Module:Infobox date field metadata
Tú kanst upprætta kunningarsíðu um hetta modulið á Module:Infobox date field metadata/doc
local main = {};
local monthIndices = {
['januar'] = 1,
['februar'] = 2,
['mars'] = 3,
['apríl'] = 4,
['mai'] = 5,
['juni'] = 6,
['juli'] = 7,
['august'] = 8,
['september'] = 9,
['oktober'] = 10,
['november'] = 11,
['desember'] = 12,
['jan'] = 1,
['feb'] = 2,
['mar'] = 3,
['apr'] = 4,
--['mai'] = 5, -- long one would have caught this already
['jun'] = 6,
['jul'] = 7,
['aug'] = 8,
['sep'] = 9,
['okt'] = 10,
['nov'] = 11,
['des'] = 12
}
local monthDays = {
[1] = 31,
[2] = 29, -- will check below
[3] = 31,
[4] = 30,
[5] = 31,
[6] = 30,
[7] = 31,
[8] = 31,
[9] = 30,
[10] = 31,
[11] = 30,
[12] = 31
}
function checkIfDayValid(day, month, year)
-- First check that the month can have at least this many days
if (day > monthDays[month]) then return false end
-- February leap year check
if (month == 2) then
-- On Feb 29, if we don't have a year (incomplete date), assume 29 can be valid once every
if (day == 29 and not year) then return true end
-- On Feb 29, check for valid year - every 4 but not 100 but 400
if (day == 29 and not ((year % 4 == 0) and (year % 100 ~= 0) or (year % 400 == 0))) then
return false
end
end
return true
end
function checkIfMonthValid(month)
return month ~= 0 and month <= 12 -- <0 never happens with [0-9] pattern
end
function checkIfYearValid(year)
return year >= 1583 -- up to 9999
end
function checkIfHourValid(hour)
return hour < 24 -- <0 never happens with [0-9] pattern
end
function checkIfMinuteValid(minute)
return minute < 60 -- <0 never happens with [0-9] pattern
end
function checkIfSecondValid(second)
return second < 60 -- <0 never happens with [0-9] pattern
end
local PARSERESULT_OKAY = 1
local PARSERESULT_FAIL = 2 -- whatever we encountered isn't expected for any pattern
local PARSERESULT_UNCRECOGNIZED = 3 -- 14 May 12, 2014 (all elements okay, no pattern)
local PARSERESULT_INCOMPLETE = 4 -- May 3
local PARSERESULT_INCOMPLETERANGE = 5 -- 3 May 2012 - June 2013
local PARSERESULT_INVALID = 6 -- May 32
local PARSERESULT_INVALIDRANGE = 7 -- May 3 - May 2
-- This will first verify that we have a valid date and time and then output an ISO datetime string
function checkAndOutput(year, month, day, hour, minute, second, year2, month2, day2, hour2, minute2, second2)
local s
if (year and not checkIfYearValid(year)) then return PARSERESULT_INVALID; end
if (month and not checkIfMonthValid(month)) then return PARSERESULT_INVALID; end
if (day and not checkIfDayValid(day, month, year)) then return PARSERESULT_INVALID; end
if (hour and not checkIfHourValid(hour)) then return PARSERESULT_INVALID; end
if (minute and not checkIfMinuteValid(minute)) then return PARSERESULT_INVALID; end
if (second and not checkIfSecondValid(second)) then return PARSERESULT_INVALID; end
if (year2 and not checkIfYearValid(year2)) then return PARSERESULT_INVALID; end
if (month2 and not checkIfMonthValid(month2)) then return PARSERESULT_INVALID; end
if (day2 and not checkIfDayValid(day2, month2, year2)) then return PARSERESULT_INVALID; end
if (hour2 and not checkIfHourValid(hour2)) then return PARSERESULT_INVALID; end
if (minute2 and not checkIfMinuteValid(minute2)) then return PARSERESULT_INVALID; end
if (second2 and not checkIfSecondValid(second2)) then return PARSERESULT_INVALID; end
-- Check that end date is actually after start date
if (year2 and year) then
if (year2 < year) then return PARSERESULT_INVALIDRANGE end
if (year2 == year) then
if (month2 and month) then
if (month2 < month) then return PARSERESULT_INVALIDRANGE end
if (month2 == month) then
if (day2 and day) then
if (day2 < day) then return PARSERESULT_INVALIDRANGE end
-- TODO: compare time
end
end
end
end
end
-- Check that the date is actually complete even if valid
if (month and month2 and not year) then return PARSERESULT_INCOMPLETERANGE end -- any of 'd-dM', 'dM-dM', 'Md-d', 'Md-Md'
if (month and not year) then return PARSERESULT_INCOMPLETE end -- 'May', 'May 15', '15 May'
if (month2 and not year2) then return PARSERESULT_INCOMPLETE end -- same but other end
-- While technically there are more cases, none should have been matched and been given to us
local date1, time1, date2, time2
-- time only
if (second and not year) then time1 = string.format('%02d:%02d:%02d', hour, minute, second)
elseif (minute and not year) then time1 = string.format('%02d:%02d', hour, minute)
elseif (hour and not year) then time1 = string.format('%02d', hour)
-- date and time
elseif (second) then date1 = string.format('%d-%02d-%02d', year, month, day) time1 = string.format('%02d:%02d:%02d', hour, minute, second)
elseif (minute) then date1 = string.format('%d-%02d-%02d', year, month, day) time1 = string.format('%02d:%02d', hour, minute)
elseif (hour) then date1 = string.format('%d-%02d-%02d', year, month, day) time1 = string.format('%02d', hour)
-- date only
elseif (day) then date1 = string.format('%d-%02d-%02d', year, month, day)
elseif (month) then date1 = string.format('%d-%02d', year, month)
elseif (year) then date1 = string.format('%d', year)
end
-- time only
if (second2 and not year2) then time2 = string.format('%02d:%02d:%02d', hour2, minute2, second2)
elseif (minute2 and not year2) then time2 = string.format('%02d:%02d', hour2, minute2)
elseif (hour2 and not year2) then time2 = string.format('%02d', hour2)
-- date and time
elseif (second2) then date2 = string.format('%d-%02d-%02d', year2, month2, day2) time2 = string.format('%02d:%02d:%02d', hour2, minute2, second2)
elseif (minute2) then date2 = string.format('%d-%02d-%02d', year2, month2, day2) time2 = string.format('%02d:%02d', hour2, minute2)
elseif (hour2) then date2 = string.format('%d-%02d-%02d', year2, month2, day2) time2 = string.format('%02d', hour2)
-- date only
elseif (day2) then date2 = string.format('%d-%02d-%02d', year2, month2, day2)
elseif (month2) then date2 = string.format('%d-%02d', year2, month2)
elseif (year2) then date2 = string.format('%d', year2)
end
return PARSERESULT_OKAY, date1, time1, date2, time2 -- this function wouldn't be called withotu matching pattern, so at least 1 value should have been filled
end
function periodHourAdd(period)
if (period == 'pm' or period == 'p.m' or period == 'pm.' or period == 'p.m.') then -- random '.' is pattern match artifact
return 12
else
return 0
end
end
local seekString -- this is our local seek string, so we don't have to pass it as parameter every time
local currentPosition -- this keeps track of where we are in seeking our current string
-- These are the element type "constants" for readability mostly
local ELEMENT_INVALID = 1
local ELEMENT_ONETWODIGITS = 2 -- '1' '12' '01'
local ELEMENT_FOURDIGITS = 3 -- '1234'
local ELEMENT_WHITESPACE = 4 -- ' ' ' '
local ELEMENT_MONTHWORD = 5 -- 'May' 'February' 'Aug'
local ELEMENT_COMMA = 6 -- ',' ', '
local ELEMENT_DASH = 7 -- '-' ' - ' ' — ' '- ' ' -'
local ELEMENT_DATESEPARATOR = 8 -- '-'
local ELEMENT_TIMESEPARATOR = 9 -- ':'
local ELEMENT_TIMEPERIOD = 10 -- 'am' 'p.m.'
local ELEMENT_PERIODWHITESPACE = 11 -- '.' or '. '
local ELEMENT_ONETWODIGITSWITHORDINAL = 12 -- '12th' '3rd'
function seekNextElement()
-- Profiler says mw.ustring.find is the bottleneck, probably because it's unicode; not sure how to improve though besides writing my own pattern matcher
-- Digits with letters
local foundPositionStart, foundPositionEnd, foundMatch, foundMatch2 = mw.ustring.find(seekString, '^([0-9]+)([a-z]+)%.?', currentPosition)
if (foundPositionStart) then
--currentPosition = foundPositionEnd + 1 -- this is our new start location -- only if we return
-- Additionally check how many digits we actually have, as arbitrary number isn't valid
if (#foundMatch <= 2) then -- most likely a day number
if (foundMatch2 == 'st' or foundMatch2 == 'nd' or foundMatch2 == 'rd' or foundMatch2 == 'th') then -- won't bother checking against a number, no false positives that I saw in 120k cases
currentPosition = foundPositionEnd + 1 -- this is our new start location (forced to do this here, since we don't always return)
return ELEMENT_ONETWODIGITSWITHORDINAL, tonumber(foundMatch), (currentPosition > mw.ustring.len(seekString))
--else -- let it capture digits again, this time '10am' '8p.m.' will be separate
-- return ELEMENT_INVALID -- not a valid ordinal indicator
end
--else -- let it capture digits again, this time '10am' '8p.m.' will be separate
-- return ELEMENT_INVALID -- just the invalid, the number of digits (3+) won't match any patterns
end
end
-- Digits
local foundPositionStart, foundPositionEnd, foundMatch = mw.ustring.find(seekString, '^([0-9]+)', currentPosition)
if (foundPositionStart) then
currentPosition = foundPositionEnd + 1 -- this is our new start location
-- Additionally check how many digits we actually have, as arbitrary number isn't valid
if (#foundMatch <= 2) then -- most likely a day number or time number
return ELEMENT_ONETWODIGITS, tonumber(foundMatch), (currentPosition > mw.ustring.len(seekString))
elseif (#foundMatch == 4) then -- most likely a year
return ELEMENT_FOURDIGITS, tonumber(foundMatch), (currentPosition > mw.ustring.len(seekString))
else
return ELEMENT_INVALID -- just the invalid, the number of digits (3 or 5+) won't match any patterns
end
end
-- Time period - a.m./p.m. (before letters)
foundPositionStart, foundPositionEnd, foundMatch = mw.ustring.find(seekString, '^%s*([ap]%.?m%.?)', currentPosition)
if (foundPositionStart) then
currentPosition = foundPositionEnd + 1 -- this is our new start location
return ELEMENT_TIMEPERIOD, foundMatch, (currentPosition > mw.ustring.len(seekString))
end
-- Word
foundPositionStart, foundPositionEnd, foundMatch = mw.ustring.find(seekString, '^([A-Za-z]+)', currentPosition)
if (foundPositionStart) then
currentPosition = foundPositionEnd + 1 -- this is our new start location
if (#foundMatch >= 3) then
-- Find the possible month name index
monthIndex = monthIndices[mw.ustring.lower(foundMatch)]
if (monthIndex) then
return ELEMENT_MONTHWORD, monthIndex, (currentPosition > mw.ustring.len(seekString))
else
return ELEMENT_INVALID -- just the invalid, the word didn't match a valid month name
end
else
-- TODO LETTERS
return ELEMENT_INVALID -- just the invalid, the word was too short to be valid month name
end
end
-- Time separator (colon without whitespace)
foundPositionStart, foundPositionEnd, foundMatch = mw.ustring.find(seekString, '^(:)', currentPosition)
if (foundPositionStart) then
currentPosition = foundPositionEnd + 1 -- this is our new start location
return ELEMENT_TIMESEPARATOR, foundMatch, (currentPosition > mw.ustring.len(seekString))
end
-- Comma and any following whitespace
foundPositionStart, foundPositionEnd, foundMatch = mw.ustring.find(seekString, '^(,%s*)', currentPosition)
if (foundPositionStart) then
currentPosition = foundPositionEnd + 1 -- this is our new start location
return ELEMENT_COMMA, foundMatch, (currentPosition > mw.ustring.len(seekString))
end
-- Period and any following whitespace ('Feb. 2010' or '29. June')
foundPositionStart, foundPositionEnd, foundMatch = mw.ustring.find(seekString, '^(%.%s*)', currentPosition)
if (foundPositionStart) then
currentPosition = foundPositionEnd + 1 -- this is our new start location
return ELEMENT_PERIODWHITESPACE, foundMatch, (currentPosition > mw.ustring.len(seekString))
end
-- Dash with possible whitespace or Date separator (dash without whitespace)
-- Both non-breaking spaces - ' - '
foundPositionStart, foundPositionEnd, foundMatch = mw.ustring.find(seekString, '^( [%-–—] )', currentPosition)
if (foundPositionStart) then
currentPosition = foundPositionEnd + 1 -- this is our new start location
return ELEMENT_DASH, foundMatch, (currentPosition > mw.ustring.len(seekString))
end
-- Non-breaking space - ' - '
foundPositionStart, foundPositionEnd, foundMatch = mw.ustring.find(seekString, '^( [%-–—]%s*)', currentPosition)
if (foundPositionStart) then
currentPosition = foundPositionEnd + 1 -- this is our new start location
return ELEMENT_DASH, foundMatch, (currentPosition > mw.ustring.len(seekString))
end
-- Dash entity code and both non-breaking spaces - ' – '
foundPositionStart, foundPositionEnd, foundMatch = mw.ustring.find(seekString, '^( &[nm]dash; )', currentPosition)
if (foundPositionStart) then
currentPosition = foundPositionEnd + 1 -- this is our new start location
return ELEMENT_DASH, foundMatch, (currentPosition > mw.ustring.len(seekString))
end
-- Dash entity code and non-breaking space - ' – '
foundPositionStart, foundPositionEnd, foundMatch = mw.ustring.find(seekString, '^( &[nm]dash;%s*)', currentPosition)
if (foundPositionStart) then
currentPosition = foundPositionEnd + 1 -- this is our new start location
return ELEMENT_DASH, foundMatch, (currentPosition > mw.ustring.len(seekString))
end
-- Dash entity code - ' – '
foundPositionStart, foundPositionEnd, foundMatch = mw.ustring.find(seekString, '^(%s*&[nm]dash;%s*)', currentPosition)
if (foundPositionStart) then
currentPosition = foundPositionEnd + 1 -- this is our new start location
return ELEMENT_DASH, foundMatch, (currentPosition > mw.ustring.len(seekString))
end
-- Regular whitespace
foundPositionStart, foundPositionEnd, foundMatch = mw.ustring.find(seekString, '^(%s*[%-–—]%s*)', currentPosition)
if (foundPositionStart) then
currentPosition = foundPositionEnd + 1 -- this is our new start location
if (foundMatch == '-') then -- nothing else is date separator, no hyphens no stuff like that
return ELEMENT_DATESEPARATOR, foundMatch, (currentPosition > mw.ustring.len(seekString))
else
return ELEMENT_DASH, foundMatch, (currentPosition > mw.ustring.len(seekString)) -- we will actually need to check for DATESEPARATOR as well, as that one stole the '-' case
end
end
-- Whitespace (after all others that capture whitespace)
foundPositionStart, foundPositionEnd, foundMatch = mw.ustring.find(seekString, '^(%s+)', currentPosition)
if (foundPositionStart) then
currentPosition = foundPositionEnd + 1 -- this is our new start location
return ELEMENT_WHITESPACE, foundMatch, (currentPosition > mw.ustring.len(seekString))
end
-- Whitespace -- same as above but using (other that for dashes)
foundPositionStart, foundPositionEnd, foundMatch = mw.ustring.find(seekString, '^ ', currentPosition)
if (foundPositionStart) then
currentPosition = foundPositionEnd + 1 -- this is our new start location
return ELEMENT_WHITESPACE, foundMatch, (currentPosition > mw.ustring.len(seekString))
end
return ELEMENT_INVALID -- just the invalid, we won't be parsing this further
end
function parseDateString(input)
-- Reset our seek string and position
seekString = input
currentPosition = 1
local elements = {}
local values = {}
-- Seek the entire string now
local numberOfElements = 0
repeat
foundElement, foundValue, eos = seekNextElement()
-- If we found something we can't process, return as unparsable
if (foundElement == ELEMENT_INVALID) then return nil end
numberOfElements = numberOfElements + 1
elements[numberOfElements] = foundElement
values[numberOfElements] = foundValue
until eos
--[[
local s = input .. ' -> ' .. numberOfElements .. ' elements: '
for currentElementIndex = 1, numberOfElements do
s = s .. ' #' .. elements[currentElementIndex] .. '=' .. values[currentElementIndex]
end
do return s end
]]
-- Now comes an uber-deep if-then-else tree
-- This is roughly the most efficient step-by-step parsing, something like log(N)
-- Doing each combination via pattern/"Regex" is way slower
-- Having each combination a clean function/preset means checking every element, so way slower
-- Only immediate big improvement is to only seekNextElement() when actually checking that deep, though this will make a (even bigger) mess
if (elements[1] == ELEMENT_ONETWODIGITS or elements[1] == ELEMENT_ONETWODIGITSWITHORDINAL) then -- '3' or '10' or '12th'
if (elements[2] == ELEMENT_WHITESPACE or elements[2] == ELEMENT_PERIODWHITESPACE) then -- '3 ' or '3. '
if (elements[3] == ELEMENT_MONTHWORD) then -- '3 May'
if (numberOfElements == 3) then return checkAndOutput(nil, values[3], values[1], nil, nil, nil) end
if (elements[4] == ELEMENT_WHITESPACE or elements[4] == ELEMENT_PERIODWHITESPACE or elements[4] == ELEMENT_COMMA) then -- '3 May ' or '3 Feb. ' or '3 May, '
if (elements[5] == ELEMENT_FOURDIGITS) then -- '3 May 2013'
if (numberOfElements == 5) then return checkAndOutput(values[5], values[3], values[1], nil, nil, nil) end
if (elements[6] == ELEMENT_WHITESPACE or elements[6] == ELEMENT_COMMA) then -- '3 May 2013, '
if (elements[7] == ELEMENT_ONETWODIGITS) then -- '3 May 2013, 10'
if (elements[8] == ELEMENT_TIMEPERIOD) then -- '3 May 2013, 10 am'
if (numberOfElements == 8) then return checkAndOutput(values[5], values[3], values[1], values[7] + periodHourAdd(values[8]), nil, nil) end
elseif (elements[8] == ELEMENT_TIMESEPARATOR) then -- '3 May 2013, 10:'
if (elements[9] == ELEMENT_ONETWODIGITS) then -- '3 May 2013, 10:38'
if (numberOfElements == 9) then return checkAndOutput(values[5], values[3], values[1], values[7], values[9], nil) end
if (elements[10] == ELEMENT_TIMEPERIOD) then -- '3 May 2013, 10:38 am'
if (numberOfElements == 10) then return checkAndOutput(values[5], values[3], values[1], values[7] + periodHourAdd(values[10]), values[9], nil) end
elseif (elements[10] == ELEMENT_TIMESEPARATOR) then -- '3 May 2013, 10:38:'
if (elements[11] == ELEMENT_ONETWODIGITS) then -- '3 May 2013, 10:38:27'
if (numberOfElements == 11) then return checkAndOutput(values[5], values[3], values[1], values[7], values[9], values[11]) end
if (elements[12] == ELEMENT_TIMEPERIOD) then -- '3 May 2013, 10:38:27 am'
if (numberOfElements == 12) then return checkAndOutput(values[5], values[3], values[1], values[7] + periodHourAdd(values[12]), values[9], values[11]) end
end
end
end
end
end
end
elseif (elements[6] == ELEMENT_DASH or elements[6] == ELEMENT_DATESEPARATOR) then -- '3 May 2013 - '
if (elements[7] == ELEMENT_ONETWODIGITS or elements[7] == ELEMENT_ONETWODIGITSWITHORDINAL) then -- '3 May 2013 - 12' or '3rd May 2013 - 12th'
if (elements[8] == ELEMENT_WHITESPACE) then -- '3 May 2013 - 12 '
if (elements[9] == ELEMENT_MONTHWORD) then -- '3 May 2013 - 12 February'
if (elements[10] == ELEMENT_WHITESPACE) then -- '3 May 2013 - 12 February '
if (elements[11] == ELEMENT_FOURDIGITS) then -- '3 May 2013 - 12 February 2014'
if (numberOfElements == 11) then return checkAndOutput(values[5], values[3], values[1], nil, nil, nil, values[11], values[9], values[7], nil, nil, nil) end
end
end
end
end
end
end
end
elseif (elements[4] == ELEMENT_DASH or elements[4] == ELEMENT_DATESEPARATOR) then -- '3 May - '
if (elements[5] == ELEMENT_ONETWODIGITS or elements[5] == ELEMENT_ONETWODIGITSWITHORDINAL) then -- '3 May - 12' or '3rd May - 12th'
if (elements[6] == ELEMENT_WHITESPACE) then -- '3 May - 12 '
if (elements[7] == ELEMENT_MONTHWORD) then -- '3 May - 12 October'
if (numberOfElements == 7) then return checkAndOutput(nil, values[3], values[1], nil, nil, nil, nil, values[7], values[5], nil, nil, nil) end
if (elements[8] == ELEMENT_COMMA or elements[8] == ELEMENT_WHITESPACE) then -- '3 May - 12 October '
if (elements[9] == ELEMENT_FOURDIGITS) then -- '3 May - 12 October 2013'
if (numberOfElements == 9) then return checkAndOutput(values[9], values[3], values[1], nil, nil, nil, values[9], values[7], values[5], nil, nil, nil) end
end
end
end
end
end
end
end
elseif (elements[2] == ELEMENT_DASH or elements[2] == ELEMENT_DATESEPARATOR) then -- '3 - '
if (elements[3] == ELEMENT_ONETWODIGITS or elements[3] == ELEMENT_ONETWODIGITSWITHORDINAL) then -- '3 - 12' or '3rd - 12th'
if (elements[4] == ELEMENT_WHITESPACE) then -- '3 - 12 '
if (elements[5] == ELEMENT_MONTHWORD) then -- '3 - 12 May'
if (numberOfElements == 5) then return checkAndOutput(nil, values[5], values[1], nil, nil, nil, nil, values[5], values[3], nil, nil, nil) end
if (elements[6] == ELEMENT_COMMA or elements[6] == ELEMENT_WHITESPACE) then -- '3 - 12 May '
if (elements[7] == ELEMENT_FOURDIGITS) then -- '3 - 12 May 2013'
if (numberOfElements == 7) then return checkAndOutput(values[7], values[5], values[1], nil, nil, nil, values[7], values[5], values[3], nil, nil, nil) end
end
end
end
end
end
end
-- Here's a case where we want to optimize or rather add readability and trim redundancy
-- Basically, any time '10am', '10:28', '10:28am', '10:28:27', '10:28:27am' can be followed by a date, which means 5 copies of 30+ lines (urgh)
-- Instead we will only check once, but using a different element index offset if needed, so date might start at element 4 or 5 or 6 etc.
-- Currently we only have '10', but if it turns out to be a time we will be checking if it's followed by a date
local wasTime = false -- by default we didn't find a valid time syntax, we have '10' and that's not a time by itself without 'am/pm' or further precision
local possibleHour, possibleMinute, possibleSecond -- temporary values that we will fill as far as we can when parsing time and use for time+date combo if one is found
local i = 0 -- this is our offset from the closest possible location for date seeking
if (elements[2] == ELEMENT_TIMESEPARATOR and elements[1] ~= ELEMENT_ONETWODIGITSWITHORDINAL) then -- '10:' but not '10th:'
possibleHour = values[1] -- only once we see ':' (or 'am' below) it is likely a time
if (elements[3] == ELEMENT_ONETWODIGITS) then -- '10:28'
if (numberOfElements == 3) then return checkAndOutput(nil, nil, nil, values[1], values[3], nil) end
possibleMinute = values[3]
wasTime = true -- this is a valid final time, so we can check date appended to this
i = 1 -- '10', ':' and '28' are three elements, so we start 1 further from 3
if (elements[4] == ELEMENT_TIMESEPARATOR) then -- '10:28:'
wasTime = false -- a time can't end with separator, so if this is last match, we aren't appending any dates
if (elements[5] == ELEMENT_ONETWODIGITS) then -- '10:28:27'
if (numberOfElements == 5) then return checkAndOutput(nil, nil, nil, values[1], values[3], values[5]) end
possibleSecond = values[5]
wasTime = true -- this is a valid final time, so we can check date appended to this
i = 3 -- '10', ':', '28', ':' and '27' are five elements, so we start 3 further from 3
if (elements[6] == ELEMENT_TIMEPERIOD) then -- '10:28:27 am'
if (numberOfElements == 6) then return checkAndOutput(nil, nil, nil, values[1] + periodHourAdd(values[6]), values[3], values[5]) end
possibleHour = values[1] + periodHourAdd(values[6]) -- hour now needs possible adjusting since we saw a time period
-- wasTime = true -- already set
i = 4 -- '10', ':', '28', ':', '27' and 'am' are six elements, so we start 4 further from 3
end
end
elseif (elements[4] == ELEMENT_TIMEPERIOD) then -- '10:28 am'
if (numberOfElements == 4) then return checkAndOutput(nil, nil, nil, values[1] + periodHourAdd(values[4]), values[3], nil) end
wasTime = true -- this is a valid final time, so we can check date appended to this
possibleHour = values[1] + periodHourAdd(values[4]) -- hour now needs possible adjusting since we saw a time period
i = 2 -- '10', ':', '28' and 'am' are four elements, so we start 2 further from 3
end
end
elseif (elements[2] == ELEMENT_TIMEPERIOD) then -- '10 am'
if (numberOfElements == 2) then return checkAndOutput(nil, nil, nil, values[1] + periodHourAdd(values[2]), nil, nil) end
possibleHour = values[1] + periodHourAdd(values[2]) -- only once we see 'am' (or ':' above) it is likely a time
wasTime = true -- this is a valid final time, so we can check date appended to this
i = 0 -- '10' and 'am' are two elements, so we start at 3 - default
end
if (wasTime) then -- '10am', '10:28', '10:28am', '10:28:27', '10:28:27am' (using just '10:28:27...' below)
-- Now we will try to append a date to the time
if (elements[3+i] == ELEMENT_WHITESPACE or elements[3+i] == ELEMENT_COMMA) then -- '10:28:27, '
if (elements[4+i] == ELEMENT_ONETWODIGITS) then -- '10:28:27, 3'
if (elements[5+i] == ELEMENT_WHITESPACE) then -- '10:28:27, 3 '
if (elements[6+i] == ELEMENT_MONTHWORD) then -- '10:28:27, 3 May'
if (elements[7+i] == ELEMENT_WHITESPACE) then -- '10:28:27, 3 May '
if (elements[8+i] == ELEMENT_FOURDIGITS) then -- '10:28:27, 3 May 2013'
if (numberOfElements == 8+i) then return checkAndOutput(values[8+i], values[6+i], values[4+i], possibleHour, possibleMinute, possibleSecond) end
end
end
end
end
elseif (elements[4+i] == ELEMENT_MONTHWORD) then -- '10:28:27, May'
if (elements[5+i] == ELEMENT_WHITESPACE) then -- '10:28:27, May '
if (elements[6+i] == ELEMENT_ONETWODIGITS) then -- '10:28:27, May 3'
if (elements[7+i] == ELEMENT_COMMA or elements[7+i] == ELEMENT_WHITESPACE) then -- '10:28:27, May 3, '
if (elements[8+i] == ELEMENT_FOURDIGITS) then -- '10:28:27, May 3, 2013'
if (numberOfElements == 8+i) then return checkAndOutput(values[8+i], values[4+i], values[6+i], possibleHour, possibleMinute, possibleSecond) end
end
end
end
end
elseif (elements[4+i] == ELEMENT_FOURDIGITS) then -- '10:28:27, 2013'
if (elements[5+i] == ELEMENT_DATESEPARATOR) then -- '10:28:27, 2013-'
if (elements[6+i] == ELEMENT_ONETWODIGITS) then -- '10:28:27, 2013-05'
if (elements[7+i] == ELEMENT_DATESEPARATOR) then -- '10:28:27, 2013-05-'
if (elements[8+i] == ELEMENT_ONETWODIGITS) then -- '10:28:27, 2013-05-03'
if (numberOfElements == 8+i) then return checkAndOutput(values[4+i], values[6+i], values[8+i], possibleHour, possibleMinute, possibleSecond) end
end
end
end
end
end
end
end
elseif (elements[1] == ELEMENT_FOURDIGITS) then -- '2013'
if (numberOfElements == 1) then return checkAndOutput(values[1], nil, nil, nil, nil, nil) end
if (elements[2] == ELEMENT_DATESEPARATOR) then -- '2013-'
if (elements[3] == ELEMENT_ONETWODIGITS) then -- '2013-05'
--if (numberOfElements == 3) then return checkAndOutput(values[1], values[3], nil, nil, nil, nil) end
-- This is actually ambiguous -- 2008-12 can be years 2008 to 2012 or it could be Decemeber 2008; few cases, so just ignoring
if (elements[4] == ELEMENT_DATESEPARATOR) then -- '2013-05-'
if (elements[5] == ELEMENT_ONETWODIGITS) then -- '2013-05-03'
if (numberOfElements == 5) then return checkAndOutput(values[1], values[3], values[5], nil, nil, nil) end
if (elements[6] == ELEMENT_WHITESPACE or elements[6] == ELEMENT_COMMA) then -- '2013-05-03, '
if (elements[7] == ELEMENT_ONETWODIGITS) then -- '2013-05-03, 10'
if (elements[8] == ELEMENT_TIMEPERIOD) then -- '2013-05-03, 10 am'
if (numberOfElements == 8) then return checkAndOutput(values[1], values[3], values[5], values[7] + periodHourAdd(values[8]), nil, nil) end
elseif (elements[8] == ELEMENT_TIMESEPARATOR) then -- '2013-05-03, 10:'
if (elements[9] == ELEMENT_ONETWODIGITS) then -- '2013-05-03, 10:38'
if (numberOfElements == 9) then return checkAndOutput(values[1], values[3], values[5], values[7], values[9], nil) end
if (elements[10] == ELEMENT_TIMEPERIOD) then -- '2013-05-03, 10:38 am'
if (numberOfElements == 10) then return checkAndOutput(values[1], values[3], values[5], values[7] + periodHourAdd(values[10]), values[9], nil) end
elseif (elements[10] == ELEMENT_TIMESEPARATOR) then -- '2013-05-03, 10:38:'
if (elements[11] == ELEMENT_ONETWODIGITS) then -- '2013-05-03, 10:38:27'
if (numberOfElements == 11) then return checkAndOutput(values[1], values[3], values[5], values[7], values[9], values[11]) end
if (elements[12] == ELEMENT_TIMEPERIOD) then -- '2013-05-03, 10:38:27 am'
if (numberOfElements == 12) then return checkAndOutput(values[1], values[3], values[5], values[7] + periodHourAdd(values[12]), values[9], values[11]) end
end
end
end
end
end
end
end
end
end
end
end -- can't elseif, because we have ELEMENT_DATESEPARATOR, which repeats above
if (elements[2] == ELEMENT_DASH or elements[2] == ELEMENT_DATESEPARATOR) then -- '2013 - '
if (elements[3] == ELEMENT_FOURDIGITS) then -- '2013 - 2014'
if (numberOfElements == 3) then return checkAndOutput(values[1], nil, nil, nil, nil, nil, values[3], nil, nil, nil, nil, nil) end
end
elseif (elements[2] == ELEMENT_WHITESPACE or elements[2] == ELEMENT_COMMA) then -- '2013 ' or '2013, '
if (elements[3] == ELEMENT_MONTHWORD) then -- '2013 May'
if (numberOfElements == 3) then return checkAndOutput(values[1], values[3], nil, nil, nil, nil) end
-- 2013 May - 2013 April (let's see first if this is ever used real-world)
if (elements[4] == ELEMENT_WHITESPACE) then -- '2013 May '
if (elements[5] == ELEMENT_ONETWODIGITS) then -- '2013 May 15'
if (numberOfElements == 5) then return checkAndOutput(values[1], values[3], values[5], nil, nil, nil) end
end
end
elseif (elements[3] == ELEMENT_ONETWODIGITS) then -- '2013 15' or '2013, 15'
if (elements[4] == ELEMENT_WHITESPACE) then -- '2013 15 '
if (elements[5] == ELEMENT_MONTHWORD) then -- '2013 15 May'
if (numberOfElements == 5) then return checkAndOutput(values[1], values[5], values[3], nil, nil, nil) end
end
end
end
end
elseif (elements[1] == ELEMENT_MONTHWORD) then -- 'May'
if (numberOfElements == 1) then return checkAndOutput(nil, values[1], nil, nil, nil, nil) end
if (elements[2] == ELEMENT_WHITESPACE or elements[2] == ELEMENT_PERIODWHITESPACE) then -- 'May ' or 'Feb. '
if (elements[3] == ELEMENT_ONETWODIGITS or elements[3] == ELEMENT_ONETWODIGITSWITHORDINAL) then -- 'May 3' or 'May 3rd'
if (numberOfElements == 3) then return checkAndOutput(nil, values[1], values[3], nil, nil, nil) end
if (elements[4] == ELEMENT_COMMA or elements[4] == ELEMENT_WHITESPACE) then -- 'May 3, '
if (elements[5] == ELEMENT_FOURDIGITS) then -- 'May 3, 2013'
if (numberOfElements == 5) then return checkAndOutput(values[5], values[1], values[3], nil, nil, nil) end
if (elements[6] == ELEMENT_WHITESPACE or elements[6] == ELEMENT_COMMA) then -- ''May 3, 2013, '
if (elements[7] == ELEMENT_ONETWODIGITS) then -- ''May 3, 2013, 10'
if (elements[8] == ELEMENT_TIMEPERIOD) then -- ''May 3, 2013, 10 am'
if (numberOfElements == 8) then return checkAndOutput(values[5], values[1], values[3], values[7] + periodHourAdd(values[8]), nil, nil) end
elseif (elements[8] == ELEMENT_TIMESEPARATOR) then -- ''May 3, 2013, 10:'
if (elements[9] == ELEMENT_ONETWODIGITS) then -- ''May 3, 2013, 10:38'
if (numberOfElements == 9) then return checkAndOutput(values[5], values[1], values[3], values[7], values[9], nil) end
if (elements[10] == ELEMENT_TIMEPERIOD) then -- ''May 3, 2013, 10:38 am'
if (numberOfElements == 10) then return checkAndOutput(values[5], values[1], values[3], values[7] + periodHourAdd(values[10]), values[9], nil) end
elseif (elements[10] == ELEMENT_TIMESEPARATOR) then -- ''May 3, 2013, 10:38:'
if (elements[11] == ELEMENT_ONETWODIGITS) then -- ''May 3, 2013, 10:38:27'
if (numberOfElements == 11) then return checkAndOutput(values[5], values[1], values[3], values[7], values[9], values[11]) end
if (elements[12] == ELEMENT_TIMEPERIOD) then -- ''May 3, 2013, 10:38:27 am'
if (numberOfElements == 12) then return checkAndOutput(values[5], values[1], values[3], values[7] + periodHourAdd(values[12]), values[9], values[11]) end
end
end
end
end
end
end
elseif (elements[6] == ELEMENT_DASH or elements[6] == ELEMENT_DATESEPARATOR) then -- 'May 3, 2013 - '
if (elements[7] == ELEMENT_MONTHWORD) then -- 'May 3, 2013 - February'
if (elements[8] == ELEMENT_WHITESPACE) then -- 'May 3, 2013 - February '
if (elements[9] == ELEMENT_ONETWODIGITS or elements[3] == ELEMENT_ONETWODIGITSWITHORDINAL) then -- 'May 3, 2013 - February 12' or 'May 3rd, 2013 - February 12th'
if (elements[10] == ELEMENT_COMMA or elements[10] == ELEMENT_WHITESPACE) then -- 'May 3, 2013 - February 12, '
if (elements[11] == ELEMENT_FOURDIGITS) then -- 'May 3, 2013 - February 12, 2014'
if (numberOfElements == 11) then return checkAndOutput(values[5], values[1], values[3], nil, nil, nil, values[11], values[7], values[9], nil, nil, nil) end
end
end
end
end
end
end
end
elseif (elements[4] == ELEMENT_DASH or elements[4] == ELEMENT_DATESEPARATOR) then -- 'May 3 - '
if (elements[5] == ELEMENT_MONTHWORD) then -- 'May 3 - June'
if (elements[6] == ELEMENT_WHITESPACE) then -- 'May 3 - June '
if (elements[7] == ELEMENT_ONETWODIGITS or elements[3] == ELEMENT_ONETWODIGITSWITHORDINAL) then -- 'May 3 - June 12' or 'May 3rd - June 12th'
if (numberOfElements == 7) then return checkAndOutput(nil, values[1], values[3], nil, nil, nil, nil, values[5], values[7], nil, nil, nil) end
if (elements[8] == ELEMENT_COMMA or elements[8] == ELEMENT_WHITESPACE) then -- 'May 3 - June 12, '
if (elements[9] == ELEMENT_FOURDIGITS) then -- 'May 3 - June 12, 2014'
if (numberOfElements == 9) then return checkAndOutput(values[9], values[1], values[3], nil, nil, nil, values[9], values[5], values[7], nil, nil, nil) end
end
end
end
end
elseif (elements[5] == ELEMENT_ONETWODIGITS or elements[3] == ELEMENT_ONETWODIGITSWITHORDINAL) then -- 'May 3 - 12' or 'May 3rd - 12th'
if (numberOfElements == 5) then return checkAndOutput(nil, values[1], values[3], nil, nil, nil, nil, values[1], values[5], nil, nil, nil) end
if (elements[6] == ELEMENT_COMMA or elements[6] == ELEMENT_WHITESPACE) then -- 'May 3 - 12, '
if (elements[7] == ELEMENT_FOURDIGITS) then -- 'May 3 - 12, 2013'
if (numberOfElements == 7) then return checkAndOutput(values[7], values[1], values[3], nil, nil, nil, values[7], values[1], values[5], nil, nil, nil) end
end
end
end
end
elseif (elements[3] == ELEMENT_FOURDIGITS) then -- 'May 2013'
if (numberOfElements == 3) then return checkAndOutput(values[3], values[1], nil, nil, nil, nil) end
if (elements[4] == ELEMENT_DASH or elements[4] == ELEMENT_DATESEPARATOR) then -- 'May 2013 -'
if (elements[5] == ELEMENT_MONTHWORD) then -- 'May 2013 - June'
if (elements[6] == ELEMENT_WHITESPACE) then -- 'May 2013 - June '
if (elements[7] == ELEMENT_FOURDIGITS) then -- 'May 2013 - June 2013'
if (numberOfElements == 7) then return checkAndOutput(values[3], values[1], nil, nil, nil, nil, values[7], values[5], nil, nil, nil, nil) end
end
end
end
end
end
elseif (elements[2] == ELEMENT_DASH or elements[2] == ELEMENT_DATESEPARATOR) then -- 'May - '
if (elements[3] == ELEMENT_MONTHWORD) then -- 'May - June'
if (elements[4] == ELEMENT_WHITESPACE) then -- 'May - June '
if (elements[5] == ELEMENT_FOURDIGITS) then -- 'May - June 2013'
if (numberOfElements == 5) then return checkAndOutput(values[5], values[1], nil, nil, nil, nil, values[5], values[3], nil, nil, nil, nil) end
end
end
end
elseif (elements[2] == ELEMENT_COMMA) then -- 'May, '
if (elements[3] == ELEMENT_FOURDIGITS) then -- 'May, 2012'
if (numberOfElements == 3) then return checkAndOutput(values[3], values[1], nil, nil, nil, nil) end
end
end
else
return PARSERESULT_UNRECOGNIZED -- the combination of elements was not a recognized one
end
end
function hasMetadataTemplates(input)
-- This is a basic list of the template names for metadata emiting tempaltes, there are inr eality more templates and more redirects
if (string.match(input, '%{%{[Ss]tart[ %-]?date')) then return true end
if (string.match(input, '%{%{[Ee]nd[ %-]?date')) then return true end
if (string.match(input, '%{%{[Bb]irth[ %-]?date')) then return true end
if (string.match(input, '%{%{[Dd]eath[ %-]?date')) then return true end
if (string.match(input, '%{%{[Bb]irth[ %-]?year')) then return true end
if (string.match(input, '%{%{[Dd]eath[ %-]?year')) then return true end
if (string.match(input, '%{%{[Ff]ilm ?date')) then return true end
if (string.match(input, '%{%{[ISO[ %-]date')) then return true end
return false
end
-- This function will return a raw string for generic checks and unit test, including defined parse errors
function main.parseDateOutputRaw(frame)
local input = frame.args[1]
-- If desired (default), unstrip and decode to have the raw markup
if (not frame.args.noUnstrip or frame.args.noUnstrip ~= 'yes') then
input = mw.text.decode(mw.text.unstrip(input))
end
-- If desired (not default), strip the field extra stuff
if (frame.args.stripExtras and frame.args.stripExtras == 'yes') then
input = stripFieldExtras(input)
end
-- If there is nothing but whitespace, don't bother
if (mw.ustring.match(input, '^%s*$')) then if (frame.args[2] == 'pretty') then return frame:preprocess('\'\'{{gray|Empty input}}\'\'') else return 'Empty input' end end
local result, startDate, startTime, endDate, endTime = parseDateString(input)
if (result == PARSERESULT_FAIL) then if (frame.args[2] == 'pretty') then return frame:preprocess('\'\'{{gray|Failed parse}}\'\'') else return 'Failed parse' end end
if (result == PARSERESULT_UNRECOGNIZED) then
local s
if (hasMetadataTemplates(input)) then s = 'Has metadata template' else s = 'Unrecognized pattern' end
if (frame.args[2] == 'pretty') then return frame:preprocess('\'\'{{gray|'..s..'}}\'\'') else return s end
end
if (result == PARSERESULT_INVALID) then if (frame.args[2] == 'pretty') then return frame:preprocess('\'\'{{maroon|Invalid date/time}}\'\'') else return 'Invalid date/time' end end
if (result == PARSERESULT_INVALIDRANGE) then if (frame.args[2] == 'pretty') then return frame:preprocess('\'\'{{maroon|Invalid date range}}\'\'') else return 'Invalid date range' end end
if (result == PARSERESULT_INCOMPLETE) then if (frame.args[2] == 'pretty') then return frame:preprocess('\'\'{{cyan|Incomplete date}}\'\'') else return 'Incomplete date' end end
if (result == PARSERESULT_INCOMPLETERANGE) then if (frame.args[2] == 'pretty') then return frame:preprocess('\'\'{{cyan|Incomplete date range}}\'\'') else return 'Incomplete date range' end end
local s
if (startDate) then s = startDate end
if (startTime) then if (startDate) then s = s .. ' ' .. startTime else s = startTime end end
if (endDate) then s = s .. '; ' .. endDate end -- currently end date implies start date
-- currently no end time
return s
end
-- Strips whitespace from an input
function trim(value)
local strip, count = string.gsub(value, '^%s*(.-)%s*$', '%1')
return strip
end
function stripFieldExtras(value)
-- todo: do progressive scan like with that seek string just for ref tags and such
-- note that we can't just replace matches with whitespace and catch them all, because it could be like '3 August<!---->20<ref/>12'
local matchStrip = value:match('^([^<]-)<ref[^>]*>[^<]*</ref><ref[^>]*>[^<]*</ref><ref[^>]*>[^<]*</ref>$') -- basic refs (quite common)
if (matchStrip) then return trim(matchStrip) end
matchStrip = value:match('^([^<]-)<ref[^>]*>[^<]*</ref><ref[^>]*>[^<]*</ref>$') -- basic refs (quite common)
if (matchStrip) then return trim(matchStrip) end
matchStrip = value:match('^([^<]-)<ref[^>]*>[^<]*</ref>$') -- basic refs (quite common)
if (matchStrip) then return trim(matchStrip) end
matchStrip = value:match('^([^<]-)<ref[^>]*/><ref[^>]*/><ref[^>]*/>$') -- basic named ref (quite common)
if (matchStrip) then return trim(matchStrip) end
matchStrip = value:match('^([^<]-)<ref[^>]*/><ref[^>]*/>$') -- basic named ref (quite common)
if (matchStrip) then return trim(matchStrip) end
matchStrip = value:match('^([^<]-)<ref[^>]*/>$') -- basic named ref (quite common)
if (matchStrip) then return trim(matchStrip) end
matchStrip = value:match('^<!--.--->([^<]+)$') -- comment before (sometimes used for notes to editors [not yet seen metadata-related comment])
if (matchStrip) then return trim(matchStrip) end
matchStrip = value:match('^([^<]-)<!--.--->$') -- comment after (sometimes used for notes to editors)
if (matchStrip) then return trim(matchStrip) end
matchStrip = value:match('^{{[Ff]lag ?icon[^}]+}}%s*{{[Ff]lag ?icon[^}]+}}([^<]+)$') -- 2 flag icons (also more common than one would think)
if (matchStrip) then return trim(matchStrip) end
matchStrip = value:match('^{{[Ff]lag ?icon[^}]+}}([^<]+)$') -- flag icon (quite common, although against MOS:ICON)
if (matchStrip) then return trim(matchStrip) end
matchStrip = value:match('^([^<]-){{[Ff]lag ?icon[^}]+}}%s*{{[Ff]lag ?icon[^}]+}}$') -- after as well
if (matchStrip) then return trim(matchStrip) end
matchStrip = value:match('^([^<]-){{[Ff]lag ?icon[^}]+}}$')
if (matchStrip) then return trim(matchStrip) end
return value -- if we didn't match anything, abort now
end
function main.emitMetadata(frame)
local input = frame.args[1]
-- If desired (default), unstrip and decode to have the raw markup
if (not frame.args.noUnstrip or frame.args.noUnstrip ~= 'yes') then
input = mw.text.decode(mw.text.unstrip(input))
end
input = stripFieldExtras(trim(input))
-- If there is nothing but whitespace, don't bother
if (mw.ustring.match(input, '^%s*$')) then return nil end
-- Then parse the date and see if we get a valid output date
local result, startDate, startTime, endDate, endTime = parseDateString(input)
if (not frame.args.noErrorCats or frame.args.noErrorCats ~= 'yes') then
if (result == PARSERESULT_FAIL and not hasMetadataTemplates(input)) then return frame:preprocess('<includeonly>[[Category:Articles that could not be parsed for automatic date metadata]]</includeonly>') end
if (result == PARSERESULT_UNRECOGNIZED and not hasMetadataTemplates(input)) then return frame:preprocess('<includeonly>[[Category:Articles that could not be parsed for automatic date metadata]]</includeonly>') end
--if (result == PARSERESULT_INVALID) then return frame:preprocess('<includeonly>[[Category:]]</includeonly>') end
--if (result == PARSERESULT_INVALIDRANGE) then return frame:preprocess('<includeonly>[[Category:]]</includeonly>') end
if (result == PARSERESULT_INCOMPLETE) then return frame:preprocess('<includeonly>[[Category:Articles with incomplete dates for automatic metadata]]</includeonly>') end
if (result == PARSERESULT_INCOMPLETERANGE) then return frame:preprocess('<includeonly>[[Category:Articles with incomplete date ranges for automatic metadata]]</includeonly>') end
-- we need to use frame:preprocess() for <includeonly> or it just gets displayed
end
-- We are only doing the rest for a valid date
if (result ~= PARSERESULT_OKAY) then return nil end
local dtstartSpan, dtendSpan
-- If we have a start value and we're told to output it
if ((startDate or startTime) and frame.args.dtstart and frame.args.dtstart == 'yes') then
if (startDate and startTime) then dtstartSpan = '<span class="dtstart">' .. startDate .. 'T' .. startTime .. '</span>'
elseif (startDate) then dtstartSpan = '<span class="dtstart">' .. startDate .. '</span>'
else dtstartSpan = '<span class="dtstart">' .. startTime .. '</span>' end
end
-- If we have an end value and we're told to output it
if ((endDate or endTime) and frame.args.dtend and frame.args.dtend == 'yes') then -- end values only happen when start values happen
if (endDate and endTime) then dtendSpan = '<span class="dtend">' .. endDate .. 'T' .. endTime .. '</span>'
elseif (endDate) then dtendSpan = '<span class="dtend">' .. endDate .. '</span>'
else dtendSpan = '<span class="dtend">' .. endTime .. '</span>' end
end
local trackingCat = ''
if (frame.args.trackingCat and frame.args.trackingCat == 'yes') then
trackingCat = '[[Category:Articles with automatically detected infobox date metadata]]'
end
if (dtstartSpan and dtendSpan) then return '<span style="display:none"> (' .. dtstartSpan .. ' - ' .. dtendSpan .. ')</span>' .. trackingCat
elseif (dtstartSpan) then return '<span style="display:none"> (' .. dtstartSpan .. ')</span>' .. trackingCat
elseif (dtendSpan) then return '<span style="display:none"> (' .. dtendSpan .. ')</span>' .. trackingCat
else return nil end
end
function main.outputRawStripped(frame)
return stripFieldExtras(trim(mw.text.decode(mw.text.unstrip(frame.args[1]))))
end
function main.reoutputme(frame)
if (frame.args.preprocess and frame.args.preprocess == 'yes') then
return frame:preprocess(mw.text.decode(mw.text.unstrip(frame.args[1]))) .. '[[Category:Dummy]]'
else
return mw.text.decode(mw.text.unstrip(frame.args[1])) .. '[[Category:Dummy]]'
end
--[[local input = mw.text.decode(mw.text.unstrip(frame.args[1]))
s = 'Len=' .. #input .. ' '
for i = 1, string.len(input) do
s = s .. string.sub(input, i, i) .. ' '
end
return s]]
end
return main