* master: (58 commits)
Update history to reflect merge of #4792 [ci skip]
Update history to reflect merge of #4793 [ci skip]
Update history to reflect merge of #4804 [ci skip]
Update history to reflect merge of #4754 [ci skip]
Update history to reflect merge of #4813 [ci skip]
Added missing single quote on rsync client side command
Add v3.0.4 and v3.1.3 to the history.
Fixed typo
Add jekyll-autoprefixer plugin
Explicitly require Filters rather than implicitly.
Update history to reflect merge of #4786 [ci skip]
Update history to reflect merge of #4789 [ci skip]
updates example domain in config template
Globalize Jekyll's Filters.
Update JRuby to 9.0.5.0; Drop the double digit test.
Update Rack-Jekyll Heroku deployment blog post url
convertible: use Document::YAML_FRONT_MATTER_REGEXP to parse transformable files
Update history to reflect merge of #4734 [ci skip]
Update history to reflect merge of #4478 [ci skip]
Fix rubocop warning.
...
As it stands Jekyll does not globalize it's filters. So anybody wishing to go
into Jekyll's context to process their own Liquid (say in a plugin) may be taken
aback when they find out that Jekyll's filters are not available.
See: jekyll/jekyll-assets#252.
Prior to this change, the related posts for the most recently rendered post
stayed set on the `site` object. This could result in pages that showed related
posts even when the page represented an entire collection of posts, such as on
an index page. This change restores the functionality from Jekyll V2.
Previously, even if the document permalink was a folder, it would look for
an extension on that. For example, if I have:
permalink: "/new-version-jekyll-v3.0.0/"
the output_ext would be ".0". Now, the output_ext honors the trailing
slash and will report based on the converters instead.
The properties of Liquid::Drops are only evaluated when they're asked for
and therefore save computation time. This prevents a lot of GC time cleaning
up objects that are not needed, because they're not created unless requested.
Additionally, this saves time for actual computation of those values because
they can be computed only if needed.
It's funny how much it helps when you only do what is needed. Far less overhead.