Wed, 15 Dec 2021 10:37:47 +0200
Add ANTI-ABUSE LICENSE
#!/usr/bin/env lua ---------------------------------------------------------------------- -- Utility functions ---------------------------------------------------------------------- local unpack = table.unpack or unpack -- Returns the result of mapping the values in table t through the function f local function map(t, f) local out = {} for k,v in pairs(t) do out[k] = f(v,k) end return out end -- Functional style if statement. (NOTE: no short circuit evaluation) local function iff(t, a, b) if t then return a else return b end end -- Splits the text into an array of separate lines. local function split(text, sep) sep = sep or "\n" local lines = {} local pos = 1 while true do local b,e = text:find(sep, pos) if not b then table.insert(lines, text:sub(pos)) break end table.insert(lines, text:sub(pos, b-1)) pos = e + 1 end return lines end -- Converts tabs to spaces local function detab(text) local tab_width = 4 local function rep(match) local spaces = -match:len() while spaces<1 do spaces = spaces + tab_width end return match .. string.rep(" ", spaces) end text = text:gsub("([^\n]-)\t", rep) return text end -- Applies string.find for every pattern in the list and returns the first match local function find_first(s, patterns, index) local res = {} for _,p in ipairs(patterns) do local match = {s:find(p, index)} if #match>0 and (#res==0 or match[1] < res[1]) then res = match end end return unpack(res) end -- If a replacement array is specified, the range [start, stop] in the array is replaced -- with the replacement array and the resulting array is returned. Without a replacement -- array the section of the array between start and stop is returned. local function splice(array, start, stop, replacement) if replacement then local n = stop - start + 1 while n > 0 do table.remove(array, start) n = n - 1 end for _,v in ipairs(replacement) do table.insert(array, start, v) end return array else local res = {} for i = start,stop do table.insert(res, array[i]) end return res end end -- Outdents the text one step. local function outdent(text) text = "\n" .. text text = text:gsub("\n ? ? ?", "\n") text = text:sub(2) return text end -- Indents the text one step. local function indent(text) text = text:gsub("\n", "\n ") return text end -- Does a simple tokenization of html data. Returns the data as a list of tokens. -- Each token is a table with a type field (which is either "tag" or "text") and -- a text field (which contains the original token data). local function tokenize_html(html) local tokens = {} local pos = 1 while true do local start = find_first(html, {"<!%-%-", "<[a-z/!$]", "<%?"}, pos) if not start then table.insert(tokens, {type="text", text=html:sub(pos)}) break end if start ~= pos then table.insert(tokens, {type="text", text = html:sub(pos, start-1)}) end local _, stop if html:match("^<!%-%-", start) then _,stop = html:find("%-%->", start) elseif html:match("^<%?", start) then _,stop = html:find("?>", start) else _,stop = html:find("%b<>", start) end if not stop then -- error("Could not match html tag " .. html:sub(start,start+30)) table.insert(tokens, {type="text", text=html:sub(start, start)}) pos = start + 1 else table.insert(tokens, {type="tag", text=html:sub(start, stop)}) pos = stop + 1 end end return tokens end ---------------------------------------------------------------------- -- Hash ---------------------------------------------------------------------- -- This is used to "hash" data into alphanumeric strings that are unique -- in the document. (Note that this is not cryptographic hash, the hash -- function is not one-way.) The hash procedure is used to protect parts -- of the document from further processing. local HASH = { -- Has the hash been inited. inited = false, -- The unique string prepended to all hash values. This is to ensure -- that hash values do not accidently coincide with an actual existing -- string in the document. identifier = "", -- Counter that counts up for each new hash instance. counter = 0, -- Hash table. table = {} } -- Inits hashing. Creates a hash_identifier that doesn't occur anywhere -- in the text. local function init_hash(text) HASH.inited = true HASH.identifier = "" HASH.counter = 0 HASH.table = {} local s = "HASH" local counter = 0 local id while true do id = s .. counter if not text:find(id, 1, true) then break end counter = counter + 1 end HASH.identifier = id end -- Returns the hashed value for s. local function hash(s) assert(HASH.inited) if not HASH.table[s] then HASH.counter = HASH.counter + 1 local id = HASH.identifier .. HASH.counter .. "X" HASH.table[s] = id end return HASH.table[s] end ---------------------------------------------------------------------- -- Protection ---------------------------------------------------------------------- -- The protection module is used to "protect" parts of a document -- so that they are not modified by subsequent processing steps. -- Protected parts are saved in a table for later unprotection -- Protection data local PD = { -- Saved blocks that have been converted blocks = {}, -- Block level tags that will be protected tags = {"p", "div", "h1", "h2", "h3", "h4", "h5", "h6", "blockquote", "pre", "table", "dl", "ol", "ul", "script", "noscript", "form", "fieldset", "iframe", "math", "ins", "del"} } -- Pattern for matching a block tag that begins and ends in the leftmost -- column and may contain indented subtags, i.e. -- <div> -- A nested block. -- <div> -- Nested data. -- </div> -- </div> local function block_pattern(tag) return "\n<" .. tag .. ".-\n</" .. tag .. ">[ \t]*\n" end -- Pattern for matching a block tag that begins and ends with a newline local function line_pattern(tag) return "\n<" .. tag .. ".-</" .. tag .. ">[ \t]*\n" end -- Protects the range of characters from start to stop in the text and -- returns the protected string. local function protect_range(text, start, stop) local s = text:sub(start, stop) local h = hash(s) PD.blocks[h] = s text = text:sub(1,start) .. h .. text:sub(stop) return text end -- Protect every part of the text that matches any of the patterns. The first -- matching pattern is protected first, etc. local function protect_matches(text, patterns) while true do local start, stop = find_first(text, patterns) if not start then break end text = protect_range(text, start, stop) end return text end -- Protects blocklevel tags in the specified text local function protect(text) -- First protect potentially nested block tags text = protect_matches(text, map(PD.tags, block_pattern)) -- Then protect block tags at the line level. text = protect_matches(text, map(PD.tags, line_pattern)) -- Protect <hr> and comment tags text = protect_matches(text, {"\n<hr[^>]->[ \t]*\n"}) text = protect_matches(text, {"\n<!%-%-.-%-%->[ \t]*\n"}) return text end -- Returns true if the string s is a hash resulting from protection local function is_protected(s) return PD.blocks[s] end -- Unprotects the specified text by expanding all the nonces local function unprotect(text) for k,v in pairs(PD.blocks) do v = v:gsub("%%", "%%%%") text = text:gsub(k, v) end return text end ---------------------------------------------------------------------- -- Block transform ---------------------------------------------------------------------- -- The block transform functions transform the text on the block level. -- They work with the text as an array of lines rather than as individual -- characters. -- Returns true if the line is a ruler of (char) characters. -- The line must contain at least three char characters and contain only spaces and -- char characters. local function is_ruler_of(line, char) if not line:match("^[ %" .. char .. "]*$") then return false end if not line:match("%" .. char .. ".*%" .. char .. ".*%" .. char) then return false end return true end -- Identifies the block level formatting present in the line local function classify(line) local info = {line = line, text = line} if line:match("^ ") then info.type = "indented" info.outdented = line:sub(5) return info end for _,c in ipairs({'*', '-', '_', '='}) do if is_ruler_of(line, c) then info.type = "ruler" info.ruler_char = c return info end end if line == "" then info.type = "blank" return info end if line:match("^(#+)[ \t]*(.-)[ \t]*#*[ \t]*$") then local m1, m2 = line:match("^(#+)[ \t]*(.-)[ \t]*#*[ \t]*$") info.type = "header" info.level = m1:len() info.text = m2 return info end if line:match("^ ? ? ?(%d+)%.[ \t]+(.+)") then local number, text = line:match("^ ? ? ?(%d+)%.[ \t]+(.+)") info.type = "list_item" info.list_type = "numeric" info.number = 0 + number info.text = text return info end if line:match("^ ? ? ?([%*%+%-])[ \t]+(.+)") then local bullet, text = line:match("^ ? ? ?([%*%+%-])[ \t]+(.+)") info.type = "list_item" info.list_type = "bullet" info.bullet = bullet info.text= text return info end if line:match("^>[ \t]?(.*)") then info.type = "blockquote" info.text = line:match("^>[ \t]?(.*)") return info end if is_protected(line) then info.type = "raw" info.html = unprotect(line) return info end info.type = "normal" return info end -- Find headers constisting of a normal line followed by a ruler and converts them to -- header entries. local function headers(array) local i = 1 while i <= #array - 1 do if array[i].type == "normal" and array[i+1].type == "ruler" and (array[i+1].ruler_char == "-" or array[i+1].ruler_char == "=") then local info = {line = array[i].line} info.text = info.line info.type = "header" info.level = iff(array[i+1].ruler_char == "=", 1, 2) table.remove(array, i+1) array[i] = info end i = i + 1 end return array end -- Forward declarations local block_transform, span_transform, encode_code -- Convert lines to html code local function blocks_to_html(lines, no_paragraphs) local out = {} local i = 1 while i <= #lines do local line = lines[i] if line.type == "ruler" then table.insert(out, "<hr/>") elseif line.type == "raw" then table.insert(out, line.html) elseif line.type == "normal" then local s = line.line while i+1 <= #lines and lines[i+1].type == "normal" do i = i + 1 s = s .. "\n" .. lines[i].line end if no_paragraphs then table.insert(out, span_transform(s)) else table.insert(out, "<p>" .. span_transform(s) .. "</p>") end elseif line.type == "header" then local s = "<h" .. line.level .. ">" .. span_transform(line.text) .. "</h" .. line.level .. ">" table.insert(out, s) else table.insert(out, line.line) end i = i + 1 end return out end -- Find list blocks and convert them to protected data blocks local function lists(array, sublist) local function process_list(arr) local function any_blanks(arr) for i = 1, #arr do if arr[i].type == "blank" then return true end end return false end local function split_list_items(arr) local acc = {arr[1]} local res = {} for i=2,#arr do if arr[i].type == "list_item" then table.insert(res, acc) acc = {arr[i]} else table.insert(acc, arr[i]) end end table.insert(res, acc) return res end local function process_list_item(lines, block) while lines[#lines].type == "blank" do table.remove(lines) end local itemtext = lines[1].text for i=2,#lines do itemtext = itemtext .. "\n" .. outdent(lines[i].line) end if block then itemtext = block_transform(itemtext, true) if not itemtext:find("<pre>") then itemtext = indent(itemtext) end return " <li>" .. itemtext .. "</li>" else local lines = split(itemtext) lines = map(lines, classify) lines = lists(lines, true) lines = blocks_to_html(lines, true) itemtext = table.concat(lines, "\n") if not itemtext:find("<pre>") then itemtext = indent(itemtext) end return " <li>" .. itemtext .. "</li>" end end local block_list = any_blanks(arr) local items = split_list_items(arr) local out = "" for _, item in ipairs(items) do out = out .. process_list_item(item, block_list) .. "\n" end if arr[1].list_type == "numeric" then return "<ol>\n" .. out .. "</ol>" else return "<ul>\n" .. out .. "</ul>" end end -- Finds the range of lines composing the first list in the array. A list -- starts with (^ list_item) or (blank list_item) and ends with -- (blank* $) or (blank normal). -- -- A sublist can start with just (list_item) does not need a blank... local function find_list(array, sublist) local function find_list_start(array, sublist) if array[1].type == "list_item" then return 1 end if sublist then for i = 1,#array do if array[i].type == "list_item" then return i end end else for i = 1, #array-1 do if array[i].type == "blank" and array[i+1].type == "list_item" then return i+1 end end end return nil end local function find_list_end(array, start) local pos = #array for i = start, #array-1 do if array[i].type == "blank" and array[i+1].type ~= "list_item" and array[i+1].type ~= "indented" and array[i+1].type ~= "blank" then pos = i-1 break end end while pos > start and array[pos].type == "blank" do pos = pos - 1 end return pos end local start = find_list_start(array, sublist) if not start then return nil end return start, find_list_end(array, start) end while true do local start, stop = find_list(array, sublist) if not start then break end local text = process_list(splice(array, start, stop)) local info = { line = text, type = "raw", html = text } array = splice(array, start, stop, {info}) end -- Convert any remaining list items to normal for _,line in ipairs(array) do if line.type == "list_item" then line.type = "normal" end end return array end -- Find and convert blockquote markers. local function blockquotes(lines) local function find_blockquote(lines) local start for i,line in ipairs(lines) do if line.type == "blockquote" then start = i break end end if not start then return nil end local stop = #lines for i = start+1, #lines do if lines[i].type == "blank" or lines[i].type == "blockquote" then elseif lines[i].type == "normal" then if lines[i-1].type == "blank" then stop = i-1 break end else stop = i-1 break end end while lines[stop].type == "blank" do stop = stop - 1 end return start, stop end local function process_blockquote(lines) local raw = lines[1].text for i = 2,#lines do raw = raw .. "\n" .. lines[i].text end local bt = block_transform(raw) if not bt:find("<pre>") then bt = indent(bt) end return "<blockquote>\n " .. bt .. "\n</blockquote>" end while true do local start, stop = find_blockquote(lines) if not start then break end local text = process_blockquote(splice(lines, start, stop)) local info = { line = text, type = "raw", html = text } lines = splice(lines, start, stop, {info}) end return lines end -- Find and convert codeblocks. local function codeblocks(lines) local function find_codeblock(lines) local start for i,line in ipairs(lines) do if line.type == "indented" then start = i break end end if not start then return nil end local stop = #lines for i = start+1, #lines do if lines[i].type ~= "indented" and lines[i].type ~= "blank" then stop = i-1 break end end while lines[stop].type == "blank" do stop = stop - 1 end return start, stop end local function process_codeblock(lines) local raw = detab(encode_code(outdent(lines[1].line))) for i = 2,#lines do raw = raw .. "\n" .. detab(encode_code(outdent(lines[i].line))) end return "<pre><code>" .. raw .. "\n</code></pre>" end while true do local start, stop = find_codeblock(lines) if not start then break end local text = process_codeblock(splice(lines, start, stop)) local info = { line = text, type = "raw", html = text } lines = splice(lines, start, stop, {info}) end return lines end -- Perform all the block level transforms function block_transform(text, sublist) local lines = split(text) lines = map(lines, classify) lines = headers(lines) lines = lists(lines, sublist) lines = codeblocks(lines) lines = blockquotes(lines) lines = blocks_to_html(lines) local text = table.concat(lines, "\n") return text end ---------------------------------------------------------------------- -- Span transform ---------------------------------------------------------------------- -- Functions for transforming the text at the span level. -- These characters may need to be escaped because they have a special -- meaning in markdown. local escape_chars = "'\\`*_{}[]()>#+-.!'" local escape_table = {} local function init_escape_table() escape_table = {} for i = 1,#escape_chars do local c = escape_chars:sub(i,i) escape_table[c] = hash(c) end end -- Adds a new escape to the escape table. local function add_escape(text) if not escape_table[text] then escape_table[text] = hash(text) end return escape_table[text] end -- Encode backspace-escaped characters in the markdown source. local function encode_backslash_escapes(t) for i=1,escape_chars:len() do local c = escape_chars:sub(i,i) t = t:gsub("\\%" .. c, escape_table[c]) end return t end -- Escape characters that should not be disturbed by markdown. local function escape_special_chars(text) local tokens = tokenize_html(text) local out = "" for _, token in ipairs(tokens) do local t = token.text if token.type == "tag" then -- In tags, encode * and _ so they don't conflict with their use in markdown. t = t:gsub("%*", escape_table["*"]) t = t:gsub("%_", escape_table["_"]) else t = encode_backslash_escapes(t) end out = out .. t end return out end -- Unescape characters that have been encoded. local function unescape_special_chars(t) local tin = t for k,v in pairs(escape_table) do k = k:gsub("%%", "%%%%") t = t:gsub(v,k) end if t ~= tin then t = unescape_special_chars(t) end return t end -- Encode/escape certain characters inside Markdown code runs. -- The point is that in code, these characters are literals, -- and lose their special Markdown meanings. function encode_code(s) s = s:gsub("%&", "&") s = s:gsub("<", "<") s = s:gsub(">", ">") for k,v in pairs(escape_table) do s = s:gsub("%"..k, v) end return s end -- Handle backtick blocks. local function code_spans(s) s = s:gsub("\\\\", escape_table["\\"]) s = s:gsub("\\`", escape_table["`"]) local pos = 1 while true do local start, stop = s:find("`+", pos) if not start then return s end local count = stop - start + 1 -- Find a matching numbert of backticks local estart, estop = s:find(string.rep("`", count), stop+1) local brstart = s:find("\n", stop+1) if estart and (not brstart or estart < brstart) then local code = s:sub(stop+1, estart-1) code = code:gsub("^[ \t]+", "") code = code:gsub("[ \t]+$", "") code = code:gsub(escape_table["\\"], escape_table["\\"] .. escape_table["\\"]) code = code:gsub(escape_table["`"], escape_table["\\"] .. escape_table["`"]) code = "<code>" .. encode_code(code) .. "</code>" code = add_escape(code) s = s:sub(1, start-1) .. code .. s:sub(estop+1) pos = start + code:len() else pos = stop + 1 end end return s end -- Encode alt text... enodes &, and ". local function encode_alt(s) if not s then return s end s = s:gsub('&', '&') s = s:gsub('"', '"') s = s:gsub('<', '<') return s end -- Forward declaration for link_db as returned by strip_link_definitions. local link_database -- Handle image references local function images(text) local function reference_link(alt, id) alt = encode_alt(alt:match("%b[]"):sub(2,-2)) id = id:match("%[(.*)%]"):lower() if id == "" then id = text:lower() end link_database[id] = link_database[id] or {} if not link_database[id].url then return nil end local url = link_database[id].url or id url = encode_alt(url) local title = encode_alt(link_database[id].title) if title then title = " title=\"" .. title .. "\"" else title = "" end return add_escape ('<img src="' .. url .. '" alt="' .. alt .. '"' .. title .. "/>") end local function inline_link(alt, link) alt = encode_alt(alt:match("%b[]"):sub(2,-2)) local url, title = link:match("%(<?(.-)>?[ \t]*['\"](.+)['\"]") url = url or link:match("%(<?(.-)>?%)") url = encode_alt(url) title = encode_alt(title) if title then return add_escape('<img src="' .. url .. '" alt="' .. alt .. '" title="' .. title .. '"/>') else return add_escape('<img src="' .. url .. '" alt="' .. alt .. '"/>') end end text = text:gsub("!(%b[])[ \t]*\n?[ \t]*(%b[])", reference_link) text = text:gsub("!(%b[])(%b())", inline_link) return text end -- Handle anchor references local function anchors(text) local function reference_link(text, id) text = text:match("%b[]"):sub(2,-2) id = id:match("%b[]"):sub(2,-2):lower() if id == "" then id = text:lower() end link_database[id] = link_database[id] or {} if not link_database[id].url then return nil end local url = link_database[id].url or id url = encode_alt(url) local title = encode_alt(link_database[id].title) if title then title = " title=\"" .. title .. "\"" else title = "" end return add_escape("<a href=\"" .. url .. "\"" .. title .. ">") .. text .. add_escape("</a>") end local function inline_link(text, link) text = text:match("%b[]"):sub(2,-2) local url, title = link:match("%(<?(.-)>?[ \t]*['\"](.+)['\"]") title = encode_alt(title) url = url or link:match("%(<?(.-)>?%)") or "" url = encode_alt(url) if title then return add_escape("<a href=\"" .. url .. "\" title=\"" .. title .. "\">") .. text .. "</a>" else return add_escape("<a href=\"" .. url .. "\">") .. text .. add_escape("</a>") end end text = text:gsub("(%b[])[ \t]*\n?[ \t]*(%b[])", reference_link) text = text:gsub("(%b[])(%b())", inline_link) return text end -- Handle auto links, i.e. <http://www.google.com/>. local function auto_links(text) local function link(s) return add_escape("<a href=\"" .. s .. "\">") .. s .. "</a>" end -- Encode chars as a mix of dec and hex entitites to (perhaps) fool -- spambots. local function encode_email_address(s) -- Use a deterministic encoding to make unit testing possible. -- Code 45% hex, 45% dec, 10% plain. local hex = {code = function(c) return "&#x" .. string.format("%x", c:byte()) .. ";" end, count = 1, rate = 0.45} local dec = {code = function(c) return "&#" .. c:byte() .. ";" end, count = 0, rate = 0.45} local plain = {code = function(c) return c end, count = 0, rate = 0.1} local codes = {hex, dec, plain} local function swap(t,k1,k2) local temp = t[k2] t[k2] = t[k1] t[k1] = temp end local out = "" for i = 1,s:len() do for _,code in ipairs(codes) do code.count = code.count + code.rate end if codes[1].count < codes[2].count then swap(codes,1,2) end if codes[2].count < codes[3].count then swap(codes,2,3) end if codes[1].count < codes[2].count then swap(codes,1,2) end local code = codes[1] local c = s:sub(i,i) -- Force encoding of "@" to make email address more invisible. if c == "@" and code == plain then code = codes[2] end out = out .. code.code(c) code.count = code.count - 1 end return out end local function mail(s) s = unescape_special_chars(s) local address = encode_email_address("mailto:" .. s) local text = encode_email_address(s) return add_escape("<a href=\"" .. address .. "\">") .. text .. "</a>" end -- links text = text:gsub("<(https?:[^'\">%s]+)>", link) text = text:gsub("<(ftp:[^'\">%s]+)>", link) -- mail text = text:gsub("<mailto:([^'\">%s]+)>", mail) text = text:gsub("<([-.%w]+%@[-.%w]+)>", mail) return text end -- Encode free standing amps (&) and angles (<)... note that this does not -- encode free >. local function amps_and_angles(s) -- encode amps not part of &..; expression local pos = 1 while true do local amp = s:find("&", pos) if not amp then break end local semi = s:find(";", amp+1) local stop = s:find("[ \t\n&]", amp+1) if not semi or (stop and stop < semi) or (semi - amp) > 15 then s = s:sub(1,amp-1) .. "&" .. s:sub(amp+1) pos = amp+1 else pos = amp+1 end end -- encode naked <'s s = s:gsub("<([^a-zA-Z/?$!])", "<%1") s = s:gsub("<$", "<") -- what about >, nothing done in the original markdown source to handle them return s end -- Handles emphasis markers (* and _) in the text. local function emphasis(text) for _, s in ipairs {"%*%*", "%_%_"} do text = text:gsub(s .. "([^%s][%*%_]?)" .. s, "<strong>%1</strong>") text = text:gsub(s .. "([^%s][^<>]-[^%s][%*%_]?)" .. s, "<strong>%1</strong>") end for _, s in ipairs {"%*", "%_"} do text = text:gsub(s .. "([^%s_])" .. s, "<em>%1</em>") text = text:gsub(s .. "(<strong>[^%s_]</strong>)" .. s, "<em>%1</em>") text = text:gsub(s .. "([^%s_][^<>_]-[^%s_])" .. s, "<em>%1</em>") text = text:gsub(s .. "([^<>_]-<strong>[^<>_]-</strong>[^<>_]-)" .. s, "<em>%1</em>") end return text end -- Handles line break markers in the text. local function line_breaks(text) return text:gsub(" +\n", " <br/>\n") end -- Perform all span level transforms. function span_transform(text) text = code_spans(text) text = escape_special_chars(text) text = images(text) text = anchors(text) text = auto_links(text) text = amps_and_angles(text) text = emphasis(text) text = line_breaks(text) return text end ---------------------------------------------------------------------- -- Markdown ---------------------------------------------------------------------- -- Cleanup the text by normalizing some possible variations to make further -- processing easier. local function cleanup(text) -- Standardize line endings text = text:gsub("\r\n", "\n") -- DOS to UNIX text = text:gsub("\r", "\n") -- Mac to UNIX -- Convert all tabs to spaces text = detab(text) -- Strip lines with only spaces and tabs while true do local subs text, subs = text:gsub("\n[ \t]+\n", "\n\n") if subs == 0 then break end end return "\n" .. text .. "\n" end -- Strips link definitions from the text and stores the data in a lookup table. local function strip_link_definitions(text) local linkdb = {} local function link_def(id, url, title) id = id:match("%[(.+)%]"):lower() linkdb[id] = linkdb[id] or {} linkdb[id].url = url or linkdb[id].url linkdb[id].title = title or linkdb[id].title return "" end local def_no_title = "\n ? ? ?(%b[]):[ \t]*\n?[ \t]*<?([^%s>]+)>?[ \t]*" local def_title1 = def_no_title .. "[ \t]+\n?[ \t]*[\"'(]([^\n]+)[\"')][ \t]*" local def_title2 = def_no_title .. "[ \t]*\n[ \t]*[\"'(]([^\n]+)[\"')][ \t]*" local def_title3 = def_no_title .. "[ \t]*\n?[ \t]+[\"'(]([^\n]+)[\"')][ \t]*" text = text:gsub(def_title1, link_def) text = text:gsub(def_title2, link_def) text = text:gsub(def_title3, link_def) text = text:gsub(def_no_title, link_def) return text, linkdb end -- Main markdown processing function local function markdown(text) init_hash(text) init_escape_table() text = cleanup(text) text = protect(text) text, link_database = strip_link_definitions(text) text = block_transform(text) text = unescape_special_chars(text) return text end ---------------------------------------------------------------------- -- End of module ---------------------------------------------------------------------- -- For compatibility, set markdown function as a global _G.markdown = markdown -- Class for parsing command-line options local OptionParser = {} OptionParser.__index = OptionParser -- Creates a new option parser function OptionParser:new() local o = {short = {}, long = {}} setmetatable(o, self) return o end -- Calls f() whenever a flag with specified short and long name is encountered function OptionParser:flag(short, long, f) local info = {type = "flag", f = f} if short then self.short[short] = info end if long then self.long[long] = info end end -- Calls f(param) whenever a parameter flag with specified short and long name is encountered function OptionParser:param(short, long, f) local info = {type = "param", f = f} if short then self.short[short] = info end if long then self.long[long] = info end end -- Calls f(v) for each non-flag argument function OptionParser:arg(f) self.arg = f end -- Runs the option parser for the specified set of arguments. Returns true if all arguments -- where successfully parsed and false otherwise. function OptionParser:run(args) local pos = 1 while pos <= #args do local arg = args[pos] if arg == "--" then for i=pos+1,#args do if self.arg then self.arg(args[i]) end return true end end if arg:match("^%-%-") then local info = self.long[arg:sub(3)] if not info then print("Unknown flag: " .. arg) return false end if info.type == "flag" then info.f() pos = pos + 1 else local param = args[pos+1] if not param then print("No parameter for flag: " .. arg) return false end info.f(param) pos = pos+2 end elseif arg:match("^%-") then for i=2,arg:len() do local c = arg:sub(i,i) local info = self.short[c] if not info then print("Unknown flag: -" .. c) return false end if info.type == "flag" then info.f() else if i == arg:len() then local param = args[pos+1] if not param then print("No parameter for flag: -" .. c) return false end info.f(param) pos = pos + 1 else local param = arg:sub(i+1) info.f(param) end break end end pos = pos + 1 else if self.arg then self.arg(arg) end pos = pos + 1 end end return true end local function read_file(path, descr) local file = io.open(path) or error("Could not open " .. descr .. " file: " .. path) local contents = file:read("*a") or error("Could not read " .. descr .. " from " .. path) file:close() return contents end -- Handles the case when markdown is run from the command line local function run_command_line(arg) -- Generate output for input s given options local function run(s, options) s = markdown(s) if not options.wrap_header then return s end local header if options.header then header = read_file(options.header, "header") else header = [[ <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd"> <html> <head> <meta http-equiv="content-type" content="text/html; charset=CHARSET" /> <title>TITLE</title> <link rel="stylesheet" type="text/css" href="STYLESHEET" /> </head> <body> ]] local title = options.title or s:match("<h1>(.-)</h1>") or s:match("<h2>(.-)</h2>") or s:match("<h3>(.-)</h3>") or "Untitled" header = header:gsub("TITLE", title) if options.inline_style then local style = read_file(options.stylesheet, "style sheet") header = header:gsub('<link rel="stylesheet" type="text/css" href="STYLESHEET" />', "<style type=\"text/css\"><!--\n" .. style .. "\n--></style>") else header = header:gsub("STYLESHEET", options.stylesheet) end header = header:gsub("CHARSET", options.charset) end local footer = "</body></html>" if options.footer then footer = read_file(options.footer, "footer") end return header .. s .. footer end -- Generate output path name from input path name given options. local function outpath(path, options) if options.append then return path .. ".html" end local m = path:match("^(.+%.html)[^/\\]+$") if m then return m end m = path:match("^(.+%.)[^/\\]*$") if m and path ~= m .. "html" then return m .. "html" end return path .. ".html" end -- Default commandline options local options = { wrap_header = true, header = nil, footer = nil, charset = "utf-8", title = nil, stylesheet = "default.css", inline_style = false } local help = [[ Usage: markdown.lua [OPTION] [FILE] Runs the markdown text markup to HTML converter on each file specified on the command line. If no files are specified, runs on standard input. No header: -n, --no-wrap Don't wrap the output in <html>... tags. Custom header: -e, --header FILE Use content of FILE for header. -f, --footer FILE Use content of FILE for footer. Generated header: -c, --charset SET Specifies charset (default utf-8). -i, --title TITLE Specifies title (default from first <h1> tag). -s, --style STYLE Specifies style sheet file (default default.css). -l, --inline-style Include the style sheet file inline in the header. Generated files: -a, --append Append .html extension (instead of replacing). Other options: -h, --help Print this help text. -t, --test Run the unit tests. ]] local run_stdin = true local op = OptionParser:new() op:flag("n", "no-wrap", function () options.wrap_header = false end) op:param("e", "header", function (x) options.header = x end) op:param("f", "footer", function (x) options.footer = x end) op:param("c", "charset", function (x) options.charset = x end) op:param("i", "title", function(x) options.title = x end) op:param("s", "style", function(x) options.stylesheet = x end) op:flag("l", "inline-style", function() options.inline_style = true end) op:flag("a", "append", function() options.append = true end) op:flag("t", "test", function() local n = arg[0]:gsub("markdown%.lua", "markdown-tests.lua") local f = io.open(n) if f then f:close() package.loaded.markdown = markdown dofile(n) else error("Cannot find markdown-tests.lua") end run_stdin = false end) op:flag("h", "help", function() print(help) run_stdin = false end) op:arg(function(path) local s = read_file(path, "input") s = run(s, options) local file = io.open(outpath(path, options), "w") or error("Could not open output file: " .. outpath(path, options)) file:write(s) file:close() run_stdin = false end ) if not op:run(arg) then print(help) run_stdin = false end if run_stdin then local s = io.read("*a") s = run(s, options) io.write(s) end end -- If we are being run from the command-line, act accordingly if arg and arg[0]:find("markdown%.lua$") then run_command_line(arg) else return markdown end