Mercurial > hg > xemacs-beta
view lisp/mule/tibetan.el @ 844:047d37eb70d7
[xemacs-hg @ 2002-05-16 13:30:23 by ben]
ui fixes for things that were bothering me
bytecode.c, editfns.c, lisp.h, lread.c: Fix save-restriction to use markers rather than pseudo-markers
(integers representing the amount of text on either side of the
region). That way, all inserts are handled correctly, not just
those inside old restriction.
Add buffer argument to save_restriction_save().
process.c: Clean up very dirty and kludgy code that outputs into a buffer --
use proper unwind protects, etc.
font-lock.c: Do save-restriction/widen around the function -- otherwise, incorrect
results will ensue when a buffer has been narrowed before a call to
e.g. `buffer-syntactic-context' -- something that happens quite often.
fileio.c: Look for a handler for make-temp-name.
window.c, winslots.h: Try to solve this annoying problem: have two frames displaying the
buffer, in different places; in one, temporarily switch away to
another buffer and then back -- and you've lost your position;
it's reset to the other one in the other frame. My current
solution involves window-level caches of buffers and points (also
a cache for window-start); when set-window-buffer is called, it
looks to see if the buffer was previously visited in the window,
and if so, uses the most recent point at that time. (It's a
marker, so it handles changes.)
#### Note: It could be argued that doing it on the frame level
would be better -- e.g. if you visit a buffer temporarily through
a grep, and then go back to that buffer, you presumably want the
grep's position rather than some previous position provided
everything was in the same frame, even though the grep was in
another window in the frame. However, doing it on the frame level
fails when you have two windows on the same frame. Perhaps we
keep both a window and a frame cache, and use the frame cache if
there are no other windows on the frame showing the buffer, else
the window's cache? This is probably something to be configurable
using a specifier. Suggestions please please please?
window.c: Clean up a bit code that deals with the annoyance of window-point
vs. point.
dialog.el: Function to ask a
multiple-choice question, automatically choosing a dialog box or
minibuffer representation as necessary. Generalized version of
yes-or-no-p, y-or-n-p.
files.el: Use get-user-response to ask "yes/no/diff" question when recovering.
"diff" means that a diff is displayed between the current file and the
autosave. (Converts/deconverts escape-quoted as necessary. No more
complaints from you, Mr. Turnbull!) One known problem: when a dialog
is used, it's modal, so you can't scroll the diff. Will fix soon.
lisp-mode.el: If we're filling a string, don't treat semicolon as a comment,
which would give very unfriendly results.
Uses `buffer-syntactic-context'.
simple.el: all changes back to the beginning. (Useful if you've saved the file
in the middle of the changes.)
simple.el: Add option kill-word-into-kill-ring, which controls whether words
deleted with kill-word, backward-kill-word, etc. are "cut" into the
kill ring, or "cleared" into nothingness. (My preference is the
latter, by far. I'd almost go so far as suggesting we make it the
default, as you can always select a word and then cut it if you want
it cut.)
menubar-items.el: Add option corresponding to kill-word-into-kill-ring.
author | ben |
---|---|
date | Thu, 16 May 2002 13:30:58 +0000 |
parents | 2923009caf47 |
children | 98af8a976fc3 |
line wrap: on
line source
;;; tibetan.el --- support for Tibetan language -*- coding: iso-2022-7bit; -*- ;; Copyright (C) 1997 Electrotechnical Laboratory, JAPAN. ;; Copyright (C) 2002 Ben Wing. ;; Licensed to the Free Software Foundation. ;; Keywords: multilingual, Tibetan ;; This file is part of XEmacs. ;; XEmacs 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 2, or (at your option) ;; any later version. ;; XEmacs 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 XEmacs; see the file COPYING. If not, write to the Free ;; Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA ;; 02111-1307, USA. ;;; Synched up with: Emacs 21.1 (language/tibetan.el). ;; Author: Toru TOMABECHI, <Toru.Tomabechi@orient.unil.ch> ;; Created: Feb. 17. 1997 ;; History: ;; 1997.03.13 Modification for special signs and punctuations. ;;; Commentary: ;;; Code: ;;; Tibetan Character set. ;;; \x2130 -- \x234a is a subset of Unicode v.2 \x0f00 - \x0fb9 ;;; with a slight modification. And there are some subjoined ;;; consonants which are not specified in Unicode. ;;; I hope I can add missing characters later. ;;; ;;; 00 01 02 03 04 05 06 07 08 09 0A 0B 0C 0D 0E 0F ;;;2120 // $(7!!(B $(7!"(B $(7!#(B $(7!$(B $(7!%(B $(7!&(B $(7!'(B $(7!((B $(7!)(B $(7!*(B $(7!+(B $(7!,(B $(7!-(B $(7!.(B $(7!/(B ; obsolete glyphs (2123-5) ;;;2130 $(7!0(B $(7!1(B $(7!2(B $(7!3(B $(7!4(B $(7!5(B $(7!6(B $(7!7(B $(7!8(B $(7!9(B $(7!:(B $(7!;(B $(7!<(B $(7!=(B $(7!>(B $(7!?(B ; Punctuations, ;;;2140 $(7!@(B $(7!A(B $(7!B(B $(7!C(B $(7!D(B $(7!E(B $(7!F(B $(7!G(B $(7!H(B $(7!I(B $(7!J(B $(7!K(B $(7!L(B $(7!M(B $(7!N(B $(7!O(B ; Digits and ;;;2150 $(7!P(B $(7!Q(B $(7!R(B $(7!S(B $(7!T(B $(7!U(B $(7!V(B $(7!W(B $(7!X(B $(7!Y(B $(7!Z(B $(7(B $(7"*(B $(7"+(B $(7",(B $(7"-(B $(7".(B $(7"/(B ; Base consonants ;;;2230 $(7"0(B $(7"1(B $(7"2(B $(7"3(B $(7"4(B $(7"5(B $(7"6(B $(7"7(B $(7"8(B $(7"9(B $(7":(B $(7";(B $(7"<(B $(7"=(B $(7">(B $(7"?(B ; and ;;;2240 $(7"@(B $(7"A(B $(7"B(B $(7"C(B $(7"D(B $(7"E(B $(7"F(B $(7"G(B $(7"H(B $(7"I(B $(7"J(B $(7"K(B $(7"L(B $(7"M(B $(7"N(B $(7"O(B ; Vowel signs. ;;;2250 $(7"P(B $(7"Q(B $(7"R(B $(7"S(B $(7"T(B $(7"U(B $(7"V(B $(7"W(B $(7"X(B $(7"Y(B $(7"Z(B $(7"[(B $(7"\(B $(7"](B $(7"^(B $(7"_(B ; (\x2251 = vowel a) ;;;2260 $(7"`(B $(7"a(B $(7"b(B $(7"c(B $(7"d(B $(7"e(B $(7"f(B $(7"g(B $(7"h(B $(7"i(B $(7"j(B $(7"k(B $(7"l(B $(7"m(B $(7"n(B $(7"o(B ; Long vowels and ;;;2270 $(7"p(B $(7"q(B $(7"r(B $(7"s(B $(7"t(B $(7"u(B $(7"v(B $(7"w(B $(7"x(B $(7"y(B $(7"z(B $(7"{(B $(7"|(B $(7"}(B $(7"~(B // ; vocalic r, l ARE ;;; ; atomically ;;; ; encoded. ;;; 00 01 02 03 04 05 06 07 08 09 0A 0B 0C 0D 0E 0F ;;;2320 // $(7#!(B $(7#"(B $(7##(B $(7#$(B $(7#%(B $(7#&(B $(7#'(B $(7#((B $(7#)(B $(7#*(B $(7#+(B $(7#,(B $(7#-(B $(7#.(B $(7#/(B ; Subjoined consonants ;;;2330 $(7#0(B $(7#1(B $(7#2(B $(7#3(B $(7#4(B $(7#5(B $(7#6(B $(7#7(B $(7#8(B $(7#9(B $(7#:(B $(7#;(B $(7#<(B $(7#=(B $(7#>(B $(7#?(B ; ;;;2340 $(7#@(B $(7#A(B $(7#B(B $(7#C(B $(7#D(B $(7#E(B $(7#F(B $(7#G(B $(7#H(B $(7#I(B $(7#J(B $(7#K(B $(7#L(B $(7#M(B $(7#N(B $(7#O(B ; ;;;2350 $(7#P(B $(7#Q(B $(7#R(B $(7#S(B $(7#T(B $(7#U(B $(7#V(B $(7#W(B $(7#X(B $(7#Y(B $(7#Z(B $(7#[(B $(7#\(B $(7#](B $(7#^(B $(7#_(B ; Hereafter, the chars ;;;2360 $(7#`(B $(7#a(B $(7#b(B $(7#c(B $(7#d(B $(7#e(B $(7#f(B $(7#g(B $(7#h(B $(7#i(B $(7#j(B $(7#k(B $(7#l(B $(7#m(B $(7#n(B $(7#o(B ; are not specified ;;;2370 $(7#p(B $(7#q(B $(7#r(B $(7#s(B $(7#t(B $(7#u(B $(7#v(B $(7#w(B $(7#x(B $(7#y(B $(7#z(B $(7#{(B $(7#|(B $(7#}(B $(7#~(B // ; in Unicode. ;;; ;;; 00 01 02 03 04 05 06 07 08 09 0A 0B 0C 0D 0E 0F ;;;2420 // $(7$!(B $(7$"(B $(7$#(B $(7$$(B $(7$%(B $(7$&(B $(7$'(B $(7$((B $(7$)(B $(7$*(B $(7$+(B $(7$,(B $(7$-(B $(7$.(B $(7$/(B ; Precomposed ;;;2430 $(7$0(B $(7$1(B $(7$2(B $(7$3(B $(7$4(B $(7$5(B $(7$6(B $(7$7(B $(7$8(B $(7$9(B $(7$:(B $(7$;(B $(7$<(B $(7$=(B $(7$>(B $(7$?(B ; consonants for ;;;2440 $(7$@(B $(7$A(B $(7$B(B $(7$C(B $(7$D(B $(7$E(B $(7$F(B $(7$G(B $(7$H(B $(7$I(B $(7$J(B $(7$K(B $(7$L(B $(7$M(B $(7$N(B $(7$O(B ; ordinary Tibetan. ;;;2450 $(7$P(B $(7$Q(B $(7$R(B $(7$S(B $(7$T(B $(7$U(B $(7$V(B $(7$W(B $(7$X(B $(7$Y(B $(7$Z(B $(7$[(B $(7$\(B $(7$](B $(7$^(B $(7$_(B ; They are decomposed ;;;2460 $(7$`(B $(7$a(B $(7$b(B $(7$c(B $(7$d(B $(7$e(B $(7$f(B $(7$g(B $(7$h(B $(7$i(B $(7$j(B $(7$k(B $(7$l(B $(7$m(B $(7$n(B $(7$o(B ; into base and ;;;2470 $(7$p(B $(7$q(B $(7$r(B $(7$s(B $(7$t(B $(7$u(B $(7$v(B $(7$w(B $(7$x(B $(7$y(B $(7$z(B $(7${(B $(7$|(B $(7$}(B $(7$~(B // ; subjoined consonants ;;; ; when written on a ;;; 00 01 02 03 04 05 06 07 08 09 0A 0B 0C 0D 0E 0F ; file in Tibetan ;;;2520 // $(7%!(B $(7%"(B $(7%#(B $(7%$(B $(7%%(B $(7%&(B $(7%'(B $(7%((B $(7%)(B $(7%*(B $(7%+(B $(7%,(B $(7%-(B $(7%.(B $(7%/(B ; coding system. ;;;2530 $(7%0(B $(7%1(B $(7%2(B $(7%3(B $(7%4(B $(7%5(B $(7%6(B $(7%7(B $(7%8(B $(7%9(B $(7%:(B $(7%;(B $(7%<(B $(7%=(B $(7%>(B $(7%?(B ; ;;;2540 $(7%@(B $(7%A(B $(7%B(B $(7%C(B $(7%D(B $(7%E(B $(7%F(B $(7%G(B $(7%H(B $(7%I(B $(7%J(B $(7%K(B $(7%L(B $(7%M(B $(7%N(B $(7%O(B ; ;;;2550 $(7%P(B $(7%Q(B $(7%R(B $(7%S(B $(7%T(B $(7%U(B $(7%V(B $(7%W(B $(7%X(B $(7%Y(B $(7%Z(B $(7%[(B $(7%\(B $(7%](B $(7%^(B $(7%_(B ; ;;;2560 $(7%`(B $(7%a(B $(7%b(B $(7%c(B $(7%d(B $(7%e(B $(7%f(B $(7%g(B $(7%h(B $(7%i(B $(7%j(B $(7%k(B $(7%l(B $(7%m(B $(7%n(B $(7%o(B ; ;;;2570 $(7%p(B $(7%q(B $(7%r(B $(7%s(B $(7%t(B $(7%u(B $(7%v(B $(7%w(B $(7%x(B $(7%y(B $(7%z(B $(7%{(B $(7%|(B $(7%}(B $(7%~(B // ; ;;; (make-charset 'tibetan-1-column "Tibetan 1 column glyph" '(dimension 2 registry "MuleTibetan-1" chars 94 columns 1 direction l2r final ?8 graphic 0 short-name "Tibetan 1-col" long-name "Tibetan 1 column" )) ;; Tibetan script. (make-charset 'tibetan "Tibetan characters" '(dimension 2 registry "MuleTibetan-2" chars 94 columns 2 direction l2r final ?7 graphic 0 short-name "Tibetan 2-col" long-name "Tibetan 2 column" )) ; (make-coding-system ; 'tibetan-iso-8bit 2 ?Q ; "8-bit encoding for ASCII (MSB=0) and TIBETAN (MSB=1)" ; '(ascii tibetan nil nil ; nil nil) ; '((safe-charsets ascii tibetan) ; (post-read-conversion . tibetan-post-read-conversion) ; (pre-write-conversion . tibetan-pre-write-conversion))) (make-coding-system 'tibetan-iso-8bit 'iso2022 "Tibetan/8-bit" '(charset-g0 ascii charset-g1 tibetan mnemonic "Tibetan" safe-charsets (ascii tibetan) post-read-conversion tibetan-post-read-conversion pre-write-conversion tibetan-pre-write-conversion documentation "8-bit encoding for ASCII (MSB=0) and TIBETAN (MSB=1)")) (define-coding-system-alias 'tibetan 'tibetan-iso-8bit) (set-language-info-alist "Tibetan" '((charset tibetan tibetan-1-column) (coding-system tibetan-iso-8bit) (coding-priority iso-2022-7bit tibetan-iso-8bit) (input-method . "tibetan-wylie") (features tibet-util) (documentation . t) (sample-text . (tibetan-compose-string (copy-sequence "Tibetan (4$(7"7r'"]0"7"]14"20"21!;4%P0"G#!"Q14"20"21!;(B) $(7!4!5!5!>4"70"714$P0"!#C"Q1!;4"Er'"S0"E"S14"G0"G1!;4"70"714"2r'"[0"2"[1!;4"Dr'"[0"D"[14"#0"#14"G0"G1!>4"Ir'"]r'"_0"I"]"_1!;4"90"9"Q1!;4"/r'"S0"/"S1!;4"50"5"Q14#2x!#9r'"[0"2#9"[1!;4"Hx!"Rx!"Ur'"c0"H"A"U"c1!>(B"))))) ;; `$(7"A(B' is included in the pattern for subjoined consonants because we ;; treat it specially in tibetan-add-components. ;; modified by Tomabechi 1999/12/10 ;; modified by Tomabechi 2000/06/08 ;; To allow infinite addition of vowels/modifiers ;; as specified in Unicode v.3 ;; $(7"A(B is removed from the class of subjoined. Tomabechi 2000/06/08 ;; (for Unicode support) (defconst tibetan-composable-pattern "[$(7"!(B-$(7"J"K(B][$(7#!(B-$(7#J#K#L#M(B]*[$(7"Q"R"S(B-$(7"^"a"b"e(B]*[$(7"_"c"d"g(B-$(7"l!I!e!g(B]*" "Regexp matching a composable sequence of Tibetan characters.") ;; Register a function to compose Tibetan characters. (put-char-table 'tibetan (list (cons tibetan-composable-pattern 'tibetan-composition-function)) composition-function-table) ;;; ;;; Definitions of conversion data. ;;; ;;; alists for tibetan char <-> transcription conversion ;;; longer transcription should come first (defconst tibetan-consonant-transcription-alist '(("tsh" . "$(7";(B") ("dzh" . "$(7"=(B") ("kSH" . "$(7"J(B") ("kh" . "$(7""(B") ("gh" . "$(7"$(B") ("ng" . "$(7"%(B") ("ch" . "$(7"'(B") ("ny" . "$(7"*(B") ("TH" . "$(7",(B") ("DH" . "$(7".(B") ("th" . "$(7"1(B") ("dh" . "$(7"3(B") ("ph" . "$(7"6(B") ("bh" . "$(7"8(B") ("ts" . "$(7":(B") ("dz" . "$(7"<(B") ("zh" . "$(7"?(B") ("sh" . "$(7"E(B") ("SH" . "$(7"F(B") ("k" . "$(7"!(B") ("g" . "$(7"#(B") ("c" . "$(7"&(B") ("j" . "$(7"((B") ("T" . "$(7"+(B") ("D" . "$(7"-(B") ("N" . "$(7"/(B") ("t" . "$(7"0(B") ("d" . "$(7"2(B") ("n" . "$(7"4(B") ("p" . "$(7"5(B") ("b" . "$(7"7(B") ("m" . "$(7"9(B") ("w" . "$(7">(B") ("z" . "$(7"@(B") ("'" . "$(7"A(B") ("y" . "$(7"B(B") ("r" . "$(7"C(B") ("l" . "$(7"D(B") ("s" . "$(7"G(B") ("h" . "$(7"H(B") ("H" . "$(7"H(B") ("A" . "$(7"I(B") ;; Added by Tomabechi 1999/12/10 ("R" . "$(7"K(B") ;; fixed form RA )) (defconst tibetan-vowel-transcription-alist '( ;; Composite Vowels ;; Added by Tomabechi 2000/06/08 ("frr" . "$(7"X(B") ("fll" . "$(7"Z(B") ("fa" . "$(7"R(B") ("fi" . "$(7"T(B") ("fu" . "$(7"V(B") ("fr" . "$(7"W(B") ("fl" . "$(7"Y(B") ("fI" . "$(7"b(B") ;; Normal Vowels ("ai" . "$(7"\(B") ("au" . "$(7"^(B") ("ee" . "$(7"\(B") ("oo" . "$(7"^(B") ("a" . "$(7"Q(B") ; invisible vowel sign (\x2251) ("i" . "$(7"S(B") ("u" . "$(7"U(B") ("e" . "$(7"[(B") ("o" . "$(7"](B") ("E" . "$(7"\(B") ("O" . "$(7"^(B") ("I" . "$(7"a(B") ("," . "$(7"e(B") ; idem. )) (defconst tibetan-modifier-transcription-alist '(("M" . "$(7"_(B") ("~" . "$(7"c(B") ("`" . "$(7"d(B") ("x" . "$(7"i(B") ("X" . "$(7"j(B") ("v" . "$(7"g(B") ("V" . "$(7"h(B") ("q" . "$(7"k(B") ("Q" . "$(7"l(B") ("_/" . "$(7!I(B") ("_o" . "$(7!g(B") ("_O" . "$(7!e(B"))) (defconst tibetan-precomposed-transcription-alist '(("phyw" . "$(7$G(B") ("tshw" . "$(7$)(B") ("rtsw" . "$(7%.(B") ("khw" . "$(7$"(B") ("nyw" . "$(7$%(B") ("tsw" . "$(7$((B") ("zhw" . "$(7$*(B") ("shw" . "$(7$.(B") ("khy" . "$(7$A(B") ("phy" . "$(7$D(B") ("khr" . "$(7$Q(B") ("thr" . "$(7$T(B") ("phr" . "$(7$W(B") ("shr" . "$(7$Z(B") ("dzr" . "$(7$^(B") ("grw" . "$(7$_(B") ("rng" . "$(7%#(B") ("rny" . "$(7%%(B") ("rts" . "$(7%+(B") ("rdz" . "$(7%,(B") ("rgw" . "$(7%-(B") ("rky" . "$(7%0(B") ("rgy" . "$(7%1(B") ("rmy" . "$(7%2(B") ("lng" . "$(7%B(B") ("sng" . "$(7%R(B") ("sny" . "$(7%S(B") ("sts" . "$(7%Z(B") ("sky" . "$(7%`(B") ("sgy" . "$(7%a(B") ("spy" . "$(7%b(B") ("sby" . "$(7%c(B") ("smy" . "$(7%d(B") ("skr" . "$(7%p(B") ("sgr" . "$(7%q(B") ("snr" . "$(7%r(B") ("spr" . "$(7%s(B") ("sbr" . "$(7%t(B") ("smr" . "$(7%u(B") ("kw" . "$(7$!(B") ("gw" . "$(7$#(B") ("cw" . "$(7$$(B") ("tw" . "$(7$&(B") ("dw" . "$(7$'(B") ("zw" . "$(7$+(B") ("rw" . "$(7$,(B") ("lw" . "$(7$-(B") ("sw" . "$(7$/(B") ("hw" . "$(7$0(B") ("ky" . "$(7$@(B") ("gy" . "$(7$B(B") ("py" . "$(7$C(B") ("by" . "$(7$E(B") ("my" . "$(7$F(B") ("kr" . "$(7$P(B") ("gr" . "$(7$R(B") ("tr" . "$(7$S(B") ("dr" . "$(7$U(B") ("pr" . "$(7$V(B") ("brk" . "$(7"7%!(B") ("brg" . "$(7"7%"(B") ("brng" . "$(7"7%#(B") ("brj" . "$(7"7%$(B") ("brny" . "$(7"7%%(B") ("brt" . "$(7"7%&(B") ("brd" . "$(7"7%'(B") ("brn" . "$(7"7%((B") ("brts" . "$(7"7%+(B") ("brdz" . "$(7"7%,(B") ("brl" . "$(7"7$d(B") ("br" . "$(7$X(B") ("mr" . "$(7$Y(B") ("sr" . "$(7$[(B") ("hr" . "$(7$\(B") ("jr" . "$(7$](B") ("kl" . "$(7$`(B") ("gl" . "$(7$a(B") ("blt" . "$(7"7%E(B") ("bld" . "$(7"7%F(B") ("bl" . "$(7$b(B") ("zl" . "$(7$c(B") ("rl" . "$(7$d(B") ("sl" . "$(7$e(B") ("rk" . "$(7%!(B") ("rg" . "$(7%"(B") ("rj" . "$(7%$(B") ("rt" . "$(7%&(B") ("rd" . "$(7%'(B") ("rn" . "$(7%((B") ("rb" . "$(7%)(B") ("rm" . "$(7%*(B") ("lk" . "$(7%@(B") ("lg" . "$(7%A(B") ("lc" . "$(7%C(B") ("lj" . "$(7%D(B") ("lt" . "$(7%E(B") ("ld" . "$(7%F(B") ("ln" . "$(7!!(B") ; dummy \x2121 ("lp" . "$(7%G(B") ("lb" . "$(7%H(B") ("lh" . "$(7%I(B") ("sk" . "$(7%P(B") ("sg" . "$(7%Q(B") ("st" . "$(7%T(B") ("sd" . "$(7%U(B") ("sn" . "$(7%V(B") ("sp" . "$(7%W(B") ("sb" . "$(7%X(B") ("sm" . "$(7%Y(B")) ) (defconst tibetan-subjoined-transcription-alist (sort '(("+k" . "$(7#!(B") ("+kh" . "$(7#"(B") ("+g" . "$(7##(B") ("+gh" . "$(7#$(B") ("+ng" . "$(7#%(B") ("+c" . "$(7#&(B") ("+ch" . "$(7#'(B") ("+j" . "$(7#((B") ("+ny" . "$(7#*(B") ("+T" . "$(7#+(B") ("+TH" . "$(7#,(B") ("+D" . "$(7#-(B") ("+DH" . "$(7#.(B") ("+N" . "$(7#/(B") ("+t" . "$(7#0(B") ("+th" . "$(7#1(B") ("+d" . "$(7#2(B") ("+dh" . "$(7#3(B") ("+n" . "$(7#4(B") ("+p" . "$(7#5(B") ("+ph" . "$(7#6(B") ("+b" . "$(7#7(B") ("+bh" . "$(7#8(B") ("+m" . "$(7#9(B") ("+ts" . "$(7#:(B") ("+tsh" . "$(7#;(B") ("+dz" . "$(7#<(B") ("+dzh" . "$(7#=(B") ("+w" . "$(7#>(B") ("+zh" . "$(7#?(B") ("+z" . "$(7#@(B") ("+'" . "$(7#A(B") ("+y" . "$(7#B(B") ("+r" . "$(7#C(B") ("+l" . "$(7#D(B") ("+sh" . "$(7#E(B") ("+SH" . "$(7#F(B") ("+s" . "$(7#G(B") ("+h" . "$(7#H(B") ("+A" . "$(7#I(B") ("+kSH" . "$(7#J(B") ;; Added by Tomabechi 1999/12/10 ("+W" . "$(7#K(B") ;; fixed form subscribed WA ("+Y" . "$(7#L(B") ;; fixed form subscribed YA ("+R" . "$(7#M(B") ;; fixed form subscribed RA ) (lambda (x y) (> (length (car x)) (length (car y)))))) ;;; ;;; alist for Tibetan base consonant <-> subjoined consonant conversion. ;;; (defconst tibetan-base-to-subjoined-alist '(("$(7"!(B" . "$(7#!(B") ("$(7""(B" . "$(7#"(B") ("$(7"#(B" . "$(7##(B") ("$(7"$(B" . "$(7#$(B") ("$(7"%(B" . "$(7#%(B") ("$(7"&(B" . "$(7#&(B") ("$(7"'(B" . "$(7#'(B") ("$(7"((B" . "$(7#((B") ("$(7"*(B" . "$(7#*(B") ("$(7"+(B" . "$(7#+(B") ("$(7",(B" . "$(7#,(B") ("$(7"-(B" . "$(7#-(B") ("$(7".(B" . "$(7#.(B") ("$(7"/(B" . "$(7#/(B") ("$(7"0(B" . "$(7#0(B") ("$(7"1(B" . "$(7#1(B") ("$(7"2(B" . "$(7#2(B") ("$(7"3(B" . "$(7#3(B") ("$(7"4(B" . "$(7#4(B") ("$(7"5(B" . "$(7#5(B") ("$(7"6(B" . "$(7#6(B") ("$(7"7(B" . "$(7#7(B") ("$(7"8(B" . "$(7#8(B") ("$(7"9(B" . "$(7#9(B") ("$(7":(B" . "$(7#:(B") ("$(7";(B" . "$(7#;(B") ("$(7"<(B" . "$(7#<(B") ("$(7"=(B" . "$(7#=(B") ("$(7">(B" . "$(7#>(B") ("$(7"?(B" . "$(7#?(B") ("$(7"@(B" . "$(7#@(B") ("$(7"A(B" . "$(7#A(B") ("$(7"B(B" . "$(7#B(B") ("$(7"C(B" . "$(7#C(B") ("$(7"D(B" . "$(7#D(B") ("$(7"E(B" . "$(7#E(B") ("$(7"F(B" . "$(7#F(B") ("$(7"G(B" . "$(7#G(B") ("$(7"H(B" . "$(7#H(B") ("$(7"I(B" . "$(7#I(B") ("$(7"J(B" . "$(7#J(B") ;; Added by Tomabechi 1999/12/10 ("$(7"K(B" . "$(7#M(B") ;; Fixed form RA (224B->234D) )) ;;; alist for Tibetan composite vowels (long i, vocalic r, etc.) ;;; New varialble. created by Tomabechi 2000/06/08 (defconst tibetan-composite-vowel-alist '(;; LONG A ;; ("$(7"R(B" . ((bc . tc) ?$(7"R(B)) ;; LONG I ("$(7"T(B" . (?$(7"R(B (tc . bc) ?$(7"S(B)) ;; LONG U ("$(7"V(B" . (?$(7"R(B (bc . tc) ?$(7"U(B)) ;; VOCALIC R ("$(7"W(B" . (?$(7#C(B (tc . bc) ?$(7"a(B)) ;; LONG VOCALIC R ("$(7"X(B" . (?$(7#C(B (bc . tc) ?$(7"R(B (tc . bc) ?$(7"a(B)) ;; VOCALIC L ("$(7"Y(B" . (?$(7#D(B (tc . bc) ?$(7"a(B)) ;;$(7!;(BLONG VOCALIC L ("$(7"Z(B" . (?$(7#D(B (bc . tc) ?$(7"R(B (tc . bc) ?$(7"a(B)) ;; LONG REVERSE I ("$(7"b(B" . (?$(7"R(B (tc . bc) ?$(7"a(B)) )) ;;; ;;; alist for Tibetan consonantic components <-> precomposed glyph conversion. ;;; (includes some punctuation conversion rules) ;;; (defconst tibetan-precomposition-rule-alist `(("$(7"6#B#>(B" . "$(7$G(B") ("$(7"##C#>(B" . "$(7$_(B") ("$(7";#>(B" . "$(7$)(B") ("$(7"C#:#>(B" . "$(7%.(B") ("$(7"C###>(B" . "$(7%-(B") ("$(7"C#!#B(B" . "$(7%0(B") ("$(7"C###B(B" . "$(7%1(B") ("$(7"C#9#B(B" . "$(7%2(B") ("$(7"G#!#B(B" . "$(7%`(B") ("$(7"G###B(B" . "$(7%a(B") ("$(7"G#5#B(B" . "$(7%b(B") ("$(7"G#7#B(B" . "$(7%c(B") ("$(7"G#9#B(B" . "$(7%d(B") ("$(7"G#!#C(B" . "$(7%p(B") ("$(7"G###C(B" . "$(7%q(B") ("$(7"G#4#C(B" . "$(7%r(B") ("$(7"G#5#C(B" . "$(7%s(B") ("$(7"G#7#C(B" . "$(7%t(B") ("$(7"G#9#C(B" . "$(7%u(B") ("$(7""#>(B" . "$(7$"(B") ("$(7"*#>(B" . "$(7$%(B") ("$(7":#>(B" . "$(7$((B") ("$(7"?#>(B" . "$(7$*(B") ("$(7"E#>(B" . "$(7$.(B") ("$(7""#B(B" . "$(7$A(B") ("$(7"6#B(B" . "$(7$D(B") ("$(7""#C(B" . "$(7$Q(B") ("$(7"1#C(B" . "$(7$T(B") ("$(7"6#C(B" . "$(7$W(B") ("$(7"E#C(B" . "$(7$Z(B") ("$(7"<#C(B" . "$(7$^(B") ("$(7"C#%(B" . "$(7%#(B") ("$(7"C#*(B" . "$(7%%(B") ("$(7"C#:(B" . "$(7%+(B") ("$(7"C#<(B" . "$(7%,(B") ("$(7"D#%(B" . "$(7%B(B") ("$(7"G#%(B" . "$(7%R(B") ("$(7"G#*(B" . "$(7%S(B") ("$(7"G#:(B" . "$(7%Z(B") ("$(7"!#>(B" . "$(7$!(B") ("$(7"##>(B" . "$(7$#(B") ("$(7"&#>(B" . "$(7$$(B") ("$(7"0#>(B" . "$(7$&(B") ("$(7"2#>(B" . "$(7$'(B") ("$(7"@#>(B" . "$(7$+(B") ("$(7"C#>(B" . "$(7$,(B") ("$(7"D#>(B" . "$(7$-(B") ("$(7"G#>(B" . "$(7$/(B") ("$(7"H#>(B" . "$(7$0(B") ("$(7"!#B(B" . "$(7$@(B") ("$(7"##B(B" . "$(7$B(B") ("$(7"5#B(B" . "$(7$C(B") ("$(7"7#B(B" . "$(7$E(B") ("$(7"9#B(B" . "$(7$F(B") ("$(7"!#C(B" . "$(7$P(B") ("$(7"##C(B" . "$(7$R(B") ("$(7"0#C(B" . "$(7$S(B") ("$(7"2#C(B" . "$(7$U(B") ("$(7"5#C(B" . "$(7$V(B") ("$(7"7#C(B" . "$(7$X(B") ("$(7"9#C(B" . "$(7$Y(B") ("$(7"G#C(B" . "$(7$[(B") ("$(7"H#C(B" . "$(7$\(B") ("$(7"(#C(B" . "$(7$](B") ("$(7"!#D(B" . "$(7$`(B") ("$(7"##D(B" . "$(7$a(B") ("$(7"7#D(B" . "$(7$b(B") ("$(7"@#D(B" . "$(7$c(B") ("$(7"C#D(B" . "$(7$d(B") ("$(7"G#D(B" . "$(7$e(B") ("$(7"C#!(B" . "$(7%!(B") ("$(7"C##(B" . "$(7%"(B") ("$(7"C#((B" . "$(7%$(B") ("$(7"C#0(B" . "$(7%&(B") ("$(7"C#2(B" . "$(7%'(B") ("$(7"C#4(B" . "$(7%((B") ("$(7"C#7(B" . "$(7%)(B") ("$(7"C#9(B" . "$(7%*(B") ("$(7"D#!(B" . "$(7%@(B") ("$(7"D##(B" . "$(7%A(B") ("$(7"D#4(B" . "$(7!!(B") ; dummy 0x2121 added 2000/06/08 for transition l -> lng ("$(7"D#&(B" . "$(7%C(B") ("$(7"D#((B" . "$(7%D(B") ("$(7"D#0(B" . "$(7%E(B") ("$(7"D#2(B" . "$(7%F(B") ("$(7"D#5(B" . "$(7%G(B") ("$(7"D#7(B" . "$(7%H(B") ("$(7"D#H(B" . "$(7%I(B") ("$(7"G#!(B" . "$(7%P(B") ("$(7"G##(B" . "$(7%Q(B") ("$(7"G#0(B" . "$(7%T(B") ("$(7"G#2(B" . "$(7%U(B") ("$(7"G#4(B" . "$(7%V(B") ("$(7"G#5(B" . "$(7%W(B") ("$(7"G#7(B" . "$(7%X(B") ("$(7"G#9(B" . "$(7%Y(B"))) (defconst tibetan-obsolete-glyphs `(("$(7!=(B" . "$(8!=(B") ; 2 col <-> 1 col ("$(7!?(B" . "$(8!?(B") ("$(7!@(B" . "$(8!@(B") ("$(7!A(B" . "$(8!A(B") ("$(7"`(B" . "$(8"`(B") ("$(7!;(B" . "$(8!;(B") ("$(7!D(B" . "$(8!D(B") ;; Yes these are dirty. But ... ("$(7!>(B $(7!>(B" . ,(compose-string "$(7!>(B $(7!>(B" 0 3 [?$(7!>(B (Br . Bl) ? (Br . Bl) ?$(7!>(B])) ("$(7!4!5!5(B" . ,(compose-string "$(7#R#S#S#S(B" 0 4 [?$(7#R(B (Br . Bl) ?$(7#S(B (Br . Bl) ?$(7#S(B (Br . Bl) ?$(7#S(B])) ("$(7!4!5(B" . ,(compose-string "$(7#R#S#S(B" 0 3 [?$(7#R(B (Br . Bl) ?$(7#S(B (Br . Bl) ?$(7#S(B])) ("$(7!6(B" . ,(compose-string "$(7#R#S!I(B" 0 3 [?$(7#R(B (Br . Bl) ?$(7#S(B (br . tr) ?$(7!I(B])) ("$(7!4(B" . ,(compose-string "$(7#R#S(B" 0 2 [?$(7#R(B (Br . Bl) ?$(7#S(B])))) (defconst tibetan-regexp (let ((l (list tibetan-precomposed-transcription-alist tibetan-consonant-transcription-alist tibetan-vowel-transcription-alist tibetan-modifier-transcription-alist tibetan-subjoined-transcription-alist)) (separator "\\|") tail pattern) (while l (setq tail (car l) l (cdr l)) (while tail (setq pattern (cons separator (cons (car (car tail)) pattern)) tail (cdr tail)))) (apply 'concat (nreverse (cdr pattern)))) "Regexp matching a Tibetan transcription of a composable Tibetan sequence. The result of matching is to be used for indexing alists at conversion from a roman transcription to the corresponding Tibetan character.") (defvar tibetan-precomposed-regexp (let ((l tibetan-precomposed-transcription-alist) temp) (setq temp "^\\(") (setq temp (concat temp (car (car l)))) (setq l (cdr l)) (while l (setq temp (concat temp "\\|" (car (car l)))) (setq l (cdr l))) (concat temp "\\)")) "Regexp string to match a romanized Tibetan complex consonant. The result of matching is to be used for indexing alists when the input key from an input method is converted to the corresponding precomposed glyph.") (defvar tibetan-precomposition-rule-regexp (let ((l tibetan-precomposition-rule-alist) temp) (setq temp "\\(") (setq temp (concat temp (car (car l)))) (setq l (cdr l)) (while l (setq temp (concat temp "\\|" (car (car l)))) (setq l (cdr l))) (concat temp "\\)")) "Regexp string to match a sequence of Tibetan consonantic components, i.e., one base consonant and one or more subjoined consonants. The result of matching is to be used for indexing alist when the component sequence is converted to the corresponding precomposed glyph. This also matches some punctuation characters which need conversion.") (defvar tibetan-decomposed nil) (defvar tibetan-decomposed-temp nil) (provide 'tibetan) ;;; tibetan.el ends here