Модуль:grc-utilities
Documentation for this module may be created at Модуль:grc-utilities/doc
local export = {}
local m_script_utils = require("Module:script utilities")
local m_string_utils = require("Module:string utilities")
local m_links = require("Module:links")
local lang = require("Module:languages").getByCode("grc")
local sc = require("Module:scripts").getByCode("Polyt")
local m_data = mw.loadData("Module:grc-utilities/data")
local groups = m_data.groups
local diacritic_order = m_data.diacritic_order
local conversions = m_data.conversions
local diacritics = m_data.diacritics
local diacritic = m_data.diacritic
local macron = diacritics.macron
local breve = diacritics.breve
local spacing_macron = diacritics.spacing_macron
local spacing_breve = diacritics.spacing_breve
local rough = diacritics.rough
local smooth = diacritics.smooth
local diaeresis = diacritics.diaeresis
local acute = diacritics.acute
local grave = diacritics.grave
local circumflex = diacritics.circum
local subscript = diacritics.subscript
local combining_diacritic = m_data.combining_diacritic
local UTF8_char = "[\1-\127\194-\244][\128-\191]*"
local basic_Greek = "[\206-\207][\128-\191]" -- excluding first line of Greek and Coptic block: ͰͱͲͳʹ͵Ͷͷͺͻͼͽ;Ϳ
local find = m_string_utils.find
local match = m_string_utils.match
local gmatch = m_string_utils.gmatch
local sub = mw.ustring.sub
local gsub = m_string_utils.gsub
local lower = m_string_utils.lower
local toNFC = mw.ustring.toNFC
local decompose = mw.ustring.toNFD
local info = {}
-- The tables are shared among different characters so that they can be checked
-- for equality if needed, and to use less space.
local vowel_t = { vowel = true }
local iota_t = { vowel = true, offglide = true }
local upsilon_t = { vowel = true, offglide = true }
-- These don't need any contents.
local rho_t = {}
-- local consonant_t = {}
local diacritic_t = { diacritic = true }
-- Needed for equality comparisons.
local breathing_t = { diacritic = true }
local function add_info(characters, t)
if type(characters) == "string" then
for character in string.gmatch(characters, UTF8_char) do
info[character] = t
end
else
for i, character in ipairs(characters) do
info[character] = t
end
end
end
add_info({ macron, breve,
diaeresis,
acute, grave, circumflex,
subscript,
}, diacritic_t)
add_info({rough, smooth}, breathing_t)
add_info("ΑΕΗΟΩαεηοω", vowel_t)
add_info("Ιι", iota_t)
add_info("Υυ", upsilon_t)
-- add_info("ΒΓΔΖΘΚΛΜΝΞΠΡΣΤΦΧΨϜϘϺϷͶϠβγδζθκλμνξπρσςτφχψϝϙϻϸͷϡ", consonant_t)
add_info("Ρρ", rho_t)
local not_recognized = {}
setmetatable(info, { __index =
function(t, key)
return not_recognized
end
})
local sparseConcat = require("Module:table").sparseConcat
local checkType = require "libraryUtil".checkType
local function _check(funcName)
return function(argIndex, arg, expectType, nilOk)
return checkType(funcName, argIndex, arg, expectType, nilOk)
end
end
-- Perform a function on each Unicode character in a string.
local function forEach(str, func)
for char in string.gmatch(str, UTF8_char) do
func(char)
end
end
-- This concatenates or inserts a character, then removes it from the text.
local function add(list, index, chars, text)
if not chars then
error("The function add cannot act on a nil character.")
end
if list[index] then
list[index] = list[index] .. chars
else
list[index] = chars
end
-- Basic string function works here.
return text:sub(#chars + 1)
end
function export.tag(term, face)
return m_script_utils.tag_text(term, lang, sc, face)
end
function export.link(term, face, alt, tr)
return m_links.full_link( { term = term, alt = alt, lang = lang, sc = sc, tr = tr }, face)
end
local function linkNoTag(term, alt)
return m_links.language_link{ term = term, lang = lang, alt = alt }
end
-- Convert spacing to combining diacritics, and nonstandard to standard polytonic Greek.
function export.standardDiacritics(text)
text = decompose(text)
text = text:gsub(UTF8_char, conversions)
return text
end
--[=[ This function arranges diacritics in the following order:
1. macron or breve
2. breathings or diaeresis
3. acute, circumflex, or grave
4. iota subscript
Used by [[Module:typing-aids]].
Returns an error if a sequence of diacritics contains more than one
of each category.
]=]
local function reorderDiacriticSequence(diacritics)
local output = {}
forEach(diacritics,
function (diacritic)
local index = diacritic_order[diacritic]
if not output[index] then
output[index] = diacritic
else
-- Place breve after macron.
if diacritic == breve then
index = index + 1
end
-- The following might have odd results when there
-- are three or more diacritics.
table.insert(output, index, diacritic)
-- [[Special:WhatLinksHere/Wiktionary:Tracking/grc-utils/too many diacritics]]
require("Module:debug").track("grc-utils/too many diacritics")
--[[
local m_templates = require("Module:grc-utilities/templates")
error("There are two diacritics, " ..
m_templates.addDottedCircle(output[index]) .. " and " ..
m_templates.addDottedCircle(diacritic) ..
" that belong in the same position. There should be only one."
)
--]]
end
end)
return sparseConcat(output)
end
function export.reorderDiacritics(text)
local d = diacritics
return (gsub(decompose(text),
combining_diacritic .. combining_diacritic .. "+",
reorderDiacriticSequence))
end
--[=[
This breaks a word into meaningful "tokens", which are
individual letters or diphthongs with their diacritics.
Used by [[Module:grc-accent]] and [[Module:grc-pronunciation]].
--]=]
local function make_tokens(text)
local tokens, prev_info = {}, {}
local token_i, vowel_count = 1, 0 -- Vowel count tracks .
local prev
for character in string.gmatch(decompose(text), UTF8_char) do
local curr_info = info[character]
-- Split vowels between tokens if not a diphthong.
if curr_info.vowel then
vowel_count = vowel_count + 1
if prev and (not (vowel_count == 2 and curr_info.offglide and prev_info.vowel)
-- υυ → υ, υ
-- ιυ → ι, υ
or prev_info.offglide and curr_info == upsilon_t or curr_info == prev_info) then
token_i = token_i + 1
if prev_info.vowel then
vowel_count = 1
end
elseif vowel_count == 2 then
vowel_count = 0
end
tokens[token_i] = (tokens[token_i] or "") .. character
elseif curr_info.diacritic then
vowel_count = 0
tokens[token_i] = (tokens[token_i] or "") .. character
if prev_info.diacritic or prev_info.vowel then
if character == diaeresis then
-- Split the diphthong in the current token if a diaeresis was found:
-- the first letter, then the second letter plus any diacritics.
local previous_vowel, vowel_with_diaeresis =
string.match(tokens[token_i],
"^(" .. basic_Greek .. ")(" .. basic_Greek .. ".+)")
if previous_vowel then
tokens[token_i], tokens[token_i + 1] = previous_vowel, vowel_with_diaeresis
token_i = token_i + 1
else
-- The vowel preceding the vowel with the diaeresis will already be
-- placed in the previous token if it has a diacritic:
-- Περικλῆῐ̈ → Π ε ρ ι κ λ ῆ ῐ̈
--[[
mw.log('Diaeresis was found in ' .. text .. ', but the previous token ' ..
require("Module:Unicode data").add_dotted_circle(tokens[token_i]) ..
' couldn’t be split because it does not consist of two Basic Greek characters followed by other characters.')
--]]
end
end
elseif prev_info == rho_t then
if curr_info ~= breathing_t then
mw.log(string.format("The character %s in %s should not have the accent %s on it.",
prev, text, require("Module:grc-utilities/templates").addDottedCircle(character)))
end
else
mw.log("The character " .. prev .. " cannot have a diacritic on it.")
end
else
vowel_count = 0
if prev then
token_i = token_i + 1
end
tokens[token_i] = (tokens[token_i] or "") .. character
end
prev = character
prev_info = curr_info
end
return tokens
end
local cache = {}
function export.tokenize(text)
local decomposed = decompose(text)
if not cache[decomposed] then
cache[decomposed] = make_tokens(text)
end
return cache[decomposed]
end
--[=[ Places diacritics in the following order:
1. breathings or diaeresis
2. acute, circumflex, or grave
3. macron or breve
4. iota subscript
Used by [[Module:grc-pronunciation]]. ]=]
function export.pronunciationOrder(text)
text = export.standardDiacritics(text)
if find(text, groups[1]) then
text = gsub(text,
diacritic .. diacritic .. "+",
function(sequence)
-- Put breathing and diaeresis first, then accents, then macron or breve
return table.concat{
match(sequence, groups[2]) or "",
match(sequence, groups[3]) or "",
match(sequence, groups[1]) or "",
match(sequence, groups[4]) or ""
}
end)
text = gsub(text, macron, spacing_macron) -- combining to spacing macron
text = gsub(text, breve, spacing_breve) -- combining to spacing breve
end
return toNFC(text)
end
-- Returns a table of any ambiguous vowels in the text, language-tagged.
function export.findAmbig(text, noTag)
if (not text) or type(text) ~= "string" then
error("The input to function findAmbig is nonexistent or not a string")
end
local lengthDiacritic = "[" .. macron .. breve .. circumflex .. subscript .. "]"
local aiu_diacritic = "^([" .. "αιυ" .. "])(" .. diacritic .. "*)$"
-- breaks the word into units
local output, vowels = {}, {}
for _, token in ipairs(export.tokenize(text)) do
if not find(token, m_data.consonant) then
local vowel, diacritics = match(
token,
aiu_diacritic
)
if vowel and (diacritics == "" or
not find(diacritics, lengthDiacritic)) then
local diacriticked_vowel
if not noTag then
diacriticked_vowel = export.tag(vowel .. diacritics)
else
diacriticked_vowel = vowel
end
table.insert(output, diacriticked_vowel)
-- Lists the vowel letters that are ambiguous, for categorization purposes.
vowels[lower(vowel)] = true
end
end
end
return output, vowels
end
return export