Pandoc filter to convert math to inline svg using latex and dvisvgm

texsvg.hs

{-# LANGUAGE OverloadedStrings #-}
import Text.Pandoc.JSON
import System.Directory
import System.FilePath ((</>))
import qualified Data.Hash.MD5 as MD5
import qualified Data.Text as T
import System.IO.Temp
import System.Process
import Control.Monad (unless)

main :: IO ()
main = toJSONFilter mathToSvg

mathToSvg :: Inline -> IO Inline
mathToSvg m@(Math mathType x) = do
  let wrap = T.unpack . removeNewline . case mathType of
                   InlineMath -> \x' -> "\\(" <> x' <> "\\)"
                   DisplayMath -> \x' -> "\\[" <> x' <> "\\]"
      preamble =[
          "\\documentclass[border=1pt,varwidth]{standalone}",
          "\\usepackage{standalone}" <>
          "\\usepackage{amsmath}" <>
          "\\usepackage{amssymb}" <>
          "\\usepackage{cancel}" <>
          "\\begin{document}"
        ]
      postamble = [ "\\end{document}" ]
      removeNewline = T.filter (`notElem` ("\r\n" :: [Char]))

  tempDir <- getTemporaryDirectory
  let cacheDir = tempDir </> "pandoc.texsvg.cache"
  createDirectoryIfMissing True cacheDir
  let mathHash = MD5.md5s $ MD5.Str $ show m
      outfilename =  cacheDir </> mathHash <> ".svg"

  fileExists <- doesFileExist outfilename

  unless fileExists $
    withSystemTempDirectory "pandoc.dir" $ \tmpDir ->
      do
        origDir <- getCurrentDirectory
        setCurrentDirectory tmpDir
        _ <- readProcess "latex" (preamble <> [wrap x] <> postamble) []
        _ <- readProcess "dvisvgm"
                 ["-b2pt", "-Z1.2", "-n", "-o", outfilename, "standalone.dvi"] []
        setCurrentDirectory origDir

  svg <- T.pack <$> readFile outfilename
  return $ RawInline (Format "html") $ case mathType of
    InlineMath -> svg
    DisplayMath -> "<p>" <> svg <> "</p>"
mathToSvg x = return x

Yeah... this filter is pretty outdated. Let me see if I can easily update it.

Okay, updated, but tbh, it'd be probably easier to write a lua filter that would essentially do the same thing. All this filter does is basically run latex and then convert its output to SVG using dvisvgm for each math element in document. Should be pretty easy to implement in pandoc Lua I think. The only caveat is it does some hashing for caching, because otherwise it's very slow, and hashing isn't that easy with Lua. Doable though.

gotit, thanks! P

Hi @lierdakil i try to make a lua filter, but i am a newer in either lua or haskell, and i failed to install pandoc via cabal...thus i try to make a lua version:

local default_template = [[
\documentclass[12pt,preview]{standalone}
{{ preamble }}
\begin{document}
\begin{preview}
{{ code }}
\end{preview}
\end{document}
]]

local default_preamble = [[
\usepackage[utf8x]{inputenc}
\usepackage{amsmath}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amstext}
\usepackage{newtxtext}
\usepackage{mhchem}
\usepackage{chemfig}
\usepackage[libertine]{newtxmath}
\DeclareOldFontCommand{\rm}{\normalfont\rmfamily}{\mathrm}
\DeclareOldFontCommand{\sf}{\normalfont\sffamily}{\mathsf}
\DeclareOldFontCommand{\tt}{\normalfont\ttfamily}{\mathtt}
\DeclareOldFontCommand{\bf}{\normalfont\bfseries}{\mathbf}
\DeclareOldFontCommand{\it}{\normalfont\itshape}{\mathit}
\DeclareOldFontCommand{\sl}{\normalfont\slshape}{\@nomath\sl}
\DeclareOldFontCommand{\sc}{\normalfont\scshape}{\@nomath\sc}
\newcommand{\N}{\mathbb{N}}
\newcommand{\R}{\mathbb{R}}
\newcommand{\Z}{\mathbb{Z}}
]]

function exec_math (elem)
    -- Convert Math El -> DVI via latex --
    mathEq = default_template:gsub("{{ preamble }}", default_preamble):gsub("{{ code }}", elem.text)
    local tex = io.open("input.tex", "w")
    tex:write(mathEq)
    io.close(tex)
    os.execute("latex input.tex")
    os.execute("dvisvgm --no-fonts input.dvi output.svg")
    local svg = io.open("output.svg", "rb")
    if not svg then
        return nil
    end
    local content = svg:read "*a"
    local output
    if elem.mathtype == 'InlineMath' then
        output = content
    else
        output = "<p>" .. content .. "<p>"
    end
    elem = pandoc.RawInline('html', output)
    return elem
end

return {
    { Math = exec_math }
}

but after generating svg with markdown math $\chemfig{C([2]-H)([6]=H)}$ and return a RawInline elem, it still renders a <span class="math inline">$\chemfig{C([2]-H)([6]=H)}$</span>, and log a warning in the console:

[WARNING] Could not convert TeX math \chemfig{C([2]-H)([6]=H)}, rendering as TeX:
  \chemfig{C([2]-H)([6]=H)}
          ^
  unexpected control sequence \chemfig
  expecting "%", "\\label", "\\tag", "\\nonumber" or whitespace

Do you have any idea how to fix it?.....

a google group link pubed by myself.. https://groups.google.com/g/pandoc-discuss/c/4CirnNd-ETs

To begin with, your call to dvisvgm is not quite right... Try using dvisvgm --no-fonts input.dvi -o output.svg. Besides, I'm pretty sure that call to table.insert will fail, since neither argument is a table. I don't think you need it to begin with.

sorry but i find that you're right... as i have read a wrong document... thx:)

Here's my gist https://gist.github.com/PassionPenguin/43fc09b942cedc428ae246bff8b6e193 :)