Jekyll with Rouge and KaTeX cannot have $ in code

I am creating a new Jekyll blog setup where I want to self-host and generate server-side everything I can (so basically everything besides Disqus and Google Analytics). I want my site to work well even on browsers with JS turned off, which is why I turned to Kramdown+Rouge for syntax highlighting and KaTex for rendeding TeX formula during site generation.

I found a problem though. When I have example like:

```bash
$ echo test
test
```

It will break inside KaTeX command:

{% katexmm %}
{{ content }}
{% endkatexmm %}

I figured out the cause:

rouge turns snippet into something like <span class="gp">$</span>
katexmm extects that each $ is paired with another $ or escaped as \$

What I want to know is how to fix this, e.g. by escaping all $ inside pre, but leaving other $ (which are intended to use as actual LaTeX snippets) intact. Or maybe somehow configuring jekyll-katex to ignore unpaired $? (throw_error: false option works after text was already matched using dollar signs, so it doesn't help).

I want to fix this in config or in place of applying katexmm, so that I wouldn't have to modify the content of any post.

Solution

I managed to get rid of errors by modifying tag from jekyll-katex to work around code (inline, use single tilde) and pre code (block made using indention or 3 tildes):

# frozen_string_literal: true

require 'jekyll'
require 'jekyll-katex/configuration'
require 'jekyll-katex/katex_js'
require 'nokogiri'

module Jekyll
  module Tags
    # Defines the custom Liquid tag for compile-time rendering of KaTeX math.
    # This differs from the katex tag in that it allows use of `$` and `$$` fencing to mark math mode blocks similar to
    # standard latex.
    #   {% katexmm %}
    #   This is a mixed environment where you can write text as normal but fence off latex math using `$`. Escape
    #   using `\$`. For example.
    #   $latex math with \$$
    #   $$display mode latex$$
    #   {% endkatexmm %}
    class KatexMathModeFixed < Liquid::Block
      LOG_TOPIC = 'KatexMathModeFixed:'
      KATEX ||= Jekyll::Katex::KATEX_JS
      LATEX_TOKEN_PATTERN = /(?<!\\)([$]{2}|[$]{1})(.+?)(?<!\\)\1/m

      def initialize(tag_name, markup, tokens)
        super
        @markup = markup
        @tokens = tokens
        @display_mode_rendering = Jekyll::Katex::Configuration.global_rendering_options.merge(displayMode: true)
        @inline_mode_rendering = Jekyll::Katex::Configuration.global_rendering_options.merge(displayMode: false)
      end

      def render(context)
        enclosed_block = super
        fixed_block = fix_code(enclosed_block)
        rendered_str = fixed_block.to_s.gsub(LATEX_TOKEN_PATTERN) do |match|
          display_mode = match.to_s.start_with? '$$'
          rendering_options = display_mode ? @display_mode_rendering : @inline_mode_rendering
          Jekyll.logger.debug LOG_TOPIC, "Rendering matched block - #{match}"
          KATEX.call('katex.renderToString', Regexp.last_match(2), rendering_options)
        end
        # KaTeX should fix escaped `$` within fenced blocks, this addresses instances outside of math mode
        rendered_str.to_s.gsub(/\\[$]/, '$').to_s
      end

      def fix_code(input)
        updated = false
        html = Nokogiri::HTML.fragment(input)
        Jekyll.logger.debug LOG_TOPIC, "Fixing - #{input}"
        html.css("code, code span").each do |c|
          if c.css('*').empty? && c.content['$']
            updated = true
            Jekyll.logger.debug LOG_TOPIC, "current tag - #{c}"
            content = c.content
            content['$'] = '\$'
            c.content = content
            Jekyll.logger.debug LOG_TOPIC, "current tag now - #{c}/#{content}"
          end
        end
        output = html.to_s
        Jekyll.logger.debug LOG_TOPIC, "Fixed - #{output}"
        if updated then html.to_s else input end
      end
    end
  end
end

Liquid::Template.register_tag('katexmmx', Jekyll::Tags::KatexMathModeFixed)

It's installable in _plugins directory.

Thing is, this is still buggy - by default kramdown still tries to use mathjax engine and generates <script type="math/tex">, so that had to be changed. When I was investigating how, I found out that kramdown supports math_engine: katex as well - with that I only had to add fonts and CSS, and jekyll-katex become completely obsolete (as well as my workaround, which I'll leave here if anyone is curious).