Here’s a quick performance comparison, converting the pandoc manual (MANUAL.txt) to HTML, with versions of the same JSON filter written in compiled Haskell ( smallcaps) and interpreted Python ( smallcaps.py): CommandĪs you can see, the Lua filter avoids the substantial overhead associated with marshaling to and from JSON over a pipe. To run it, save it in a file, say a, and invoke pandoc with -lua-filter=a. This says: walk the AST, and when you find a Strong element, replace it with a SmallCaps element with the same content. Here is an example of a Lua filter that converts strong emphasis to small caps:įunction Strong (elem ) return pandoc. Pandoc data types are marshaled to Lua directly, avoiding the overhead of writing JSON to stdout and reading it from stdin.

A Lua interpreter (version 5.3) and a Lua library for creating pandoc filters is built into the pandoc executable. Starting with version 2.0, pandoc makes it possible to write filters in Lua without any external dependencies at all. One cannot simply provide a filter that can be used by anyone who has a certain version of the pandoc executable. A filter may require an interpreter for a certain programming language to be available, as well as a library for manipulating the pandoc AST in JSON form.

Second, whether a filter will work will depend on details of the user’s environment.

First, there is some overhead in writing JSON to stdout and reading it from stdin (twice, once on each side of the filter). They may be written in any programming language, and invoked from pandoc using the -filter option.Īlthough traditional filters are very flexible, they have a couple of disadvantages. Traditional pandoc filters accept a JSON representation of the pandoc AST and produce an altered JSON representation of the AST. Pandoc has long supported filters, which allow the pandoc abstract syntax tree (AST) to be manipulated between the parsing and the writing phase.

Lua table insert multiple values