Module:Jv-translit

維基詞典,自由的多語言詞典
跳到导航 跳到搜索
文档图示 模块文档[创建]
local export = {}

local conv = {
	-- finals (U+A980 - U+A983):
	
	["ꦀ"] = "m",
	["ꦁ"] = "ng",
	["ꦂ"] = "r",
	["ꦃ"] = "h",
	
	-- independent vowels (U+A984 - U+A98E):
	
	["ꦄ"] = "a",   ["ꦄꦴ"] = "a",
	["ꦅ"] = "i",
	["ꦆ"] = "i",
	["ꦇ"] = "i",
	["ꦈ"] = "u",   ["ꦈꦴ"] = "u",
	["ꦉ"] = "re",  ["ꦉꦴ"] = "reu",
	["ꦊ"] = "le",
	["ꦋ"] = "leu",
	["ꦌ"] = "e",
	["ꦍ"] = "ai",
	["ꦎ"] = "o",
	
	-- independent consonants (U+A98F - U+A9B2):
	
	["ꦏ"] = "k",   ["ꦏ꦳"] = "kh",
	["ꦐ"] = "q",
	["ꦑ"] = "kh",
	["ꦒ"] = "g",   ["ꦒ꦳"] = "gh",
	["ꦓ"] = "gh",
	["ꦔ"] = "ng",  ["ꦔ꦳"] = "'",
	
	["ꦕ"] = "c",
	["ꦖ"] = "ch",
	["ꦗ"] = "j",   ["ꦗ꦳"] = "z",
	["ꦘ"] = "jny",
	["ꦙ"] = "jh",
	["ꦚ"] = "n",
	
	["ꦛ"] = "th",
	["ꦜ"] = "th",
	["ꦝ"] = "dh",
	["ꦞ"] = "dh",
	["ꦟ"] = "nn",
	
	["ꦠ"] = "t",
	["ꦡ"] = "th",
	["ꦢ"] = "d",   ["ꦢ꦳"] = "dz",
	["ꦣ"] = "dh",
	["ꦤ"] = "n",
	
	["ꦥ"] = "p",   ["ꦥ꦳"] = "f",
	["ꦦ"] = "ph",
	["ꦧ"] = "b",
	["ꦨ"] = "bh",
	["ꦩ"] = "m",
	
	["ꦪ"] = "y",
	["ꦫ"] = "r",
	["ꦬ"] = "r",
	["ꦭ"] = "l",
	["ꦮ"] = "w",   ["ꦮ꦳"] = "v",
	["ꦯ"] = "sh",
	["ꦰ"] = "ss",
	["ꦱ"] = "s",   ["ꦱ꦳"] = "sy",
	["ꦲ"] = "h",   ["ꦲ꦳"] = "h",
	
	-- cecak_telu/nukta (U+A9B3):
	
	["꦳"] = "",
	
	-- dependent vowels (U+A9B4 - A9BD):
	
	["ꦴ"] = "a", -- tarung
	["ꦵ"] = "o",
	["ꦶ"] = "i",
	["ꦷ"] = "i",
	["ꦸ"] = "u",
	["ꦹ"] = "u",
	["ꦺ"] = "é",   ["ꦺꦴ"] = "o",
	["ꦻ"] = "ai",  ["ꦻꦴ"] = "au",
	["ꦼ"] = "e",
	["ꦽ"] = "re",  ["ꦽꦴ"] = "reu",
	
	-- medials (U+A9BE - U+A9BF):
	["ꦾ"] = "y",
	["ꦿ"] = "r",
	
	-- pangkon/virama (U+A9C0):
	
	["꧀"] = "",
	
	-- punctuation (U+A9C1 - U+A9CF):
	
	["꧁"] = "(starts title)",
	["꧂"] = "(ends title)",
	["꧃"] = "(letter to younger age or lower rank)",
	["꧄"] = "(letter to equal age or equal rank)",
	["꧅"] = "(letter to older age or higher rank)",
	["꧆"] = "(pada windu)",
	["꧇"] = ":", -- number indicator
	["꧈"] = ",",
	["꧉"] = ".",
	["꧊"] = "\"",
	["꧋"] = "//",
	["꧌"] = "(",
	["꧍"] = ")",
	["ꧏ"] = "<sup>2</sup>",
	
	-- digits (U+A9D0 - U+A9D9):
	
	["꧐"] = "0",
	["꧑"] = "1",
	["꧒"] = "2",
	["꧓"] = "3",
	["꧔"] = "4",
	["꧕"] = "5",
	["꧖"] = "6",
	["꧗"] = "7",
	["꧘"] = "8",
	["꧙"] = "9",
	
	-- ellipsis (U+A9DE - U+A9DF):
	
	["꧞"] = "-",
	["꧟"] = "-",
}

function export.tr(text, lang, sc)
	local CSVC = {
		initial = "([ꦏ-ꦲ]꦳?)",
		medial = "([ꦾꦿ]?)",
		nucleus = "([ꦴ-ꦽ꧀]?ꦴ?)",
		final = "([ꦀ-ꦃ]?)",
	}
	local VC = {
		nucleus = "([ꦄ-ꦎ]ꦴ?)",
		final = "([ꦀ-ꦃ]*)",
	}
	
	local number_indicator = "꧇"
	local digits = "[꧐-꧙]"
	
	local initial = true
	
	text = mw.ustring.gsub(
		text,
		CSVC.initial .. CSVC.medial .. CSVC.nucleus.. CSVC.final,
		function(a, b, c, d)
			a = conv[a] or error("Initial not recognized: " .. a)
			b = b == "" and "" or conv[b] or error("Medial not recognized: " .. b)
			c = c == "" and "a" or conv[c] or error("Nucleus not recognized: " .. c)
			d = d == "" and "" or conv[d] or error("Final not recognized: " .. d)
			if initial and a == "h" then
				a = ""
			end
			initial = false
			return a .. b .. c .. d
		end
	)
	
	text = mw.ustring.gsub(
		text,
		VC.nucleus .. VC.final,
		function(a, b)
			a = conv[a]
			b = (b == "" and "" or conv[b])
			initial = false
			return a .. b
		end
	)
	
	text = mw.ustring.gsub(
		text,
		number_indicator .. "(" .. digits .. "+)" .. number_indicator,
		function(a)
			a = mw.ustring.gsub(a, ".", conv)
			initial = true
			return a
		end
	)
	
	text = mw.ustring.gsub(text, ".", conv)
	
	return text
end

return export