\subsection{\translation{Markdown syntax extension}{Markdown syntaxuitbreiding}}

\package{markdown} lets a document add rules to the grammar of its reader. This file adds one, and it
exists because of a limit that has no way around it in Markdown itself: an identifier can be attached
to a heading and to a bracketed span, and a bracketed span may not contain another span or a link
written with brackets. A paragraph that carries a label and cites a definition therefore cannot be
written as one span, and the label has to be pushed onto the opening words of the sentence.

With this extension the identifier stands on its own, before the text it labels:

\begin{lstlisting}[style=md]
1. {#lid:lorem} Een [persoonsgegeven](#pg) wordt verwerkt, zie <#lid:eerder>.
\end{lstlisting}

What is added is syntax, not typesetting: a renderer decides how a construct that the reader already
recognises is set, and no renderer can make the reader recognise something new. The extension is read
by \package{markdown} itself, in Lua, before any of that.

The interface is a table with three fields. Both versions are checked, and \texttt{grammar\_version}
for equality rather than for a lower bound, so an update of \package{markdown} that changes its
grammar makes this file an error instead of a surprise.
\begin{lstlisting}[language={[5.3]Lua},numbers=left]
local regulatory_syntax = {
  api_version = 2,
  grammar_version = 4,

\end{lstlisting}
\texttt{finalize\_grammar} is handed the reader, whose grammar can be added to. The pattern is an
identifier between braces, spelled the way Pandoc spells one, and what it produces is a call of a
renderer that this bundle defines in \package{regulatory-md}.

The characters allowed in an identifier are the ones a label of this bundle is made of: letters,
digits, and the punctuation that separates a type from a name, \texttt{art:lorem} and
\texttt{ex2-lid:lorem}.
\begin{lstlisting}[language={[5.3]Lua},firstnumber=last,numbers=left]
  finalize_grammar = function(reader)
    local labelchar = lpeg.R("az", "AZ", "09") + lpeg.S(":.-_")
    local label = lpeg.P("{#") * lpeg.C(labelchar^1) * lpeg.P("}")
      / function(identifier)
          return {"\\markdownRendererRegulatoryLabel{", identifier, "}"}
        end

\end{lstlisting}
The rule is spliced in front of the one that reads links and emphasis, so that a brace opening an
identifier is seen before anything else can claim it. A brace is not a character the reader stops at
on its own, which is what the second call says: without it the brace would be swallowed by the run of
ordinary text around it and the pattern would never be tried.

Only \verb|{#| starts the pattern, so a brace anywhere else in the text is left alone.
\begin{lstlisting}[language={[5.3]Lua},firstnumber=last,numbers=left]
    reader.insert_pattern("Inline before LinkAndEmph", label, "RegulatoryLabel")
    reader.add_special_character("{")
  end
}

return regulatory_syntax
\end{lstlisting}