system 'iso-latin-7
|
||
"ISO 2022 based 8-bit encoding for Latin-7 (MIME:ISO-8859-13)."
|
||
:coding-type 'charset
|
||
:mnemonic ?9
|
||
:charset-list '(iso-8859-13)
|
||
:mime-charset 'iso-8859-13)
|
||
|
||
(define-coding-system-alias 'iso-8859-13 'iso-latin-7)
|
||
(define-coding-system-alias 'latin-7 'iso-latin-7)
|
||
|
||
(set-language-info-alist
|
||
"Latin-7" '((charset iso-8859-13)
|
||
(coding-system iso-latin-7)
|
||
(coding-priority iso-latin-7)
|
||
(nonascii-translation . iso-8859-13)
|
||
(unibyte-display . iso-latin-7)
|
||
;; Fixme: input method.
|
||
(documentation . "Support for generic Latin-7 (Baltic Rim)."))
|
||
'("European"))
|
||
|
||
;; Latin-8 (ISO-8859-14)
|
||
|
||
(define-coding-system 'iso-latin-8
|
||
"ISO 2022 based 8-bit encoding for Latin-8 (MIME:ISO-8859-14)."
|
||
:coding-type 'charset
|
||
;; `W' for `Welsh', since `C' for `Celtic' is taken.
|
||
:mnemonic ?W
|
||
:charset-list '(iso-8859-14)
|
||
:mime-charset 'iso-8859-14)
|
||
|
||
(define-coding-system-alias 'iso-8859-14 'iso-latin-8)
|
||
(define-coding-system-alias 'latin-8 'iso-latin-8)
|
||
|
||
(set-language-info-alist
|
||
"Latin-8" '((charset iso-8859-14)
|
||
(coding-system iso-latin-8)
|
||
(coding-priority iso-latin-8)
|
||
(nonascii-translation . iso-8859-14)
|
||
(unibyte-display . iso-latin-8)
|
||
(input-method . "latin-8-prefix")
|
||
;; Fixme: Welsh/Ga{e}lic greetings
|
||
(sample-text . ",_"(B $(D+q(B $(D*t(B")
|
||
(documentation . "\
|
||
This language environment is a generic one for the Latin-8 (ISO-8859-14)
|
||
character set which supports the Celtic languages, including those not
|
||
covered by other ISO-8859 character sets:
|
||
Welsh, Manx Gaelic and Irish Gaelic (old orthography)."))
|
||
'("European"))
|
||
|
||
;; Latin-9 (ISO-8859-15)
|
||
|
||
(define-coding-system 'iso-latin-9
|
||
"ISO 2022 based 8-bit encoding for Latin-9 (MIME:ISO-8859-15)."
|
||
:coding-type 'charset
|
||
;; `0' for `Latin-0'
|
||
:mnemonic ?0
|
||
:charset-list '(iso-8859-15)
|
||
:mime-charset 'iso-8859-15)
|
||
|
||
(define-coding-system-alias 'iso-8859-15 'iso-latin-9)
|
||
(define-coding-system-alias 'latin-9 'iso-latin-9)
|
||
(define-coding-system-alias 'latin-0 'iso-latin-9)
|
||
|
||
(set-language-info-alist
|
||
"Latin-9" '((charset iso-8859-15)
|
||
(coding-system iso-latin-9)
|
||
(coding-priority iso-latin-9)
|
||
(nonascii-translation . iso-8859-15)
|
||
(unibyte-display . iso-latin-9)
|
||
(input-method . "latin-9-prefix")
|
||
(sample-text
|
||
. "AVE. $(D*^+^*v+v)-)M*s(B $(Q)!(B")
|
||
(documentation . "\
|
||
This language environment is a generic one for the Latin-9 (ISO-8859-15)
|
||
character set which supports the same languages as Latin-1 with the
|
||
addition of the Euro sign and some additional French and Finnish letters.
|
||
Latin-9 is sometimes nicknamed `Latin-0'."))
|
||
'("European"))
|
||
|
||
(set-language-info-alist
|
||
"Esperanto" '((tutorial . "TUTORIAL.eo")
|
||
(charset iso-8859-3)
|
||
(coding-system iso-latin-3)
|
||
(coding-priority iso-latin-3)
|
||
(nonascii-translation . latin-iso8859-3)
|
||
(unibyte-syntax . "latin-3")
|
||
(unibyte-display . iso-latin-3)
|
||
(input-method . "latin-3-prefix")
|
||
(documentation . "Support for Esperanto with ISO-8859-3 character set."))
|
||
'("European"))
|
||
|
||
|
||
(define-coding-system 'windows-1250
|
||
"windows-1250 (Central European) encoding (MIME: WINDOWS-1250)"
|
||
:coding-type 'charset
|
||
:mnemonic ?*
|
||
:charset-list '(windows-1250)
|
||
:mime-charset 'windows-1250)
|
||
(define-coding-system-alias 'cp1250 'windows-1250)
|
||
|
||
(define-coding-system 'windows-1252
|
||
"windows-1252 (Western European) encoding (MIME: WINDOWS-1252)"
|
||
:coding-type 'charset
|
||
:mnemonic ?*
|
||
:charset-list '(windows-1252)
|
||
:mime-charset 'windows-1252)
|
||
(define-coding-system-alias 'cp1252 'windows-1252)
|
||
|
||
(define-coding-system 'windows-1254
|
||
"windows-1254 (Turkish) encoding (MIME: WINDOWS-1254)"
|
||
:coding-type 'charset
|
||
:mnemonic ?*
|
||
:charset-list '(windows-1254)
|
||
:mime-charset 'windows-1254)
|
||
(define-coding-system-alias 'cp1254 'windows-1254)
|
||
|
||
(define-coding-system 'windows-1257
|
||
"windows-1257 (Baltic) encoding (MIME: WINDOWS-1257)"
|
||
:coding-type 'charset
|
||
:mnemonic ?*
|
||
:charset-list '(windows-1257)
|
||
:mime-charset 'windows-1257)
|
||
(define-coding-system-alias 'cp1257 'windows-1257)
|
||
|
||
(define-coding-system 'cp775
|
||
"DOS codepage 775 (PC Baltic, MS-DOS Baltic Rim)"
|
||
:coding-type 'charset
|
||
:mnemonic ?D
|
||
:charset-list '(cp775)
|
||
:mime-charset 'cp775)
|
||
(define-coding-system-alias 'ibm775 'cp775)
|
||
|
||
(define-coding-system 'cp850
|
||
"DOS codepage 850 (Western European)"
|
||
:coding-type 'charset
|
||
:mnemonic ?D
|
||
:charset-list '(cp850)
|
||
:mime-charset 'cp850)
|
||
(define-coding-system-alias 'ibm850 'cp850)
|
||
|
||
(define-coding-system 'cp852
|
||
"DOS codepage 852 (Slavic)"
|
||
:coding-type 'charset
|
||
:mnemonic ?D
|
||
:charset-list '(cp852)
|
||
:mime-charset 'cp852)
|
||
(define-coding-system-alias 'ibm852 'cp852)
|
||
|
||
(define-coding-system 'cp857
|
||
"DOS codepage 857 (Turkish)"
|
||
:coding-type 'charset
|
||
:mnemonic ?D
|
||
:charset-list '(cp857)
|
||
:mime-charset 'cp857)
|
||
(define-coding-system-alias 'ibm857 'cp857)
|
||
|
||
(define-coding-system 'cp858
|
||
"Codepage 858 (Multilingual Latin I + Euro)"
|
||
:coding-type 'charset
|
||
:mnemonic ?D
|
||
:charset-list '(cp858)
|
||
:mime-charset 'cp858)
|
||
|
||
(define-coding-system 'cp860
|
||
"DOS codepage 860 (Portuguese)"
|
||
:coding-type 'charset
|
||
:mnemonic ?D
|
||
:charset-list '(cp860)
|
||
:mime-charset 'cp860)
|
||
(define-coding-system-alias 'ibm860 'cp860)
|
||
|
||
(define-coding-system 'cp861
|
||
"DOS codepage 861 (Icelandic)"
|
||
:coding-type 'charset
|
||
:mnemonic ?D
|
||
:charset-list '(cp861)
|
||
:mime-charset 'cp861)
|
||
(define-coding-system-alias 'ibm861 'cp861)
|
||
|
||
(define-coding-system 'cp863
|
||
"DOS codepage 863 (French Canadian)"
|
||
:coding-type 'charset
|
||
:mnemonic ?D
|
||
:charset-list '(cp863)
|
||
:mime-charset 'cp863)
|
||
(define-coding-system-alias 'ibm863 'cp863)
|
||
|
||
(define-coding-system 'cp865
|
||
"DOS codepage 865 (Norwegian/Danish)"
|
||
:coding-type 'charset
|
||
:mnemonic ?D
|
||
:charset-list '(cp865)
|
||
:mime-charset 'cp865)
|
||
(define-coding-system-alias 'ibm865 'cp865)
|
||
|
||
(define-coding-system 'cp437
|
||
"DOS codepage 437"
|
||
:coding-type 'charset
|
||
:mnemonic ?D
|
||
:charset-list '(cp437)
|
||
:mime-charset 'cp437)
|
||
(define-coding-system-alias 'ibm437 'cp437)
|
||
|
||
(set-language-info-alist
|
||
"Dutch" '((tutorial . "TUTORIAL.nl")
|
||
(charset iso-8859-1)
|
||
(coding-system iso-latin-1 iso-latin-9)
|
||
(coding-priority iso-latin-1)
|
||
(nonascii-translation . iso-8859-1)
|
||
(unibyte-display . iso-latin-1)
|
||
(input-method . "dutch")
|
||
(sample-text . "Er is een aantal manieren waarop je dit kan doen")
|
||
(documentation . "\
|
||
This language environment is almost the same as Latin-1,
|
||
but it selects the Dutch tutorial and input method."))
|
||
'("European"))
|
||
|
||
(set-language-info-alist
|
||
"German" '((tutorial . "TUTORIAL.de")
|
||
(charset iso-8859-1)
|
||
(coding-system iso-latin-1 iso-latin-9)
|
||
(coding-priority iso-latin-1)
|
||
(nonascii-translation . iso-8859-1)
|
||
(input-method . "german-postfix")
|
||
(unibyte-display . iso-latin-1)
|
||
(sample-text . "\
|
||
German (Deutsch Nord) Guten Tag
|
||
German (Deutsch S$(D+d(Bd) Gr$(D+d)N(B Gott")
|
||
(documentation . "\
|
||
This language environment is almost the same as Latin-1,
|
||
but sets the default input method to \"german-postfix\".
|
||
Additionally, it selects the German tutorial."))
|
||
'("European"))
|
||
|
||
(set-language-info-alist
|
||
"French" '((tutorial . "TUTORIAL.fr")
|
||
(charset iso-8859-1)
|
||
(coding-system iso-latin-1 iso-latin-9)
|
||
(coding-priority iso-latin-1)
|
||
(nonascii-translation . iso-8859-1)
|
||
(unibyte-display . iso-latin-1)
|
||
(input-method . "latin-1-prefix")
|
||
(sample-text . "French (Fran$(D+.(Bais) Bonjour, Salut")
|
||
(documentation . "\
|
||
This language environment is almost the same as Latin-1,
|
||
but it selects the French tutorial and input method."))
|
||
'("European"))
|
||
|
||
(set-language-info-alist
|
||
"Italian" '((tutorial . "TUTORIAL.it")
|
||
(charset iso-8859-1)
|
||
(coding-system iso-latin-1 iso-latin-9)
|
||
(coding-priority iso-latin-1)
|
||
(nonascii-translation . iso-8859-1)
|
||
(unibyte-display . iso-latin-1)
|
||
(input-method . "italian-postfix")
|
||
(sample-text . "Salve, ciao!")
|
||
(documentation . "\
|
||
This language environment is almost the same as Latin-1,
|
||
but sets the default input method to \"italian-postfix\".
|
||
Additionally, it selects the Italian tutorial."))
|
||
'("European"))
|
||
|
||
(set-language-info-alist
|
||
"Slovenian" '((charset iso-8859-2)
|
||
(coding-system . (iso-8859-2 windows-1250))
|
||
(coding-priority . (iso-8859-2))
|
||
(nonascii-translation . iso-8859-2)
|
||
(input-method . "slovenian")
|
||
(unibyte-display . iso-8859-2)
|
||
(tutorial . "TUTORIAL.sl")
|
||
(sample-text . "$(D*v(Belimo vam uspe$(D+^(Ben dan!")
|
||
(documentation . "\
|
||
This language environment is almost the same as Latin-2,
|
||
but it selects the Slovenian tutorial and input method."))
|
||
'("European"))
|
||
|
||
(set-language-info-alist
|
||
"Spanish" '((tutorial . "TUTORIAL.es")
|
||
(charset iso-8859-1)
|
||
(coding-system iso-latin-1 iso-latin-9)
|
||
(coding-priority iso-latin-1)
|
||
(input-method . "spanish-postfix")
|
||
(nonascii-translation . iso-8859-1)
|
||
(unibyte-display . iso-latin-1)
|
||
(sample-text . "Spanish (Espa$(D+P(Bol) $(D"B(BHola!")
|
||
(documentation . "\
|
||
This language environment is almost the same as Latin-1,
|
||
but it sets the default input method to \"spanish-postfix\",
|
||
and it selects the Spanish tutorial."))
|
||
'("European"))
|
||
|
||
;; For Turkish, the character set ISO-8859-9 (Latin-5) is used. But,
|
||
;; before the introduction of ISO-8859-9 in 1988, ISO-8859-3 (Latin-3)
|
||
;; was used for Turkish. Those who use Latin-3 for Turkish should use
|
||
;; "Latin-3" language environment.
|
||
|
||
(set-language-info-alist
|
||
"Turkish" '((charset iso-8859-9)
|
||
(coding-system iso-latin-5 windows-1254 iso-latin-3)
|
||
(coding-priority iso-latin-5)
|
||
(nonascii-translation . iso-8859-9)
|
||
(unibyte-display . iso-latin-5)
|
||
(input-method . "turkish-postfix")
|
||
(sample-text . "Turkish (T$(D+d(Brk$(D+.(Be) Merhaba")
|
||
(setup-function . turkish-case-conversion-enable)
|
||
(setup-function . turkish-case-conversion-disable)
|
||
(documentation . "Support for Turkish.
|
||
Differs from the Latin-5 environment in using the `turkish-postfix' input
|
||
method and applying Turkish case rules for the characters i, I, $(D)E(B, $(D*D(B.")))
|
||
|
||
(defun turkish-case-conversion-enable ()
|
||
"Set up Turkish case conversion of `i' and `I' into `$(D*D(B' and `$(D)E(B'."
|
||
(let ((table (standard-case-table)))
|
||
(set-case-syntax-pair ?$(D*D(B ?i table)
|
||
(set-case-syntax-pair ?I ?$(D)E(B table)))
|
||
|
||
(defun turkish-case-conversion-disable ()
|
||
"Set up normal (non-Turkish) case conversion of `i' into `I'."
|
||
(let ((table (standard-case-table)))
|
||
(set-case-syntax-pair ?I ?i table)
|
||
(set-case-syntax ?$(D*D(B "w" table)
|
||
(set-case-syntax ?$(D)E(B "w" table)))
|
||
|
||
;; Polish ISO 8859-2 environment.
|
||
;; Maintainer: Wlodek Bzyl <matwb@univ.gda.pl>
|
||
;; Keywords: multilingual, Polish
|
||
|
||
(set-language-info-alist
|
||
"Polish" '((charset iso-8859-2)
|
||
(coding-system iso-8859-2 windows-1250)
|
||
(coding-priority iso-8859-2)
|
||
(input-method . "polish-slash")
|
||
(nonascii-translation . iso-8859-2)
|
||
(unibyte-display . iso-8859-2)
|
||
(tutorial . "TUTORIAL.pl")
|
||
(sample-text . "P$(D+Q(Bjd$(D+u(B, ki$(D+M(B-$(D+w(Be t$(D+8(B chmurno$(D+\++(B w g$(D)H+((Bb flaszy")
|
||
(documentation . t))
|
||
'("European"))
|
||
|
||
(set-language-info-alist
|
||
"Welsh" `((coding-system utf-8 latin-8) ; the input method is Unicode-based
|
||
(coding-priority utf-8 latin-8)
|
||
(nonascii-translation . iso-8859-14)
|
||
(input-method . "welsh")
|
||
(documentation . "Support for Welsh, using Unicode."))
|
||
'("European"))
|
||
|
||
(set-language-info-alist
|
||
"Latin-6" `((coding-system latin-6)
|
||
(coding-priority latin-6)
|
||
(nonascii-translation . ,(get 'decode-iso-latin-6 'translation-table))
|
||
(input-method . "latin-prefix")
|
||
(features code-pages)
|
||
(documentation . "Support for Latin-6."))
|
||
'("European"))
|
||
|
||
(set-language-info-alist
|
||
"Latin-7" `((coding-system latin-7)
|
||
(coding-priority latin-7)
|
||
(nonascii-translation . iso-8859-13)
|
||
(input-method . "latin-prefix")
|
||
(documentation . "Support for Latin-7, e.g. Latvian, Lithuanian."))
|
||
'("European"))
|
||
|
||
(set-language-info-alist
|
||
"Lithuanian" `((coding-system latin-7 windows-1257)
|
||
(coding-priority latin-7)
|
||
(nonascii-translation . iso-8859-13)
|
||
(input-method . "lithuanian-keyboard")
|
||
(documentation . "Support for Lithuanian."))
|
||
'("European"))
|
||
|
||
(set-language-info-alist
|
||
"Latvian" `((coding-system latin-7 windows-1257)
|
||
(coding-priority latin-7)
|
||
(nonascii-translation . iso-8859-13)
|
||
(input-method . "latvian-keyboard")
|
||
(documentation . "Support for Latvian."))
|
||
'("European"))
|
||
|
||
(set-language-info-alist
|
||
"Swedish" '((tutorial . "TUTORIAL.sv")
|
||
(charset iso-8859-1)
|
||
(coding-system iso-latin-1)
|
||
(coding-priority iso-latin-1)
|
||
(nonascii-translation . iso-8859-1)
|
||
(unibyte-display . iso-latin-1)
|
||
(sample-text . "Goddag Hej")
|
||
(documentation . "Support for Swedish"))
|
||
'("European"))
|
||
|
||
(set-language-info-alist
|
||
"Croatian" '((charset iso-8859-2)
|
||
(coding-system iso-8859-2)
|
||
(coding-priority iso-8859-2)
|
||
(input-method . "croatian")
|
||
(nonascii-translation . iso-8859-2)
|
||
(unibyte-display . iso-8859-2)
|
||
(documentation . "Support for Croatian with Latin-2 encoding."))
|
||
'("European"))
|
||
|
||
(set-language-info-alist
|
||
"Brazilian Portuguese" '((tutorial . "TUTORIAL.pt_BR")
|
||
(charset iso-8859-1)
|
||
(coding-system iso-latin-1 iso-latin-9)
|
||
(coding-priority iso-latin-1)
|
||
(nonascii-translation . iso-8859-1)
|
||
(unibyte-display . iso-8859-1)
|
||
(input-method . "latin-1-prefix")
|
||
(sample-text . "Oi")
|
||
(documentation . "Support for Brazilian Portuguese."))
|
||
'("European"))
|
||
|
||
|
||
(define-coding-system 'mac-roman
|
||
"Mac Roman Encoding (MIME:MACINTOSH)."
|
||
:coding-type 'charset
|
||
:mnemonic ?M
|
||
:charset-list '(mac-roman)
|
||
:mime-charset 'macintosh)
|
||
(define-coding-system-alias 'macintosh 'mac-roman)
|
||
|
||
(define-coding-system 'next
|
||
"NeXTstep encoding"
|
||
:coding-type 'charset
|
||
:mnemonic ?*
|
||
:charset-list '(next)
|
||
:mime-charset 'next)
|
||
|
||
(define-coding-system 'hp-roman8
|
||
"Hewlet-Packard roman-8 encoding (MIME:ROMAN-8)"
|
||
:coding-type 'charset
|
||
:mnemonic ?*
|
||
:charset-list '(hp-roman8)
|
||
:mime-charset 'hp-roman8)
|
||
(define-coding-system-alias 'roman8 'hp-roman8)
|
||
|
||
(define-coding-system 'adobe-standard-encoding
|
||
"Adobe `standard' encoding for PostScript"
|
||
:coding-type 'charset
|
||
:mnemonic ?*
|
||
:charset-list '(adobe-standard-encoding)
|
||
:mime-charset 'adobe-standard-encoding)
|
||
|
||
(provide 'european)
|
||
|
||
;;; european.el ends here
|