GitHub uses what we're calling "GitHub Flavored Markdown" (GFM) for messages, issues, and comments. It differs from standard Markdown (SM) in a few significant ways and adds some additional functionality.

If you're not already familiar with Markdown, you should spend 15 minutes and go over the excellent Markdown Syntax Guide at Daring Fireball.

If you prefer to learn by example, see the following source and result:

You can also use our live javascript preview to see GFM in action.

Differences from traditional Markdown

Newlines

The biggest difference that GFM introduces is in the handling of linebreaks. With SM you can hard wrap paragraphs of text and they will be combined into a single paragraph. We find this to be the cause of a huge number of unintentional formatting errors. GFM treats newlines in paragraph-like content as real line breaks, which is probably what you intended.

The next paragraph contains two phrases separated by a single newline character:

Roses are red
Violets are blue

becomes

Roses are red
Violets are blue

Multiple underscores in words

It is not reasonable to italicize just part of a word, especially when you're dealing with code and names often appear with multiple underscores. Therefore, GFM ignores multiple underscores in words.

perform_complicated_task
do_this_and_do_that_and_another_thing

becomes

perform_complicated_task
do_this_and_do_that_and_another_thing

URL autolinking

GFM will autolink standard URLs, so if you want to link to a URL (instead of setting link text), you can simply enter the URL and it will be turned into a link to that URL.

A bit of the GitHub spice

In addition to the changes in the previous section, certain references are auto-linked:

* SHA: be6a8cc1c1ecfe9489fb51e4869af15a13fc2cd2
* User@SHA ref: mojombo@be6a8cc1c1ecfe9489fb51e4869af15a13fc2cd2
* User/Project@SHA: mojombo/god@be6a8cc1c1ecfe9489fb51e4869af15a13fc2cd2
* \#Num: #1
* User/#Num: mojombo#1
* User/Project#Num: mojombo/god#1

becomes

Code

The newline and underscore modification code can be seen below. If you find a bug in the rendering, we'd love to hear about it.

require 'digest/md5'

def gfm(text)
  # Extract pre blocks
  extractions = {}
  text.gsub!(%r{<pre>.*?</pre>}m) do |match|
    md5 = Digest::MD5.hexdigest(match)
    extractions[md5] = match
    "{gfm-extraction-#{md5}}"
  end

  # prevent foo_bar_baz from ending up with an italic word in the middle
  text.gsub!(/(^(?! {4}|\t)\w+_\w+_\w[\w_]*)/) do |x|
    x.gsub('_', '\_') if x.split('').sort.to_s[0..1] == '__'
  end

  # in very clear cases, let newlines become <br /> tags
  text.gsub!(/^[\w\<][^\n]*\n+/) do |x|
    x =~ /\n{2}/ ? x : (x.strip!; x << " \n")
  end

  # Insert pre block extractions
  text.gsub!(/\{gfm-extraction-([0-9a-f]{32})\}/) do
    "\n\n" + extractions[$1]
  end

  text
end

if $0 == __FILE__
  require 'test/unit'
  require 'shoulda'

  class GFMTest < Test::Unit::TestCase
    context "GFM" do
      should "not touch single underscores inside words" do
        assert_equal "foo_bar", gfm("foo_bar")
      end

      should "not touch underscores in code blocks" do
        assert_equal " foo_bar_baz", gfm(" foo_bar_baz")
      end

      should "not touch underscores in pre blocks" do
        assert_equal "\n\n<pre>\nfoo_bar_baz\n</pre>", gfm("<pre>\nfoo_bar_baz\n</pre>")
      end

      should "not treat pre blocks with pre-text differently" do
        a = "\n\n<pre>\nthis is `a\\_test` and this\\_too\n</pre>"
        b = "hmm<pre>\nthis is `a\\_test` and this\\_too\n</pre>"
        assert_equal gfm(a)[2..-1], gfm(b)[3..-1]
      end

      should "escape two or more underscores inside words" do
        assert_equal "foo\\_bar\\_baz", gfm("foo_bar_baz")
      end

      should "turn newlines into br tags in simple cases" do
        assert_equal "foo \nbar", gfm("foo\nbar")
      end

      should "convert newlines in all groups" do
        assert_equal "apple \npear \norange\n\nruby \npython \nerlang",
                     gfm("apple\npear\norange\n\nruby\npython\nerlang")
      end

      should "convert newlines in even long groups" do
        assert_equal "apple \npear \norange \nbanana\n\nruby \npython \nerlang",
                     gfm("apple\npear\norange\nbanana\n\nruby\npython\nerlang")
      end

      should "not convert newlines in lists" do
        assert_equal "# foo\n# bar", gfm("# foo\n# bar")
        assert_equal "* foo\n* bar", gfm("* foo\n* bar")
      end
    end
  end
end
view raw gfm.rb This Gist brought to you by GitHub.