commonmarker: Error "incompatible character encodings: UTF-8 and ASCII-8BIT" when combined with a rails app

I think this might not be a commonmarker problem, BUT the error is not raised when using pandoc-ruby nor redcarpet, so it has something to do with commonmarker.

Here you can see a test run from the command line with both cmark and commonmarker and there’s no problem:

$ cat test-curly-quotes.md
This curly quote “makes commonmarker throw an exception”.

$ cmark --version
cmark 0.20.0 - CommonMark converter
(C) 2014, 2015 John MacFarlane

$ cmark test-curly-quotes.md
<p>This curly quote “makes commonmarker throw an exception”.</p>

$ gem list --local commonmarker

*** LOCAL GEMS ***

commonmarker (0.2.0)

$ cat test-curly-quotes.md | ruby -r commonmarker -e "puts CommonMarker.render_html(gets)"
<p>This curly quote “makes commonmarker throw an exception”.</p>

That said, I’m testing different markdown parsers/renderers for our rails 4.1.12 (ruby 2.2.2) based app and I’m getting the following error:

ActionView::Template::Error (incompatible character encodings: UTF-8 and ASCII-8BIT):
    12:       - if user_signed_in?
    13:         .outline-content
    14:           = commonmarker_markdown(@quimbee_outline.source)
  app/views/outlines/show.html.slim:15:in `_app_views_outlines_show_html_slim___3317075370232322437_70158621096300'


  Rendered /Users/oboxodo/.rbenv/versions/2.2.2/lib/ruby/gems/2.2.0/gems/actionpack-4.1.12/lib/action_dispatch/middleware/templates/rescues/_trace.html.erb (2.9ms)
  Rendered /Users/oboxodo/.rbenv/versions/2.2.2/lib/ruby/gems/2.2.0/gems/actionpack-4.1.12/lib/action_dispatch/middleware/templates/rescues/_request_and_response.html.erb (1.7ms)
  Rendered /Users/oboxodo/.rbenv/versions/2.2.2/lib/ruby/gems/2.2.0/gems/actionpack-4.1.12/lib/action_dispatch/middleware/templates/rescues/template_error.html.erb within rescues/layout (69.1ms)

I have these helpers:

# encoding: UTF-8
module ApplicationHelper
  def commonmarker_markdown(text)
    CommonMarker.render_html(text, :smart).html_safe
  end

  def pandoc_markdown(text)
    converter = PandocRuby.new(text, from: :markdown, to: :html)
    converter.convert.html_safe
  end

  def redcarpet_markdown(text)
    # ...
  end
end

Changing the call to commonmarker_markdown to either pandoc_markdown or redcarpet_markdown renders the expected result with no errors.

It’s not a DB (postgresql) encoding problem either as hardcoding the test phrase in place of the text variable (no DB involved) causes the same problem.

Any ideas about what could be happening?