Module:Wikitext Parsing

require("strict")
local p = {}

--Helper functions
local function startswith(text, subtext)
	return string.sub(text, 1, #subtext) == subtext
end
local function endswith(text, subtext)
	return string.sub(text, -#subtext, -1) == subtext
end
local function allcases(s)
	return s:gsub("%a", function(c) 
		return "["..c:upper()..c:lower().."]"
	end)
end

--[[ Implementation notes
---- NORMAL HTML TAGS ----
Tags are very strict on how they want to start, but loose on how they end.
The start must strictly follow <[tAgNaMe](%s|>) with no room for whitespace in
the tag's name, but may then flow as they want afterwards, making
<div\nclass\n=\n"\nerror\n"\n> valid

There's no sense of escaping < or >
E.g.
 <div class="error\>"> will end at \> despite it being inside a quote
 <div class="<span class="error">error</span>"> will not process the larger div

If a tag has no end, it will consume all text instead of not processing

---- NOPROCESSING TAGS (nowiki, pre, syntaxhighlight, source) ----
(In most comments, <source> will not be mentioned. This is because it is the
deprecated version of <syntaxhighlight>)

No-Processing tags have some interesting differences to the above rules.
For example, their syntax is a lot stricter. While an opening tag appears to
follow the same set of rules, A closing tag can't have any sort of extra
formatting period. While </div a/a> is valid, </nowiki a/a> isn't - only
newlines and spaces are allowed in closing tags (except in <pre> tags, which
follow the rules of a regular html tag for formatting).

Both the content inside the tag pair and the content inside each side of the
pair is not processed. E.g. <nowiki |}}>|}}</nowiki> would have both of the |}}
escaped in practice.

When something in the code is referenced to as a "Nowiki Tag", it means a tag
which causes wiki text to not be processed, which includes <nowiki>, <pre>,
and <syntaxhighlight>

Since we only care about these tags, we can ignore the idea of an intercepting
tag preventing processing, and just go straight for the first ending we can find
If there is no ending to find, the tag will NOT consume the rest of the text in
terms of processing behaviour (though <pre> will appear to have an effect).
Even if there is no end of the tag, the content inside the opening half will
still be unprocessed, meaning {{X20|<nowiki }}>}} wouldn't end at the first }}
despite there being no ending to the tag.

Note that there are some tags, like <math>, which also function like <nowiki>
which are included in this aswell. Some other tags, like <ref>, have far too
unpredictable behaviour to be handled currently (they'd have to be split and
processed as something seperate - its complicated, but maybe not impossible.)

---- HTML COMMENTS AND INCLUDEONLY ----
HTML Comments are about as basic as it could get for this
Start at <!--, end at -->, no extra conditions. Simple enough
If a comment has no end, it will eat all text instead of not being processed

includeonly tags function mostly like a regular nowiki tag, with the exception
that the tag will actually consume all future text if not given an ending as
opposed to simply giving up and not changing anything. Due to complications and
the fact that this is far less likely to be present on a page, aswell as being
something that may not want to be escaped, includeonly tags are ignored during
processing
--]]
local validtags = {nowiki=1, pre=1, syntaxhighlight=1, source=1, math=1}
--This function expects the string to start with the tag
local function TestForNowikiTag(text)
	local tagName = (string.match(text, "^<([^\n />]+)") or ""):lower()
	if not validtags[tagName] then
		return nil
	end
	local nextOpener = string.find(text, "<", 2) or -1
	local nextCloser = string.find(text, ">", 2) or -1
	if nextCloser > -1 and (nextOpener == -1 or nextCloser < nextOpener) then
		local startingTag = string.sub(text, 1, nextCloser)
		--We have our starting tag (E.g. '<pre style="color:red">')
		--Now find our ending...
		if endswith(startingTag, "/>") then --self-closing tag (we are our own ending)
			return {
				Tag = tagName,
				Start = startingTag,
				Content = "", End = "",
				Length = #startingTag
			}

		else
			local endingTag
			if tagName == "pre" then --Looser restrictions for <pre>
				endingTag = --no | so we just use 2 matches
					string.match(text, "</[Pp][Rr][Ee]>") or
					string.match(text, "</[Pp][Rr][Ee][ \t\n/][^<]*>")
			else
				endingTag = string.match(text, "</"..allcases(tagName).."[ \t\n]*>")
			end

			if endingTag then --Regular tag formation
				local endingTagPosition = string.find(text, endingTag, nextCloser, true)
				local tagContent = string.sub(text, nextCloser+1, endingTagPosition-1)
				return {
					Tag = tagName,
					Start = startingTag,
					Content = tagContent,
					End = endingTag,
					Length = #startingTag + #tagContent + #endingTag
				}

			else --Content inside still needs escaping (also linter error!)
				return {
					Tag = tagName,
					Start = startingTag,
					Content = "", End = "",
					Length = #startingTag
				}
			end
		end
	end
	return nil
end
local function TestForComment(text) --Like TestForNowikiTag but for <!-- -->
	if startswith(text, "<!--") then
		local commentEnd = string.find(text, "-->", 5, true)
		if commentEnd then
			return {
				Start = "<!--", End = "-->",
				Content = string.sub(text, 5, commentEnd-1),
				Length = commentEnd+2
			}
		else --Consumes all text if not given an ending
			return {
				Start = "<!--", End = "",
				Content = string.sub(text, 5),
				Length = #text
			}
		end
	end
	return nil
end

--[[ Implementation notes
The goal of this function is to escape all text that wouldn't be parsed if it
was preprocessed (see above implementation notes).

Using keepComments will keep all HTML comments instead of removing them. They
will still be escaped regardless to avoid processing errors
--]]
local function PrepareText(text, keepComments)
	local newtext = ""
	while text ~= "" do
		local NextCheck = string.find(text,"<[NnSsPpMm!]") --Advance to the next potential tag we care about
		if not NextCheck then --Done
			newtext = newtext .. text
			break
		end
		newtext = newtext .. string.sub(text,1,NextCheck-1)
		text = string.sub(text, NextCheck)
		local Comment = TestForComment(text)
		if Comment then
			if keepComments then
				newtext = newtext .. Comment.Start .. mw.text.nowiki(Comment.Content) .. Comment.End
			end
			text = string.sub(text, Comment.Length+1)
		else
			local Tag = TestForNowikiTag(text)
			if Tag then
				local newTagStart = "<" .. mw.text.nowiki(string.sub(Tag.Start,2,-2)) .. ">"
				local newTagEnd = 
					Tag.End == "" and "" or --Respect no tag ending
					"</" .. mw.text.nowiki(string.sub(Tag.End,3,-2)) .. ">"
				local newContent = mw.text.nowiki(Tag.Content)
				newtext = newtext .. newTagStart .. newContent .. newTagEnd
				text = string.sub(text, Tag.Length+1)
			else --Nothing special, move on...
				newtext = newtext .. string.sub(text, 1, 1)
				text = string.sub(text, 2)
			end
		end
	end
	return newtext
end

--[=[ Implementation notes (This needs re-doing pending new changes...........)
This function is an alternative to Transcluder's getParameters which considers
the potential for a singular { or } or other odd syntax that %b doesn't like to
be in a parameter's value. Also theoretically faster as it does a singular pass
through the text instead of multiple gsub runs.

When handling the difference between {{ and {{{, mediawiki will attempt to match
as many sequences of {{{ as possible before matching a {{
E.g.
 {{{{A}}}} -> { {{{A}}} }
 {{{{{{{{Text|A}}}}}}}} -> {{ {{{ {{{Text|A}}} }}} }}
The trouble here is when the amount of opening and closing braces don't match
E.g.
 {{{!}} -> { {{!}} -> {|
 {{{{{A}}}} -> {{ {{{A}}} }
The rough idea is to match based off of the smallest amount of braces from the
two sides and treat the excess as regular text.

Templates beginning with the characters listed
[[Wikipedia:Page name#Technical restrictions and limitations|here]] won't be
considered as valid syntax inside {{ }}
Since we can't account for literally every scenario without it getting insane
very fast, syntax like {{{{{A}}}}} will be considered invalid, claiming {{{A}}}
is an invalid name regardless of what it resolves to. This means we don't have
to consider some complications. While this obviously isn't perfect, it's the
best approach I can form right now.

Setting dontEscape will prevent running the input text through EET. Only do this
if it's already been passed through manually.
--]=]
local function ParseTemplates(text, dontEscape) --Need to think about how the execution logic on this is gonna work, so its not done yet
	if not dontEscape then
		text = PrepareText(text)
	end
	--Step 1: Find and escape the content of all wikilinks on the page, which are stronger than templates (see implementation notes)
	local scannerPosition = 1
	local wikilinks = {}
	local openWikilinks = {}
	while text ~= "" do
		local NextOpen = string.find(text,"%[%[", scannerPosition) or 9e9
		local NextClose = string.find(text,"]]", scannerPosition) or 9e9
		--mw.log("Debug time | Step 1")
		--mw.log("NO, NC =", NextOpen, ",", NextClose)
		--mw.logObject(wikilinks)
		--mw.logObject(openWikilinks)
		if NextOpen == NextClose then --Done (both 9e9)
			break
		end
		local NextCheck = math.min(NextOpen or 9e9, NextClose or 9e9) --Skip to next notable block
		scannerPosition = NextCheck+2 --+2 to pass the [[ / ]]
		if NextOpen < NextClose then --Add a [[ to the pending wikilink queue
			table.insert(openWikilinks, NextOpen)
		else --Pair up the ]] to any available [[
			if #openWikilinks >= 1 then
				local start = table.remove(openWikilinks) --Pop the latest [[
				table.insert(wikilinks, {start, NextClose+1}) --Note the pair
				--mw.log("Suspected text:", string.sub(text, start, NextClose+1))
			end
		end
	end
	if true then return end --not done!
	
	--Step 2: Find the bounds of every valid template, figuring out if a set should be treated as {{ or {{{ as needed
	local templates = {}
	local templateData = {}
	while text ~= "" do
		local NextTemplate = string.find(text,"{{") --Skip to next notable block
		if not NextTemplate then --Done
			break
		end
		newtext = newtext .. string.sub(text,1,NextCheck-1)
		text = string.sub(text, NextCheck)
	end
	
	--Step 3: Re-trace the templates using their known bounds, collecting our parameters with ease
	return templates, templateData
end

--Main entry points
p.PrepareText = PrepareText
p.ParseTemplates = ParseTemplates
--Extra entry points, not really required
p.TestForNowikiTag = TestForNowikiTag
p.TestForComment = TestForComment

return p

--[==[ console tests

local s = [=[Hey!{{Text|<nowiki | ||>
Hey! }}
A</nowiki>|<!--AAAAA|AAA-->Should see|Shouldn't see}}]=]
local out = p.PrepareText(s)
mw.logObject(out)

local s = [=[<!--
Hey!
-->A]=]
local out = p.TestForComments(s)
mw.logObject(out); mw.log(string.sub(s, 1, out.Length))

p.ParseTemplates([=[
]]Hey!
A[[a]][[b]]B
Done[[
]=])

]==]