mirror of
https://gitlab.com/parroquia-san-leandro/cancionero-25.git
synced 2024-11-23 11:50:20 +01:00
1070 lines
39 KiB
Lua
1070 lines
39 KiB
Lua
-- Copyright (C) 2017 Kevin W. Hamlen
|
|
--
|
|
-- This program 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
|
|
-- of the License, or (at your option) any later version.
|
|
--
|
|
-- This program 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 this program; if not, write to the Free Software
|
|
-- Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston,
|
|
-- MA 02110-1301, USA.
|
|
--
|
|
-- The latest version of this program can be obtained from
|
|
-- http://songs.sourceforge.net.
|
|
|
|
|
|
VERSION = "3.0"
|
|
BIBLEDEFAULT = "bible.can"
|
|
|
|
-- fileopen(<filename>)
|
|
-- Open <filename> for reading, returning a filestate table on success or
|
|
-- nil on failure.
|
|
function fileopen(fnam)
|
|
local handle
|
|
if fnam ~= "-" then
|
|
local msg,errno
|
|
handle,msg,errno = io.open(fnam, "r")
|
|
if not handle then
|
|
io.stderr:write("songidx: Unable to open ",fnam," for reading.\n",
|
|
"Error ",errno,": ",msg,"\n")
|
|
return nil
|
|
end
|
|
else
|
|
handle = io.stdin
|
|
fnam = "stdin"
|
|
end
|
|
return { f=handle, filename=fnam, lineno=0 }
|
|
end
|
|
|
|
-- filereadln(<fstate>)
|
|
-- Read a line of input, returning the string or nil if at file-end.
|
|
function filereadln(fs)
|
|
fs["lineno"] = fs["lineno"] + 1
|
|
local s = fs["f"]:read()
|
|
if not s then return nil end
|
|
return s
|
|
end
|
|
|
|
-- closeout(<handle>)
|
|
-- Close an output file <handle>.
|
|
function closeout(handle)
|
|
if handle == io.stdout then handle:flush()
|
|
else handle:close() end
|
|
end
|
|
|
|
-- errorout(<handle>)
|
|
-- Respond to a write failure to <handle>.
|
|
function errorout(handle)
|
|
closeout(handle)
|
|
io.stderr:write("songidx: Error writing to output file. Aborting.\n")
|
|
return 2
|
|
end
|
|
|
|
-- fileclose(<fstate>)
|
|
-- Close file table <fstate>.
|
|
function fileclose(fs)
|
|
if (fs["f"] ~= io.stdin) then fs["f"]:close() end
|
|
fs["f"] = nil
|
|
fs["filename"] = nil
|
|
fs["lineno"] = 0
|
|
end
|
|
|
|
-- numprefix(<string>)
|
|
-- Find the longest prefix of <string> that represents a number (according to
|
|
-- the current locale) and contains no whitespace. Return the number and the
|
|
-- suffix of <string> not parsed. If there is no such prefix, return false
|
|
-- and the original <string>.
|
|
function numprefix(s)
|
|
local _, len = unicode.utf8.find(s,"^%S*")
|
|
for i=len, 1, -1 do
|
|
local n = tonumber(s:sub(1,i))
|
|
if n then return n, s:sub(i+1) end
|
|
end
|
|
return false, s
|
|
end
|
|
|
|
-- cleantitle(<song>)
|
|
-- Remove macros and braces from <song>'s title, and convert it to uppercase.
|
|
-- Macro-spaces ("\ ") are converted to regular spaces (" "). Cache the
|
|
-- result to avoid re-cleaning during sorting.
|
|
function cleantitle(s)
|
|
if not s["clean"] then
|
|
local t = s["title"]:gsub("\\[^%a%s]","")
|
|
:gsub("\\(%s)","%1")
|
|
:gsub("\\%a+%s*","")
|
|
:gsub("{%s*","")
|
|
:gsub("}","")
|
|
s["clean"] = unicode.utf8.upper(t)
|
|
end
|
|
return s["clean"]
|
|
end
|
|
|
|
-- songcmp(<song1>,<song2>)
|
|
-- Return true if <song1> is less than <song2>, and false otherwise. The
|
|
-- ordering is first by title, then by index. This function is suitable for
|
|
-- use with table.sort().
|
|
function songcmp(s1,s2)
|
|
local t1 = cleantitle(s1)
|
|
local t2 = cleantitle(s2)
|
|
|
|
while true do
|
|
-- Find the next word or number in each string.
|
|
t1, t2 = unicode.utf8.match(t1,"%w.*"), unicode.utf8.match(t2,"%w.*")
|
|
|
|
-- If there is no next word/number in both, sort by index. If there is
|
|
-- no next word/number in one but not the other, sort the shorter string
|
|
-- before the longer one.
|
|
if not t1 then
|
|
if not t2 then break end
|
|
return true
|
|
elseif not t2 then return false end
|
|
|
|
-- If one is a number, sort the number before the word. If both are
|
|
-- numbers (and are not equal), then sort in numerical order.
|
|
local n1, n2
|
|
n1, t1 = numprefix(t1)
|
|
n2, t2 = numprefix(t2)
|
|
if n1 then
|
|
if not n2 or n1 < n2 then return true end
|
|
if n1 > n2 then return false end
|
|
elseif n2 then return false
|
|
else
|
|
-- Otherwise, both are words. Lexicographically compare the words
|
|
-- according to the current locale's collation conventions. If the
|
|
-- locale considers them "equal" (i.e., w1<w2 is false but w1<=w2 is true)
|
|
-- then continue the loop to consider the next pair of words.
|
|
local w1, w2 = unicode.utf8.match(t1,"^[%a'`]*"),
|
|
unicode.utf8.match(t2,"^[%a'`]*")
|
|
local diff = w1 < w2
|
|
if diff or not (w1 <= w2) then return diff end
|
|
t1, t2 = t1:sub(#w1+1), t2:sub(#w2+1)
|
|
end
|
|
end
|
|
|
|
-- If all corresponding words/numbers are equal, then sort alternate-form
|
|
-- entries (e.g., lyrics) after normal entries (e.g., titles)
|
|
local alt1, alt2 = s1["title"]:sub(1,1), s2["title"]:sub(1,1)
|
|
if alt1 == "*" and alt2 ~= "*" then return false end
|
|
if alt1 ~= "*" and alt2 == "*" then return true end
|
|
|
|
-- If everything is the same, sort by the right-hand sides of the index
|
|
-- entries (e.g., the song or page numbers).
|
|
return s1["idx"] < s2["idx"]
|
|
end
|
|
|
|
-- setstartchars(<array>)
|
|
-- Try to find canonical "start characters" for each block of songs in a
|
|
-- sorted <array> of songs. Since Lua doesn't currently have any means of
|
|
-- imparting a locale's alphabet, we adopt the following strategy: Extract
|
|
-- the first unicode character from each title in the sorted list of titles
|
|
-- until reaching one that the locale's collation algorithm says is "bigger"
|
|
-- than the last. Find the "smallest" first character in that set using
|
|
-- NON-UNICODE lexicographic sort. This tends to be the most canonical one,
|
|
-- since unicode tends to put canonical (e.g., no-accent) glyphs at lower
|
|
-- code points than non-canonical (e.g., accented) ones. (Unfortunately, if
|
|
-- none of the titles in the block start with the desired canonical glyph,
|
|
-- there's no way to guess it; we just use the best one available.)
|
|
function setstartchars(songs)
|
|
local start = 1
|
|
local best
|
|
|
|
for i=1, #songs do
|
|
local c = unicode.utf8.match(cleantitle(songs[i]),"%w")
|
|
if c then
|
|
c = unicode.utf8.upper(c)
|
|
if not best then
|
|
songs[start]["newblk"], start, best = "\\#", i, c
|
|
elseif best < c then
|
|
songs[start]["newblk"], start, best = best, i, c
|
|
elseif #c < #best then best = c
|
|
elseif #c == #best then
|
|
for j=1,#c do
|
|
if c:byte(j) < best:byte(j) then best = c; break
|
|
elseif c:byte(j) > best:byte(j) then break end
|
|
end
|
|
end
|
|
elseif best then
|
|
songs[start]["newblk"], start, best = best, i, nil
|
|
end
|
|
end
|
|
|
|
if start <= #songs then
|
|
if best then
|
|
songs[start]["newblk"] = best
|
|
else
|
|
songs[start]["newblk"] = "\\#"
|
|
end
|
|
end
|
|
end
|
|
|
|
prelist = { A=true, THE=true }
|
|
wt_and = { AND=true }
|
|
wt_by = { BY=true }
|
|
wt_unknown = { UNKNOWN=true }
|
|
|
|
-- rotate(<title>)
|
|
-- If the first word of <title> is any word in prelist, then shift that word
|
|
-- to the end of the string, preceded by a comma and a space. So for example,
|
|
-- if prelist contains "The", then rotate("The title") returns "Title, The".
|
|
-- Words in prelist are matched case-insensitively, and the new first word
|
|
-- becomes capitalized. If <title> begins with the marker character '*',
|
|
-- that character is ignored and left unchanged.
|
|
function rotate(s)
|
|
local t = unicode.utf8.upper(s)
|
|
local n = 0
|
|
if s:sub(1,1) == "*" then n = 1 end
|
|
for pre in pairs(prelist) do
|
|
if t:sub(1+n,n+#pre) == pre and
|
|
unicode.utf8.find(t, "^%s+%S", n+#pre+1) then
|
|
local len = unicode.utf8.len(pre)
|
|
local x, y, z = unicode.utf8.match(unicode.utf8.sub(s,n+len+1),"^%s+(%W*)(%w?)(.*)$")
|
|
return s:sub(1,n) .. x .. unicode.utf8.upper(y) .. z:gsub("\\%s","%0\1"):match("^(.-)%s*$"):gsub("\1","") .. ",~" .. unicode.utf8.sub(s,1+n,n+len)
|
|
end
|
|
end
|
|
return s
|
|
end
|
|
|
|
-- matchany(<string>,<init>,<wordtable>)
|
|
-- If a word in <wordtable> case-insensitively matches to <string> starting
|
|
-- at index <init>, and if the match concludes with whitespace, then return
|
|
-- the index of the whitespace; otherwise return <init>.
|
|
function matchany(s,init,wt)
|
|
local t = s:sub(init)
|
|
local u = unicode.utf8.upper(t)
|
|
for w,_ in pairs(wt) do
|
|
if u:sub(1,#w) == w and
|
|
(#w == #u or unicode.utf8.find(u, "^%s", #w+1)) then
|
|
return init + #(unicode.utf8.sub(t,1,unicode.utf8.len(w)))
|
|
end
|
|
end
|
|
return init
|
|
end
|
|
|
|
-- issuffix(<string>,<init>)
|
|
-- Return true if the abbreviation "Jr" or a roman numeral (followed by
|
|
-- nothing, space, period, comma, or semicolon) appears at position <init>
|
|
-- within <string>. Return false otherwise.
|
|
function issuffix(s,i)
|
|
return unicode.utf8.find(s, "^Jr$", i) or
|
|
unicode.utf8.find(s, "^Jr[%s,;%.]", i) or
|
|
unicode.utf8.find(s, "^[IVX]+$", i) or
|
|
unicode.utf8.find(s, "^[IVX]+[%s,;%.]", i)
|
|
end
|
|
|
|
-- grabauthor(<string>)
|
|
-- Return a string of the form "Sirname, Restofname" denoting the full name
|
|
-- of the first author found in <string>; or return nil if no author name
|
|
-- can be found. Also return an index to the suffix of <string> that was
|
|
-- not parsed, and just the "Sirname" part as a stand-alone string.
|
|
--
|
|
-- Precondition: Caller must first sanitize <string> as follows:
|
|
-- <string>:gsub("[\1\2\3]","")
|
|
-- :gsub("\\\\","\\\1"):gsub("\\{","\\\2"):gsub("\\}","\\\3")
|
|
--
|
|
-- Postcondition: Caller must then unsanitize the returned <string> with:
|
|
-- <string>:gsub("\1","\\"):gsub("\2","{"):gsub("\3","}")
|
|
--
|
|
-- This is to allow grabauthor() to safely use Lua's %b pattern to find
|
|
-- balanced brace pairs without getting confused by escaped braces.
|
|
--
|
|
-- Heuristics:
|
|
-- * Names are separated by punctuation (other than hyphens, periods,
|
|
-- apostrophes, or backslashes) or by the word "and" (or whatever words
|
|
-- are in wt_and).
|
|
-- Special case: If a comma is followed by the abbreviation "Jr" or by a
|
|
-- roman numeral, then the comma does NOT end the author's name.
|
|
-- * If a name contains the word "by" (or anything in wt_by), then
|
|
-- everything before it is not considered part of the name. (Let's hope
|
|
-- nobody is named "By".)
|
|
-- * The author's last name is always the last capitalized word in the
|
|
-- name unless the last capitalized word is "Jr." or a roman numeral.
|
|
-- In that case the author's last name is the penultimate captialized
|
|
-- word.
|
|
-- * If an author appears to have only a first name, or if the last name
|
|
-- found according to the above heuristics is an abbreviation (ending in
|
|
-- a period), look ahead in <string> until we find someone with a last
|
|
-- name and use that one. This allows us to identify the first author in
|
|
-- a string like "Joe, Billy E., and Bob Smith" to be "Joe Smith".
|
|
-- * If the resultant name contains the word "unknown" (or any word in
|
|
-- wt_unknown), it's probably not a real name. Recursively attempt
|
|
-- to parse the next author.
|
|
function grabauthor(authline,i)
|
|
i = unicode.utf8.find(authline, "[^%s,;]", i)
|
|
if not i then return nil end
|
|
i = unicode.utf8.find(authline, "%S", matchany(authline,i,wt_and))
|
|
if not i then return nil end
|
|
i = unicode.utf8.find(authline, "%S", matchany(authline,i,wt_by))
|
|
if not i then return nil end
|
|
local skip = (matchany(authline,i,wt_unknown) > i)
|
|
|
|
-- Set "first" to the index of the start of the first name, "last" to the
|
|
-- index of the start of the sirname, "suffix" to the index of any suffix
|
|
-- like "Jr." or "III" (or nil if there is none), and i to the index of the
|
|
-- first character beyond the end of this author's name.
|
|
local first,last,suffix = i,nil,nil
|
|
while i <= #authline do
|
|
while true do
|
|
local j = select(2,authline:find("^\\%A", i)) or
|
|
select(2,authline:find("^\\%a+%s*", i)) or
|
|
select(2,authline:find("^%b{}", i))
|
|
if not j then break end
|
|
i = j + 1
|
|
end
|
|
if i > #authline then break
|
|
elseif authline:sub(i,i) == "," then
|
|
local j = unicode.utf8.find(authline, "%S", i+1)
|
|
if j and issuffix(authline,j) then i = i + 1
|
|
else break end
|
|
elseif authline:sub(i,i) == ";" then break
|
|
elseif unicode.utf8.find(authline, "^%s", i) then
|
|
i = unicode.utf8.find(authline, "%S", i)
|
|
if not i then i = #authline + 1; break end
|
|
if matchany(authline, i, wt_and) > i then break end
|
|
skip = skip or (matchany(authline, i, wt_unknown) > i)
|
|
local j = matchany(authline, i, wt_by)
|
|
if j > i then
|
|
j = unicode.utf8.find(authline, "%S", j)
|
|
if not j then last = i; break end -- last name of "By"?
|
|
i,first,last,suffix = j,j,nil,nil
|
|
elseif issuffix(authline,i) then
|
|
suffix = i
|
|
elseif unicode.utf8.find(authline:sub(i):gsub("\\%A",""):gsub("\\%a+%s*",""),"^[%s{}'`\"]*%u") then
|
|
last,suffix = i,nil
|
|
end
|
|
else
|
|
i = select(2, unicode.utf8.find(authline, ".", i)) + 1
|
|
end
|
|
end
|
|
|
|
-- If an "unknown" word appeared, skip this entry and parse the next.
|
|
if skip then return grabauthor(authline,i) end
|
|
|
|
-- Find the sirname.
|
|
local sirname, fullname
|
|
if last then
|
|
sirname = unicode.utf8.gsub(authline:sub(last,(suffix or i)-1),
|
|
"([^%s,;\\])[%s,;]+$", "%1")
|
|
end
|
|
if not sirname or unicode.utf8.find(sirname, "%a%.$") then
|
|
-- Here's where it gets tough. We either have a single-word name, or the
|
|
-- last name ends in a "." which means maybe it's just a middle initial or
|
|
-- other abbreviation. We could be dealing with a line like, "Billy,
|
|
-- Joe E., and Bob Smith", in which case we have to go searching for the
|
|
-- real last name. To handle this case, we will try a recursive call.
|
|
local _,_,r = grabauthor(authline,i)
|
|
if r or not sirname then
|
|
fullname = unicode.utf8.gsub(authline:sub(first,i-1),
|
|
"([^%s,;\\])[%s,;]+$", "%1")
|
|
if r then return (r .. ", " .. fullname), i, r
|
|
else return fullname, i, nil end
|
|
end
|
|
end
|
|
|
|
-- Add the first name.
|
|
fullname = sirname .. ", " ..
|
|
unicode.utf8.gsub(authline:sub(first,(last or suffix or i)-1),
|
|
"([^%s,;\\])[%s,;]+$", "%1")
|
|
|
|
-- Add the suffix, if any.
|
|
if suffix then
|
|
fullname = fullname .. " " ..
|
|
unicode.utf8.gsub(authline:sub(suffix,i-1), "([^%s,;\\])[%s,;]+$", "%1")
|
|
end
|
|
|
|
return fullname, i, sirname
|
|
end
|
|
|
|
-- genindex(<fstate>,<outname>)
|
|
-- Reads a title (if <authorindex>=false) or author (if <authorindex>=true)
|
|
-- index data file from file table <fstate> and generates a new file named
|
|
-- <outfile> containing a LaTeX title/author index.
|
|
-- Return Value: 0 on success, 1 on warnings, or 2 on failure
|
|
function genindex(fs,outname,authorindex)
|
|
local songs = {}
|
|
local seen = {}
|
|
local wt = { wt_sep, wt_after, wt_prefix, wt_ignore }
|
|
local typ = authorindex and "author" or "title"
|
|
|
|
io.stderr:write("songidx: Parsing ",typ," index data file ",fs["filename"],"...\n")
|
|
|
|
while true do
|
|
local buf = filereadln(fs)
|
|
if not buf then break end
|
|
if buf:sub(1,1) == "%" then
|
|
local j = buf:match("^()%%sep ") or
|
|
buf:match("^%%()after ") or
|
|
buf:match("^%%p()refix ") or
|
|
buf:match("^%%ig()nore ")
|
|
if j then
|
|
if not seen[j] then wt[j], seen[j] = {}, true end
|
|
wt[j][unicode.utf8.upper(buf:sub(buf:find(" ")+1))] = true
|
|
end
|
|
else
|
|
local snum = filereadln(fs)
|
|
if not snum then
|
|
io.stderr:write("songidx:",fs["filename"],":",fs["lineno"],": incomplete song entry (orphan ",typ,")\n")
|
|
return 2
|
|
end
|
|
local link = filereadln(fs)
|
|
if not link then
|
|
io.stderr:write("songidx:",fs["filename"],":",fs["lineno"],": incomplete song entry (missing hyperlink)\n")
|
|
return 2
|
|
end
|
|
if authorindex then
|
|
local i,a = 1
|
|
buf = buf:gsub("[\1\2\3]",""):gsub("\\\\","\\\1")
|
|
:gsub("\\{","\\\2"):gsub("\\}","\\\3")
|
|
while true do
|
|
a,i = grabauthor(buf, i)
|
|
if not a then break end
|
|
a = a:gsub("\1","\\"):gsub("\2","{"):gsub("\3","}")
|
|
table.insert(songs, {title=a, num=snum, linkname=link, idx=#songs})
|
|
end
|
|
else
|
|
buf = rotate(unicode.utf8.gsub(unicode.utf8.gsub(buf,"([^%s\\])%s+$","%1"),"^(%*?)%s+","%1"))
|
|
table.insert(songs, {title=buf, num=snum, linkname=link, idx=#songs})
|
|
end
|
|
end
|
|
end
|
|
fileclose(fs)
|
|
|
|
-- Sort the song table.
|
|
table.sort(songs, songcmp)
|
|
-- Find the index blocks.
|
|
setstartchars(songs)
|
|
|
|
-- Write the sorted data out to the output file.
|
|
io.stderr:write("songidx: Generating ",typ," index TeX file ",outname,"...\n")
|
|
local f,msg,errno
|
|
if outname == "-" then
|
|
f, outname = io.stdout, "stdout"
|
|
else
|
|
f,msg,errno = io.open(outname, "w")
|
|
if not f then
|
|
io.stderr:write("songidx: Unable to open ",outname," for writing.\n",
|
|
"Error ",errno,": ",msg,"\n")
|
|
return 2
|
|
end
|
|
end
|
|
for i=1, #songs do
|
|
if i>1 and songs[i]["title"] == songs[i-1]["title"] then
|
|
if not f:write("\\\\\\songlink{",songs[i]["linkname"],"}{",songs[i]["num"],"}") then return errorout(f) end
|
|
else
|
|
if songs[i]["newblk"] then
|
|
if i>1 then
|
|
if not f:write("}\n\\end{idxblock}\n") then return errorout(f) end
|
|
end
|
|
if not f:write("\\begin{idxblock}{",songs[i]["newblk"]) then return errorout(f) end
|
|
end
|
|
if songs[i]["title"]:find("^%*") then
|
|
if not f:write("}\n\\idxaltentry{",songs[i]["title"]:sub(2)) then return errorout(f) end
|
|
else
|
|
if not f:write("}\n\\idxentry{",songs[i]["title"]) then return errorout(f) end
|
|
end
|
|
if not f:write("}{\\songlink{",songs[i]["linkname"],"}{",songs[i]["num"],"}") then return errorout(f) end
|
|
end
|
|
end
|
|
if #songs > 0 then
|
|
if not f:write("}\n\\end{idxblock}\n") then return errorout(f) end
|
|
end
|
|
|
|
return 0
|
|
end
|
|
|
|
bible = {}
|
|
chapX = 0
|
|
|
|
-- readbible(<filename>)
|
|
-- Read bible data file <filename> into the bible table. Return nil on error
|
|
-- or true on success.
|
|
function readbible(filename)
|
|
local fs = fileopen(filename)
|
|
if not fs then return nil end
|
|
bible = {}
|
|
|
|
while true do
|
|
local buf = filereadln(fs)
|
|
if not buf then break end
|
|
if buf:sub(1,1) ~= "#" and buf:find("%S") then
|
|
local t, vbuf = { name = buf:match("^[^|]*"),
|
|
aliases = "|" .. unicode.utf8.upper(buf) .. "|" }
|
|
repeat
|
|
vbuf = filereadln(fs)
|
|
if not vbuf then
|
|
io.stderr:write("songidx:",fs["filename"],":",fs["lineno"],": incomplete bible book entry (book title with no verses)\n")
|
|
fileclose(fs)
|
|
return nil
|
|
end
|
|
until vbuf:sub(1,1) ~= "#" and buf:find("%S")
|
|
if vbuf:find("[^%d%s]") then
|
|
io.stderr:write("songidx:",fs["filename"],":",fs["lineno"],": verse count includes a non-digit\n")
|
|
fileclose(fs)
|
|
return nil
|
|
end
|
|
for n in vbuf:gmatch("%d+") do
|
|
local i = tonumber(n)
|
|
if not i then
|
|
io.stderr:write("songidx:",fs["filename"],":",fs["lineno"],": invalid number ",n,"\n")
|
|
fileclose(fs)
|
|
return nil
|
|
end
|
|
i = math.floor(i)
|
|
if chapX < i+1 then chapX = i+1 end
|
|
table.insert(t, i)
|
|
end
|
|
table.insert(bible, t)
|
|
end
|
|
end
|
|
|
|
fileclose(fs)
|
|
return true
|
|
end
|
|
|
|
-- parseref(<string>,<init>,<book>,<chapter>)
|
|
-- Interpret the characters starting at index <init> of <string> as a
|
|
-- scripture reference, and return four values: (1) the index of the first
|
|
-- character after <init> not parsed, (2) the book number parsed, (3) the
|
|
-- chapter number parsed (or 1 if the book has only verses), and (4) the
|
|
-- verse number parsed. Arguments <book> and <chapter> are the PREVIOUS book
|
|
-- number and chapter parsed, or -1 if none. If book or chapter information
|
|
-- is missing from <string>, they will be drawn from <book> and <chapter>.
|
|
-- That way, successive calls can correctly parse a run-on string like
|
|
-- "Philippians 3:1,5; 4:3", infering that "5" refers to "Philippians 3" and
|
|
-- "4:3" refers to "Philippians". If the parser encounters an error in
|
|
-- processing the book name (e.g., a book name was specified but not
|
|
-- recognized), then #bible+1 is returned for the book. If no chapter or no
|
|
-- verse is provided (e.g., the reference is just "Philippians" or
|
|
-- "Philippians 3") then the chapter and/or verse are returned as -1.
|
|
function parseref(s,i,book,ch)
|
|
local v = -1
|
|
i = unicode.utf8.find(s,"%S",i)
|
|
if not i then return nil end
|
|
local j = unicode.utf8.find(s,"[%d:]*%s*[,;%-]",i) or
|
|
unicode.utf8.find(s,"[%d:]*%s*$",i)
|
|
local bookname = "|" .. unicode.utf8.upper(unicode.utf8.match(s:sub(i,j-1), "(.-)%s*$")) .. "|"
|
|
i = j
|
|
if bookname ~= "||" then
|
|
book, ch = #bible+1, -1
|
|
for b,t in pairs(bible) do
|
|
if t["aliases"]:find(bookname, 1, true) then book = b; break end
|
|
end
|
|
end
|
|
j = unicode.utf8.find(s,"%D",i) or (#s+1)
|
|
if s:sub(j,j) == ":" then
|
|
ch, i = (tonumber(s:sub(i,j-1)) or -1), j+1
|
|
j = unicode.utf8.find(s,"%D",i) or (#s+1)
|
|
end
|
|
if ch<=0 and book>0 and #bible[book] == 1 then
|
|
-- Special case: This book has only one chapter.
|
|
ch = 1
|
|
end
|
|
if ch <= 0 then
|
|
ch = tonumber(s:sub(i,j-1)) or -1
|
|
else
|
|
v = tonumber(s:sub(i,j-1)) or -1
|
|
end
|
|
i = unicode.utf8.find(s,"%S",j)
|
|
if not i then i = #s+1
|
|
elseif not s:find("^[,;%-]",i) then return nil end
|
|
return i, book, ch, v
|
|
end
|
|
|
|
-- vlt(<chapter1>,<verse1>,<chapter2>,<verse2>)
|
|
-- Return true if <chapter1>:<verse1> precedes <chapter2>:<verse2> and false
|
|
-- otherwise.
|
|
function vlt(ch1,v1,ch2,v2)
|
|
return ch1 < ch2 or (ch1 == ch2 and v1 < v2)
|
|
end
|
|
|
|
-- vinc(<book>,<chapter>,<verse>)
|
|
-- Return the chapter,verse pair of the verse immediately following
|
|
-- <chapter>:<verse> in <book>. If <chapter>:<verse> is the last verse in
|
|
-- <book>, the returned chapter will not exist in <book>.
|
|
function vinc(book,ch,v)
|
|
if v < bible[book][ch] then return ch, v+1 end
|
|
return ch+1, 1
|
|
end
|
|
|
|
-- vdec(<book>,<chapter>,<verse>)
|
|
-- Return the chapter,verse pair of the verse immediately preceding
|
|
-- <chapter>:<verse> in <book>. If <chapter> and <verse> are both 1,
|
|
-- then 0,nil will be returned.
|
|
function vdec(book,ch,v)
|
|
if v > 1 then return ch, v-1 end
|
|
return ch-1, bible[book][ch-1]
|
|
end
|
|
|
|
-- unpack_cv(<cv>)
|
|
-- Decompose a chapter-verse key (computed as cv*chapX+v) into the original
|
|
-- chapter and verse numbers.
|
|
function unpack_cv(cv)
|
|
local v = cv % chapX
|
|
return (cv-v)/chapX, v
|
|
end
|
|
|
|
-- eqdom(<table1>,<table2>)
|
|
-- Return true if two tables have identical domains; false otherwise.
|
|
function eqdom(t1,t2)
|
|
for k,_ in pairs(t1) do if not t2[k] then return false end end
|
|
for k,_ in pairs(t2) do if not t1[k] then return false end end
|
|
return true
|
|
end
|
|
|
|
-- insertref(<is_add>,<changeset>,<chapter>,<verse>,<song>,<link>,<key>)
|
|
-- Insert song <song>,<link>,<key> into the set of "adds" (if <is_add>=true)
|
|
-- or "drops" (if <is_add>=false) for verse <chapter>:<verse> in <changeset>.
|
|
-- A <changeset> is a table that maps cv's to {adds=<refset>, drops=<refset>}
|
|
-- tables, where cv's are chapter-verse pairs encoded as chapter*chapX+verse.
|
|
-- Each such entry denotes a verse where the set of songs that reference it
|
|
-- changes. The "adds" field lists the songs that refer to this verse but
|
|
-- not the previous one. The "drops" field lists the songs that refer to
|
|
-- this verse but not the next. This formulation allows us to efficiently
|
|
-- represent range-references (e.g., "Psalms 1:1-8") without creating a
|
|
-- separate table entry for each verse in the range.
|
|
-- Create a new entry for <chapter>:<verse> in <changeset> if it doesn't
|
|
-- already exist. Return the new <changeset>.
|
|
function insertref(is_add,set,ch,v,n,l,k)
|
|
if not set then set = {} end
|
|
local cv = ch*chapX+v
|
|
if not set[cv] then set[cv] = { adds={}, drops={} } end
|
|
set[cv][is_add and "adds" or "drops"][k] = { num=n, link=l }
|
|
return set
|
|
end
|
|
|
|
-- print_vrange(<file>,<book>,<ch1>,<v1>,<ch2>,<v2>,<lastchapter>)
|
|
-- Output LaTeX material to file <file> for verse range <ch1>:<v1>--<ch2>:<v2>
|
|
-- of book number <book>. Depending on <lastchapter>, the outputted material
|
|
-- might be the start of a new index entry or the continuation of a previous
|
|
-- entry. If <lastchapter> is positive, continue the previous entry and
|
|
-- print the chapter of <ch1>:<v1> only if it differs from <lastchapter>. If
|
|
-- <lastchapter> is negative, continue the previous entry and always print
|
|
-- the chapter number of <ch1>:<v1>.
|
|
function print_vrange(f,b,ch1,v1,ch2,v2,lch)
|
|
local r = f:write(lch == 0 and "\\idxentry{" or ",")
|
|
|
|
if v1 <= 0 then
|
|
if lch ~= 0 then r = r and f:write("\\thinspace ") end
|
|
r = r and f:write(ch1)
|
|
elseif 0 <= b and b < #bible and #bible[b] == 1 then
|
|
-- This book has only one chapter.
|
|
if lch ~= 0 then r = r and f:write("\\thinspace ") end
|
|
r = r and f:write(v1)
|
|
elseif lch <= 0 or lch ~= ch1 or ch1 ~= ch2 then
|
|
if lch ~= 0 then r = r and f:write(" ") end
|
|
r = r and f:write(ch1,":",v1)
|
|
else
|
|
if lch ~= 0 then r = r and f:write("\\thinspace ") end
|
|
r = r and f:write(v1)
|
|
end
|
|
|
|
if vlt(ch1,v1,ch2,v2) then
|
|
if v2 <= 0 then
|
|
r = r and f:write("--",ch2)
|
|
elseif ch1 ~= ch2 then
|
|
r = r and f:write("--",ch2,":",v2)
|
|
else
|
|
r = r and f:write("--",v2)
|
|
end
|
|
end
|
|
|
|
return r
|
|
end
|
|
|
|
-- print_reflist(<file>,<refset>)
|
|
-- Output the list of song references given by <refset> in sorted order
|
|
-- to file <file>.
|
|
function print_reflist(f,t)
|
|
local r = true
|
|
local s = {}
|
|
for k,_ in pairs(t) do table.insert(s,k) end
|
|
table.sort(s)
|
|
|
|
local first = true
|
|
for _,k in ipairs(s) do
|
|
if first then first = false else r = r and f:write("\\\\") end
|
|
r = r and f:write("\\songlink{",t[k]["link"],"}{",t[k]["num"],"}")
|
|
end
|
|
return r
|
|
end
|
|
|
|
function debug_print_reflist(r)
|
|
local first = true
|
|
io.stderr:write("{")
|
|
for k,_ in pairs(r) do
|
|
if first then first=false else io.stderr:write(",") end
|
|
io.stderr:write(k)
|
|
end
|
|
io.stderr:write("}")
|
|
end
|
|
|
|
function debug_print_changeset(x)
|
|
if not x then io.stderr:write("nil") else
|
|
for cv,r in pairs(x) do
|
|
local ch,v = unpack_cv(cv)
|
|
io.stderr:write("{",ch,":",v," --> adds=")
|
|
debug_print_reflist(r["adds"])
|
|
io.stderr:write(", drops=")
|
|
debug_print_reflist(r["drops"])
|
|
io.stderr:write("}")
|
|
end
|
|
end
|
|
end
|
|
|
|
-- genscriptureindex(<fstate>,<outname>,<biblename>)
|
|
-- Generate a LaTeX file named <outname> containing material suitable to
|
|
-- typeset the scripture index data found in input file <fstate>. Input
|
|
-- bible data from an ascii file named <biblename>. Return 0 on success,
|
|
-- 1 if there were warnings, and 2 if there was a fatal error.
|
|
function genscriptureindex(fs,outname,biblename)
|
|
local hadwarnings = 0
|
|
local idx = {}
|
|
|
|
io.stderr:write("songidx: Parsing scripture index data file ",fs["filename"],"...\n")
|
|
|
|
-- Read the bible data file into the bible array.
|
|
if not readbible(biblename) then return 2 end
|
|
|
|
-- Walk through the input file and construct a <changeset> for each book
|
|
-- of the bible. Each changeset represents the set of verses in that book
|
|
-- referred to by songs in the song book.
|
|
local key = 0
|
|
while true do
|
|
local ref = filereadln(fs)
|
|
if not ref then break end
|
|
local n = filereadln(fs)
|
|
if not n then
|
|
io.stderr:write("songidx:",fs["filename"],":",fs["lineno"],": incomplete song entry (orphan reference line)\n")
|
|
fileclose(fs)
|
|
return 2
|
|
end
|
|
local l = filereadln(fs)
|
|
if not l then
|
|
io.stderr:write("songidx:",fs["filename"],":",fs["lineno"],": incomplete song entry (missing hyperlink)\n")
|
|
fileclose(fs)
|
|
return 2
|
|
end
|
|
key = key + 1
|
|
|
|
local i = 1
|
|
local book, ch1, v1, ch2, v2 = -1, -1, -1, -1, -1
|
|
while i <= #ref do
|
|
i,book,ch1,v1 = parseref(ref,i,book,ch1)
|
|
ch2,v2 = ch1,v1
|
|
if not i then
|
|
io.stderr:write("songidx:",fs["filename"],":",fs["lineno"],": WARNING: Malformed scripture reference for song ",n,". Ignoring it.\n")
|
|
hadwarnings = 1
|
|
break
|
|
end
|
|
if book < 1 then
|
|
io.stderr:write("songidx:",fs["filename"],":",fs["lineno"],": WARNING: Scripture reference for song ",n," doesn't include a book name. Ignoring it.\n")
|
|
hadwarnings = 1
|
|
break
|
|
end
|
|
if book > #bible then
|
|
io.stderr:write("songidx:",fs["filename"],":",fs["lineno"],": WARNING: Scripture reference for song ",n," references unknown book. Ignoring it.\n")
|
|
hadwarnings = 1
|
|
break
|
|
end
|
|
if ch1 < 1 then ch1 = 1 end
|
|
if ch1 > #bible[book] then
|
|
io.stderr:write("songidx:",fs["filename"],":",fs["lineno"],": WARNING: Scripture reference for song ",n," refers to ",bible[book]["name"]," ",ch1,", which doesn't exist. Ignoring it.\n")
|
|
hadwarnings = 1
|
|
break
|
|
end
|
|
if v1 < 1 then v1 = 1 end
|
|
if v1 > bible[book][ch1] then
|
|
io.stderr:write("songidx:",fs["filename"],":",fs["lineno"],": WARNING: Scripture reference for song ",n," refers to ",bible[book]["name"]," ",ch1,":",v1,", which doesn't exist. Ignoring it.\n")
|
|
hadwarnings = 1
|
|
break
|
|
end
|
|
|
|
if ref:sub(i,i) == "-" then
|
|
-- If the reference ends in a "-", it starts an explicit range.
|
|
-- Parse the next reference to find the range's end.
|
|
i = unicode.utf8.find(ref, "[^%s%-]", i)
|
|
if not i then
|
|
io.stderr:write("songidx:",fs["filename"],":",fs["lineno"],": WARNING: Scripture reference for song ",n," has range with no limit. Ignoring it.\n")
|
|
hadwarnings = 1
|
|
break
|
|
end
|
|
local book2
|
|
i,book2,ch2,v2 = parseref(ref,i,book,ch1)
|
|
if not i then
|
|
io.stderr:write("songidx:",fs["filename"],":",fs["lineno"],": WARNING: Malformed scripture reference for song ",n,". Ignoring it.\n")
|
|
hadwarnings = 1
|
|
break
|
|
end
|
|
if book2 ~= book then
|
|
io.stderr:write("songidx:",fs["filename"],":",fs["lineno"],": WARNING: Scripture reference for song ",n," appears to span books! Ignoring it.\n")
|
|
hadwarnings = 1
|
|
break
|
|
end
|
|
end
|
|
if ch2 < 1 then ch2 = #bible[book] end
|
|
if ch2 > #bible[book] then
|
|
io.stderr:write("songidx:",fs["filename"],":",fs["lineno"],": WARNING: Scripture reference for song ",n," refers implicitly to ",bible[book]["name"]," ",ch2,", which doesn't exist. Ignoring it.\n")
|
|
hadwarnings = 1
|
|
break
|
|
end
|
|
if v2 < 1 then v2 = bible[book][ch2]
|
|
elseif v2 > bible[book][ch2] then
|
|
io.stderr:write("songidx:",fs["filename"],":",fs["lineno"],": WARNING: Scripture reference for song ",n," refers implicitly to chapter ",ch2," of ",bible[book]["name"],", which doesn't exist. Ignoring it.\n")
|
|
hadwarnings = 1
|
|
break
|
|
end
|
|
if vlt(ch2,v2,ch1,v1) then
|
|
io.stderr:write("songidx:",fs["filename"],":",fs["lineno"],": WARNING: Scripture reference for song ",n," contains backwards range ",bible[book]["name"]," ",ch1,":",v1,"-",ch2,":",v2,". Ignoring it.\n")
|
|
hadwarnings = 1
|
|
break
|
|
end
|
|
if i < #ref then i = i + 1 end
|
|
|
|
idx[book] = insertref(true, idx[book], ch1, v1, n, l, key)
|
|
idx[book] = insertref(false, idx[book], ch2, v2, n, l, key)
|
|
end
|
|
end
|
|
fileclose(fs)
|
|
|
|
-- Now create the index .sbx file.
|
|
io.stderr:write("songidx: Generating scripture index TeX file ",outname,"...\n")
|
|
local f,msg,errno
|
|
if outname == "-" then
|
|
f, outname = io.stdout, "stdout"
|
|
else
|
|
f,msg,errno = io.open(outname, "w")
|
|
if not f then
|
|
io.stderr:write("songidx: Unable to open ",outname," for writing.\n",
|
|
"Error ",errno,": ",msg,"\n")
|
|
return 2
|
|
end
|
|
end
|
|
|
|
-- For each book of the bible the has songs that reference it, go through its
|
|
-- <changeset> and generate a sequence of index entries. Wherever possible,
|
|
-- compact adjacent entries that have identical <refset>'s so that we never
|
|
-- have two consecutive index entries with identical right-hand sides.
|
|
for b=1,#bible do
|
|
local x = idx[b]
|
|
if x then
|
|
-- io.stderr:write("idx[",b,"] = ")
|
|
-- debug_print_changeset(x)
|
|
local s, t = {}, {}
|
|
for cv,_ in pairs(x) do table.insert(s,cv) end
|
|
table.sort(s)
|
|
local lch = 0 -- 0=none, -1=force printing of chapter
|
|
local cv1 = s[1]
|
|
local ch1,v1 = unpack_cv(cv1)
|
|
if not f:write("\\begin{idxblock}{",bible[b]["name"],"}\n") then return errorout(f) end
|
|
for i,cv in ipairs(s) do
|
|
local this = x[cv]
|
|
local ch,v = unpack_cv(cv)
|
|
local ncv,nxt,nch,nv = s[i+1]
|
|
if ncv then
|
|
nxt,nch,nv = x[ncv], unpack_cv(ncv)
|
|
end
|
|
for k,r in pairs(this["adds"]) do t[k] = r end
|
|
local skip = false
|
|
if ncv and eqdom(this["drops"], nxt["adds"]) then
|
|
-- Set of drops here equals set of adds next time. There's at least
|
|
-- a chance that we can combine this item and the next one into a
|
|
-- single index entry.
|
|
local ch2, v2 = vinc(b,ch,v)
|
|
if not vlt(ch2,v2, nch,nv) then
|
|
-- If the next item is adjacent to this one, do nothing. Just let
|
|
-- the range in progress be extended. We'll output a single entry
|
|
-- for all of these adjacent verses when we reach the end.
|
|
skip = true
|
|
elseif eqdom(t, this["drops"]) then
|
|
-- Otherwise, if the next item is not adjacent but all refs are
|
|
-- dropped here, then print a partial entry to be continued with a
|
|
-- comma next time.
|
|
if not print_vrange(f,b,ch1,v1,ch,v,lch) then return errorout(f) end
|
|
lch = (ch1 == ch) and ch or -1
|
|
ch1, v1 = nch, nv
|
|
skip = true
|
|
end
|
|
end
|
|
if not skip then
|
|
if next(this["drops"]) then
|
|
-- Some songs get dropped here, and either the next item is not
|
|
-- adjacent to this one, or it's adjacent and the set of adds is not
|
|
-- the same. In either case, that means the set of refs changes at
|
|
-- this point, so we need to output a full entry (or finish the one
|
|
-- in progress).
|
|
if not (print_vrange(f,b,ch1,v1,ch,v,lch) and
|
|
f:write("}{") and
|
|
print_reflist(f,t) and
|
|
f:write("}\n")) then return errorout(f) end
|
|
for k,_ in pairs(this["drops"]) do t[k] = nil end
|
|
lch = 0
|
|
if not next(t) and ncv then
|
|
ch1, v1 = nch, nv
|
|
else
|
|
ch1, v1 = vinc(b,ch,v)
|
|
end
|
|
end
|
|
if next(t) and ncv and next(nxt["adds"]) and vlt(ch1,v1,nch,nv) then
|
|
-- There are verses between this item and the next which have refs,
|
|
-- but the refs change at the beginning of the next item. Make an
|
|
-- entry for the intermediate block of verses.
|
|
local ch2, v2 = vdec(b,nch,nv)
|
|
if not (print_vrange(f,b,ch1,v1,ch2,v2,lch) and
|
|
f:write("}{") and
|
|
print_reflist(f,t) and
|
|
f:write("}\n")) then return errorout(f) end
|
|
lch, ch1, v1 = 0, nch, nv
|
|
end
|
|
end
|
|
end
|
|
if not f:write("\\end{idxblock}\n") then return errorout(f) end
|
|
end
|
|
end
|
|
|
|
closeout(f)
|
|
return hadwarnings
|
|
end
|
|
|
|
-- Main program entry point
|
|
function main()
|
|
local fs, biblename, inname, outname, locale
|
|
|
|
local i = 1
|
|
while arg[i] do
|
|
if arg[i] == "-v" or arg[i] == "--version" then
|
|
io.write("songidx ", VERSION, "\n",
|
|
"Copyright (C) 2017 Kevin W. Hamlen\n",
|
|
"License GPLv2: GNU GPL version 2 or later",
|
|
" <http://gnu.org/licenses/gpl.html>\n",
|
|
"This is free software: you are free to change and redistribute it.\n",
|
|
"There is NO WARRANTY, to the extent permitted by law.\n")
|
|
return 0
|
|
elseif arg[i] == "-h" or arg[i] == "--help" then
|
|
io.write("Syntax: ",arg[-1]," ",arg[0]," [options] input.sxd [output.sbx]\n",
|
|
"Available options:\n",
|
|
" -b FILE Set the bible format when generating a scripture index\n",
|
|
" --bible FILE (default: ", BIBLEDEFAULT, ")\n",
|
|
"\n",
|
|
" -l LOCALE Override the default system locale (affecting how non-\n",
|
|
" --locale LOCALE English characters are sorted). See your system help\n",
|
|
" for valid LOCALEs.\n",
|
|
"\n",
|
|
" -h Display this help file and stop.\n",
|
|
" --help\n",
|
|
"\n",
|
|
" -v Print version information and stop.\n",
|
|
" --version\n",
|
|
"\n",
|
|
"If omitted, [output.sbx] defaults to the input filename but with the file\n",
|
|
"extension renamed to '.sbx'. To read or write to stdin or stdout, use '-'\n",
|
|
"in place of input.sxd or output.sbx.\n",
|
|
"\n",
|
|
"See http://songs.sourceforge.net for support.\n")
|
|
return 0
|
|
elseif arg[i] == "-b" or arg[i] == "--bible" then
|
|
if biblename then
|
|
io.stderr:write("songidx: multiple bible files specified\n")
|
|
return 2
|
|
end
|
|
i = i + 1
|
|
if arg[i] then
|
|
biblename = arg[i]
|
|
else
|
|
io.stderr:write("songidx: ",arg[i-1]," option requires an argument\n")
|
|
return 2
|
|
end
|
|
elseif arg[i] == "-l" or arg[i] == "--locale" then
|
|
if locale then
|
|
io.stderr:write("songidx: multiple locales specified\n")
|
|
return 2
|
|
end
|
|
i = i + 1
|
|
if arg[i] then
|
|
locale = arg[i]
|
|
else
|
|
io.stderr:write("songidx: ",arg[i-1]," requires an argument\n")
|
|
return 2
|
|
end
|
|
elseif arg[i] == "-o" or arg[i] == "--output" then
|
|
if outname then
|
|
io.stderr:write("songidx: multiple output files specified\n")
|
|
return 2
|
|
end
|
|
i = i + 1
|
|
if arg[i] then
|
|
outname = arg[i]
|
|
else
|
|
io.stderr:write("songidx: ",arg[i-1]," option requires an argument\n")
|
|
return 2
|
|
end
|
|
elseif arg[i]:sub(1,1) == "-" and arg[i] ~= "-" then
|
|
io.stderr:write("songidx: unknown option ",arg[i],"\n")
|
|
return 2
|
|
elseif not inname then inname = arg[i]
|
|
elseif not outname then outname = arg[i]
|
|
else
|
|
io.stderr:write("songidx: too many command line arguments\n")
|
|
return 2
|
|
end
|
|
i = i + 1
|
|
end
|
|
|
|
if not locale then
|
|
os.setlocale("")
|
|
elseif not os.setlocale(locale) then
|
|
io.stderr:write("songidx: invalid locale: ",locale,"\n")
|
|
return 2
|
|
end
|
|
|
|
if not inname then
|
|
io.stderr:write("songidx: no input file specified\n")
|
|
return 2
|
|
end
|
|
if not outname then
|
|
if inname == "-" then
|
|
outname = "-"
|
|
else
|
|
local n
|
|
outname,n = inname:gsub("%.[^%./\\]*$", ".sbx")
|
|
if n == 0 then outname = inname .. ".sbx" end
|
|
end
|
|
end
|
|
if not biblename then biblename = BIBLEDEFAULT end
|
|
|
|
fs = fileopen(inname)
|
|
if not fs then return 2 end
|
|
|
|
local retval = 2
|
|
local buf = filereadln(fs)
|
|
if not buf then
|
|
io.stderr:write("songidx:",fs["filename"],": file is empty\n")
|
|
fileclose(fs)
|
|
elseif buf == "TITLE INDEX DATA FILE" then
|
|
retval = genindex(fs,outname,false)
|
|
elseif buf == "SCRIPTURE INDEX DATA FILE" then
|
|
retval = genscriptureindex(fs,outname,biblename)
|
|
elseif buf == "AUTHOR INDEX DATA FILE" then
|
|
retval = genindex(fs,outname,true)
|
|
else
|
|
io.stderr:write("songidx:",fs["filename"],":",fs["lineno"],": file has unrecognized format\n")
|
|
fileclose(fs)
|
|
end
|
|
|
|
if retval == 0 then
|
|
io.stderr:write("songidx: Done!\n")
|
|
elseif retval == 1 then
|
|
io.stderr:write("songidx: COMPLETED WITH ERRORS. SEE ABOVE.\n")
|
|
else
|
|
io.stderr:write("songidx: FAILED. SEE ERROR MESSAGES ABOVE.\n")
|
|
end
|
|
|
|
return retval
|
|
end
|
|
|
|
os.exit(main())
|
|
|