The variables keywords, keys, values seem to be used just once across the entire lexer. You've two choices:

• You can either instruct the Ruby interpreter to interpolate a regex just once (for performance reasons) by using the o flag: %r/#{foo}/o.
• Or if you don't foresee a possibility for a new lexer being subclassed from the HOCON class, you may obliterate those variables entirely and have static regexps right away:

  # values
  rule %r/[^\$\"{}\[\]:=,\+#`\^\?!@\*&]+?/, Literal

I've removed the variables.

@davidtwco Thanks for submitting this again and I'm sorry you've had to wait so long :(

I'm looking at the informal spec and, given HOCON is a superset of JSON, it looks like this lexer would be best implemented as a subclass of Rouge::Lexers::JSON. You'd avoid needing to reimplement all the stuff that carries across from JSON and just be able to focus on the stuff that's different.

In addition to simply redefining a rule that you need to operate differently, Rouge comes with the methods prepend and append that you can use to manipulate an existing rule. You can see an example of using the JSON lexer as a superclass in the JSONDOC lexer.

@pyrmont I've implemented this as a subclass of JSON, I think I've eliminated all the redundant rules and those that remain are necessary.

Wow, that's fast! I'll have a look over it and let you know if I spot anything :)

Pushed again to (hopefully) fix Travis failures and fix a missing word in a comment.