When using Ruby 1.8.7 and Kramdown 0.14.0 and newer, the following build error
would occur if any page in your site contained a `<name@example.com>` email
address link:
Generating... Conversion error: There was an error converting 'contribute.markdown'.
/Users/nick/Documents/puppet-docs/vendor/bundle/ruby/1.8/gems/kramdown-1.0.2/lib/kramdown/converter/html.rb:404:in `obfuscate': undefined method `encoding' for "mailto":String (NoMethodError)
See also:
http://rubyforge.org/tracker/index.php?func=detail&aid=29750&group_id=7403&atid=28673
This problem traced back to the following line in the Kramdown source:
result.force_encoding(text.encoding) if result.respond_to?(:force_encoding)
Strings aren't supposed to respond to the :force_encoding method in Ruby < 1.9,
but lib/jekyll/core_ext.rb was modifying the string class like so:
def force_encoding(enc)
self
end
Strings still won't respond to :encoding, though, so we get an error when
Kramdown tries to read the incoming encoding and everything will blow up.
An ack of the codebase suggests that this was only added so we could force an
encoding for rdiscount on 1.9 without having to check whether we were running
under 1.8. Since testing for said method to learn whether one is running under a
1.9-like encoding regime seems to be a thing in libraries we rely on, we
shouldn't insert this dummy method without also dummying every other part of the
Ruby 1.9+ encoding system.
This commit removes the dummy :force_encoding method to stop poisoning core
classes for libraries we use, and moves the adjustment for 1.9-like encoding
regimes to the one place where it's needed.
This commit makes Jekyll threadsafe (or at least makes it possible to be so).
It also makes it a ton easier to use Jekyll as a library for scripted site
transformation. I did, however, break all the tests. =(
Parker Moore
Shigeya Suzuki
Nick Fagerlund
zachgersh
larrylv
mccxj
Kris Brown
Nick Quaranto
Tom Preston-Werner