မာတိကာသို့ ခုန်သွားရန်

မော်ဂျူး:IPA

အဝ်ႏ ဝိစ်သိဉ်နရီ ကို

Documentation for this module may be created at မော်ဂျူး:IPA/doc

local export = {}
-- [[Module:IPA/data]]

local m_data = mw.loadData('Module:IPA/data') -- [[Module:IPA/data]]
local m_symbols = mw.loadData('Module:IPA/data/symbols') -- [[Module:IPA/data/symbols]]
local m_syllables -- [[Module:syllables]]; loaded below if needed

local sub = mw.ustring.sub
local find = mw.ustring.find
local gsub = mw.ustring.gsub
local match = mw.ustring.match
local gmatch = mw.ustring.gmatch
local U = mw.ustring.char

function export.format_IPA_full(lang, items, err, separator, sortKey, no_count)
	local IPA_key, key_link, err_text, prefix, IPAs, category
	local hasKey = m_data.langs_with_infopages
	local namespace = mw.title.getCurrentTitle().nsText
	
	if err then
		err_text = '<span class="error">' .. err .. '</span>'
	else
		if hasKey[lang:getCode()] then
			IPA_key = "Appendix:" .. lang:getCanonicalName() .. " pronunciation"
		else
			IPA_key = "wikipedia:" .. lang:getCanonicalName() .. " phonology"
		end
		
		key_link = "[[" .. IPA_key .. "|key]]"
	end

	
	local prefix = "[[:en:Wiktionary:International Phonetic Alphabet|IPA]]<sup>(" .. ( key_link or err_text ) .. ")</sup>:&#32;"
	
	IPAs = export.format_IPA_multiple(lang, items, separator, no_count)
	
	if lang and (namespace == "" or namespace == "Reconstruction") then
		sortKey = sortKey or lang:makeSortKey(mw.title.getCurrentTitle().text)
		sortKey = sortKey and ("|" .. sortKey) or ""
		category = "[[Category:" .. "IPAအငဝ်းထန်ႏကိုကပါဒါႏ" .. lang:getCanonicalName() .. "ငဝ်းငွါဖုံႏ" .. sortKey .. "]]"
	else
		category = ""
	end

	return prefix .. IPAs .. category
end

local function determine_repr(pron)
	local repr_mark = {}
	local repr, reconstructed
	
	-- remove initial asterisk before representation marks, used on some Reconstruction pages
	if find(pron, "^%*") then
		reconstructed = true
		pron = sub(pron, 2)
	end
	
	local representation_types = {
		['/'] = { right = '/', type = 'phonemic', },
		['['] = { right = ']', type = 'phonetic', },
		['⟨'] = { right = '⟩', type = 'orthographic', },
		['-'] = { type = 'rhyme' },
	}
	
	repr_mark.i, repr_mark.f, repr_mark.left, repr_mark.right = find(pron, '^(.).-(.)$')
	
	local representation_type = representation_types[repr_mark.left]
	
	if representation_type then
		if representation_type.right then
			if repr_mark.right == representation_type.right then
				repr = representation_type.type
			end
		else
			repr = representation_type.type
		end
	else
		repr = nil
	end
	
	return repr, reconstructed
end

local function hasInvalidSeparators(transcription)
	if find(transcription, "%.[ˈˌ]") then
		return true
	else
		return false
	end
end

function export.format_IPA_multiple(lang, items, separator, no_count)
	local notes = {}
	local categories = {}
	separator = separator or ', '
	
	-- Format
	if not items[1] then
		if mw.title.getCurrentTitle().nsText == "တမ်းပလေက်" then
			table.insert(items, {pron = "/aɪ piː ˈeɪ/"})
		else
			table.insert(categories, "[[Category:Pronunciation templates without a pronunciation]]")
		end
	end
	
	local bits = {}
	
	for _, item in ipairs(items) do
		local bit = export.format_IPA(lang, item.pron)
		
		if item.qualifiers and item.qualifiers[1] then
			bit = require("Module:qualifier").format_qualifier(item.qualifiers) .. " " .. bit
		end
		
		if item.note then
			bit = bit .. mw.getCurrentFrame():extensionTag("ref", item.note)
		end
		
		table.insert(bits, bit)
		
		--[=[	[[Special:WhatLinksHere/Template:tracking/IPA/syntax-error]]
				The length or gemination symbol should not appear after a syllable break or stress symbol.	]=]
		
		if find(item.pron, "[ˈˌ%.][ːˑ]") then
			require("Module:debug").track("IPA/syntax-error")
		end
		
		if lang then
			-- Add syllable count if the language's diphthongs are listed in [[Module:syllables]].
			-- Don't do this if the term has spaces or a liaison mark (‿).
			if not no_count and mw.title.getCurrentTitle().namespace == 0 then
				m_syllables = m_syllables or require('Module:syllables')
				if m_data.langs_to_generate_syllable_count_categories[lang:getCode()] and
						(determine_repr(item.pron) == "phonemic" or lang:getCode() == "ru") and
						not find(item.pron, "[ ‿]") then
					local syllable_count = m_syllables.getVowels(item.pron, lang)
					if syllable_count then

					end	
				end
			end

			if lang:getCode() == "en" and hasInvalidSeparators(item.pron) then
				table.insert(categories, "[[ကဏ္ဍ:ပအိုဝ်ႏဘာႏသာႏငဝ်းငွါတာႏသွုံꩻIPA .ˈ တဝ်းလဲ့ˌ]]")
			end
		end
	end

	return table.concat(bits, separator) .. table.concat(categories)
end

-- Takes an IPA pronunciation and formats it and adds cleanup categories.
function export.format_IPA(lang, pron, split_output)
	local err = {}
	local categories = {}
	
	-- Remove wikilinks, so that wikilink brackets are not misinterpreted as
	-- indicating phonemic transcription
	local str_gsub = string.gsub
	local without_links = str_gsub(pron, '%[%[[^|%]]+|([^%]]+)%]%]', '%1')
	without_links = str_gsub(without_links, '%[%[[^%]]+%]%]', '%1')
	
	-- Detect whether this is a phonemic or phonetic transcription
	local repr, reconstructed = determine_repr(without_links)
	
	if reconstructed then
		pron = sub(pron, 2)
	end
	
	-- If valid, strip the representation marks
	if repr == "phonemic" then
		pron = sub(pron, 2, -2)
		without_links = sub(without_links, 2, -2)
	elseif repr == "phonetic" then
		pron = sub(pron, 2, -2)
		without_links = sub(without_links, 2, -2)
	elseif repr == "orthographic" then
		pron = sub(pron, 2, -2)
		without_links = sub(without_links, 2, -2)
	elseif repr == "rhyme" then
		pron = sub(pron, 2)
		without_links = sub(without_links, 2)
	else
		table.insert(categories, "[[Category:IPA pronunciations with invalid representation marks]]")
		-- table.insert(err, "invalid representation marks")
		-- Removed because it's annoying when previewing pronunciation pages.
	end
	
	if pron == "" then
		table.insert(categories, "[[ကဏ္ဍ:IPAအက္ခရာႏတမဲဉ်ႏတဝ်းဖုံႏ]]")
	end
	
	-- Check for obsolete and nonstandard symbols
	for i, symbol in ipairs(m_data.nonstandard) do
		local result
		for nonstandard in gmatch(pron, symbol) do
			if not result then
				result = {}
			end
			table.insert(result, nonstandard)
			table.insert(categories, "[[ကဏ္ဍ:ငဝ်းပွုံႏခွိုင်ꩻတဝ်းလဲ့IPAတဲမ်းနယ်ꩻအမာႏဖုံႏ|" .. nonstandard .. "]]")
		end
		
		if result then
			table.insert(err, "obsolete or nonstandard characters (" .. table.concat(result) .. ")")
			break
		end
	end
	
	--[[ Check for invalid symbols after removing the following:
			1. wikilinks (handled above)
			2. paired HTML tags
			3. bolding
			4. italics
			5. HTML entity for space
			6. asterisk at beginning of transcription
			7. comma followed by spacing characters
			8. superscripts enclosed in superscript parentheses		]]
	local found_HTML
	local result = str_gsub(without_links, "<(%a+)[^>]*>([^<]+)</%1>",
		function(tagName, content)
			found_HTML = true
			return content
		end)
	result = str_gsub(result, "'''([^']*)'''", "%1")
	result = str_gsub(result, "''([^']*)''", "%1")
	result = str_gsub(result, "&[^;]+;", "") -- This may catch things that are not valid character entities.
	result = str_gsub(result, "^%*", "")
	result = gsub(result, ",%s+", "")
	result = gsub(result, "⁽[".. m_symbols.superscripts .. "]+⁾", "")
	result = gsub(result, '[' .. m_symbols.valid .. ']', '')
	if result ~= '' then
		local suggestions = {}
		mw.log(pron, result)
		local namespace = mw.title.getCurrentTitle().namespace
		local category
		if namespace == 0 then
			-- main namespace
			category = "IPAအက္ခရာႏတွမ်ႏIPAအငဝ်းထန်ႏ အမာႏအဝ်ႏနဝ်ꩻသွူ"
		elseif namespace == 118 then
			-- reconstruction namespace
			category = "IPAအက္ခရာႏအမာႏ/မာꩻချာယင်းအတသာ"
		else
			category = "IPAအက္ခရာႏအမာႏအဝ်ႏနဝ်ꩻသွူ/non_mainspace"
		end
		for character in gmatch(result, ".") do
			local suggestion = m_symbols.suggestions[character]
			if suggestion then
				table.insert(suggestions, character .. " with " .. suggestion)
			end
			table.insert(categories, "[[Category:" .. category .. "|" .. character .. "]]")
		end
		table.insert(err, "IPAအက္ခရာႏအမာႏအဝ်ႏနဝ်ꩻသွူ (" .. result .. ")")
		if suggestions[1] then
			table.insert(err, "replace " .. table.concat(suggestions, ", "))
		end
	end
	
	if found_HTML then
		table.insert(categories, "[[Category:IPA pronunciations with paired HTML tags]]")
	end
	
	-- Reference inside IPA template usage
	-- FIXME: Doesn't work; you can't put HTML in module output.
	--if mw.ustring.find(pron, '</ref>') then

	--end
	
	if repr == "phonemic" or repr == "rhyme" then
		if lang and m_data.phonemes[lang:getCode()] then
			local valid_phonemes = m_data.phonemes[lang:getCode()]
			local rest = pron
			local phonemes = {}
			
			while mw.ustring.len(rest) > 0 do
				local longestmatch = ""
				
				if sub(rest, 1, 1) == "(" or sub(rest, 1, 1) == ")" then
					longestmatch = sub(rest, 1, 1)
				else
					for _, phoneme in ipairs(valid_phonemes) do
						if mw.ustring.len(phoneme) > mw.ustring.len(longestmatch) and sub(rest, 1, mw.ustring.len(phoneme)) == phoneme then
							longestmatch = phoneme
						end
					end
				end
				
				if mw.ustring.len(longestmatch) > 0 then
					table.insert(phonemes, longestmatch)
					rest = sub(rest, mw.ustring.len(longestmatch) + 1)
				else
					local phoneme = sub(rest, 1, 1)
					table.insert(phonemes, "<span style=\"color: red\">" .. phoneme .. "</span>")
					rest = sub(rest, 2)
					table.insert(categories, "[[ကဏ္ဍ:IPAအက္ခရာႏတမဲဉ်ႏတဝ်းဖုံႏ/" .. lang:getCode() .. "]]")
					require("Module:debug").track("IPA/invalid phonemes/" .. phoneme)
				end
			end
			
			pron = table.concat(phonemes)
		end
		
		if repr == "phonemic" then
			pron = "/" .. pron .. "/"
		else
			pron = "-" .. pron
		end
	elseif repr == "phonetic" then
		pron = "[" .. pron .. "]"
	elseif repr == "orthographic" then
		pron = "⟨" .. pron .. "⟩"
	end
	
	if reconstructed then
		pron = "*" .. pron
	end
	
	if err[1] then
		err = '<span class="previewonly error" style="font-size: small;>&#32;' .. table.concat(err, ', ') .. '</span>'
	else
		err = ""
	end
	
	if split_output then -- for use of IPA in links 
		return '<span class="IPA">' .. pron .. '</span>', table.concat(categories), err
	else
		return '<span class="IPA">' .. pron .. '</span>' .. err .. table.concat(categories)
	end
end

function export.example(frame)
	local output = {}
	
	local m_links = require('Module:links')
	local m_languages = require('Module:languages')
	
	table.insert(
		output,
[[
{| class="wikitable"
! Term !! IPA !! Generated X-SAMPA !! Regenerated IPA !! Matched?
]]
	)
	local row =
[[
|-
| link || IPA || XSAMPA || regenerated_IPA || matched
]]
	
	local examples = mw.text.split(frame.args[1], ",%s*")
	
	local m_XSAMPA = require("Module:IPA/X-SAMPA")
	
	for _, example in pairs(examples) do
		local lang, word = match(example, "(%l%l%l?):(.+) [/%[]")
		
		if lang then
			lang = m_languages.getByCode(lang) or error('"' .. lang .. '" is not a valid language code.')
		end
		
		local IPA = match(example, "/[^/]+/")
			or match(example, "%[[^%]]+%]")
			or error('No IPA transcription found in "' .. example .. '".')
		local XSAMPA = m_XSAMPA.IPA_to_XSAMPA(IPA)
		local regenerated_IPA = m_XSAMPA.XSAMPA_to_IPA(XSAMPA)
		
		content = {
			link = lang and word and m_links.full_link{ term = word, lang = lang },
			matched = IPA == regenerated_IPA
				and '<span style="color: green;">yes</span>'
				or '<span style="color: red;">no</span>',
			IPA = '<span class="IPA">' .. IPA .. '</span>',
			XSAMPA = '<code>' .. XSAMPA .. '</span>',
			regenerated_IPA = '<span class="IPA">' .. regenerated_IPA .. '</span>'
		}
		
		local function add_content(item)
			return content[item] or ""
		end
		local row = gsub(row, "[%a_]+", add_content)
		table.insert(output, row)
	end
	
	table.insert(output, "|}")
	
	return table.concat(output)
end

return export