Markdown is a plain text format for writing structured documents, based on conventions used for indicating formatting in email and usenet posts. It was developed in 2004 by John Gruber, who wrote the first Markdown-to-HTML converter in perl, and it soon became widely used in websites. By 2014 there were dozens of implementations in many languages. Some of them extended basic Markdown syntax with conventions for footnotes, definition lists, tables, and other constructs, and some allowed output not just in HTML but in LaTeX and many other formats.
John Gruber’s canonical description of Markdown’s syntax does not specify the syntax unambiguously. Here are some examples of questions it does not answer:
How much indentation is needed for a sublist? The spec says that
continuation paragraphs need to be indented four spaces, but is
not fully explicit about sublists. It is natural to think that
they, too, must be indented four spaces, but Markdown.pl
does
not require that. This is hardly a “corner case,” and divergences
between implementations on this issue often lead to surprises for
users in real documents. (See this comment by John
Gruber.)
Is a blank line needed before a block quote or heading? Most implementations do not require the blank line. However, this can lead to unexpected results in hard-wrapped text, and also to ambiguities in parsing (note that some implementations put the heading inside the blockquote, while others do not). (John Gruber has also spoken in favor of requiring the blank lines.)
Is a blank line needed before an indented code block?
(Markdown.pl
requires it, but this is not mentioned in the
documentation, and some implementations do not require it.)
paragraph
code?
What is the exact rule for determining when list items get
wrapped in <p>
tags? Can a list be partially “loose” and partially
“tight”? What should we do with a list like this?
1. one
2. two
3. three
Or this?
1. one
- a
- b
2. two
(There are some relevant comments by John Gruber here.)
Can list markers be indented? Can ordered list markers be right-aligned?
8. item 1
9. item 2
10. item 2a
Is this one list with a thematic break in its second item, or two lists separated by a thematic break?
* a
* * * * *
* b
When list markers change from numbers to bullets, do we have two lists or one? (The Markdown syntax description suggests two, but the perl scripts and many other implementations produce one.)
1. fee
2. fie
- foe
- fum
What are the precedence rules for the markers of inline structure? For example, is the following a valid link, or does the code span take precedence ?
[a backtick (`)](/url) and [another backtick (`)](/url).
What are the precedence rules for markers of emphasis and strong emphasis? For example, how should the following be parsed?
*foo *bar* baz*
What are the precedence rules between block-level and inline-level structure? For example, how should the following be parsed?
- `a long code span can contain a hyphen like this
- and it can screw things up`
Can list items include section headings? (Markdown.pl
does not
allow this, but does allow blockquotes to include headings.)
- # Heading
Can list items be empty?
* a
*
* b
Can link references be defined inside block quotes or list items?
> Blockquote [foo].
>
> [foo]: /url
If there are multiple definitions for the same reference, which takes precedence?
[foo]: /url1
[foo]: /url2
[foo][]
In the absence of a spec, early implementers consulted Markdown.pl
to resolve these ambiguities. But Markdown.pl
was quite buggy, and
gave manifestly bad results in many cases, so it was not a
satisfactory replacement for a spec.
Because there is no unambiguous spec, implementations have diverged considerably. As a result, users are often surprised to find that a document that renders one way on one system (say, a github wiki) renders differently on another (say, converting to docbook using pandoc). To make matters worse, because nothing in Markdown counts as a “syntax error,” the divergence often isn’t discovered right away.
This document attempts to specify Markdown syntax unambiguously.
It contains many examples with side-by-side Markdown and
HTML. These are intended to double as conformance tests. An
accompanying script spec_tests.py
can be used to run the tests
against any Markdown program:
python test/spec_tests.py --spec spec.txt --program PROGRAM
Since this document describes how Markdown is to be parsed into an abstract syntax tree, it would have made sense to use an abstract representation of the syntax tree instead of HTML. But HTML is capable of representing the structural distinctions we need to make, and the choice of HTML for the tests makes it possible to run the tests against an implementation without writing an abstract syntax tree renderer.
This document is generated from a text file, spec.txt
, written
in Markdown with a small extension for the side-by-side tests.
The script tools/makespec.py
can be used to convert spec.txt
into
HTML or CommonMark (which can then be converted into other formats).
In the examples, the →
character is used to represent tabs.
Any sequence of characters is a valid CommonMark document.
A character is a Unicode code point. Although some code points (for example, combining accents) do not correspond to characters in an intuitive sense, all code points count as characters for purposes of this spec.
This spec does not specify an encoding; it thinks of lines as composed of characters rather than bytes. A conforming parser may be limited to a certain encoding.
A line is a sequence of zero or more characters
other than newline (U+000A
) or carriage return (U+000D
),
followed by a line ending or by the end of file.
A line ending is a newline (U+000A
), a carriage return
(U+000D
) not followed by a newline, or a carriage return and a
following newline.
A line containing no characters, or a line containing only spaces
(U+0020
) or tabs (U+0009
), is called a blank line.
The following definitions of character classes will be used in this spec:
A whitespace character is a space
(U+0020
), tab (U+0009
), newline (U+000A
), line tabulation (U+000B
),
form feed (U+000C
), or carriage return (U+000D
).
Whitespace is a sequence of one or more whitespace characters.
A Unicode whitespace character is
any code point in the Unicode Zs
class, or a tab (U+0009
),
carriage return (U+000D
), newline (U+000A
), or form feed
(U+000C
).
Unicode whitespace is a sequence of one or more Unicode whitespace characters.
A space is U+0020
.
A non-whitespace character is any character that is not a whitespace character.
An ASCII punctuation character
is !
, "
, #
, $
, %
, &
, '
, (
, )
,
*
, +
, ,
, -
, .
, /
, :
, ;
, <
, =
, >
, ?
, @
,
[
, \
, ]
, ^
, _
, `
, {
, |
, }
, or ~
.
A punctuation character is an ASCII
punctuation character or anything in
the Unicode classes Pc
, Pd
, Pe
, Pf
, Pi
, Po
, or Ps
.
Tabs in lines are not expanded to spaces. However, in contexts where indentation is significant for the document’s structure, tabs behave as if they were replaced by spaces with a tab stop of 4 characters.
- foo
- bar
→ - baz
<ul>
<li>foo
<ul>
<li>bar
<ul>
<li>baz</li>
</ul>
</li>
</ul>
</li>
</ul>
For security reasons, the Unicode character U+0000
must be replaced
with the REPLACEMENT CHARACTER (U+FFFD
).
We can think of a document as a sequence of blocks—structural elements like paragraphs, block quotations, lists, headings, rules, and code blocks. Some blocks (like block quotes and list items) contain other blocks; others (like headings and paragraphs) contain inline content—text, links, emphasized text, images, code, and so on.
Indicators of block structure always take precedence over indicators of inline structure. So, for example, the following is a list with two items, not a list with one item containing a code span:
This means that parsing can proceed in two steps: first, the block structure of the document can be discerned; second, text lines inside paragraphs, headings, and other block constructs can be parsed for inline structure. The second step requires information about link reference definitions that will be available only at the end of the first step. Note that the first step requires processing lines in sequence, but the second can be parallelized, since the inline parsing of one block element does not affect the inline parsing of any other.
We can divide blocks into two types: container blocks, which can contain other blocks, and leaf blocks, which cannot.
This section describes the different kinds of leaf block that make up a Markdown document.
A line consisting of 0-3 spaces of indentation, followed by a sequence
of three or more matching -
, _
, or *
characters, each followed
optionally by any number of spaces, forms a
thematic break.
Wrong characters:
Not enough characters:
One to three spaces indent are allowed:
Four spaces is too many:
More than three characters may be used:
Spaces are allowed between the characters:
Spaces are allowed at the end:
However, no other characters may occur in the line:
It is required that all of the non-whitespace characters be the same. So, this is not a thematic break:
Thematic breaks do not need blank lines before or after:
Thematic breaks can interrupt a paragraph:
If a line of dashes that meets the above conditions for being a thematic break could also be interpreted as the underline of a setext heading, the interpretation as a setext heading takes precedence. Thus, for example, this is a setext heading, not a paragraph followed by a thematic break:
When both a thematic break and a list item are possible interpretations of a line, the thematic break takes precedence:
If you want a thematic break in a list item, use a different bullet:
An ATX heading
consists of a string of characters, parsed as inline content, between an
opening sequence of 1–6 unescaped #
characters and an optional
closing sequence of any number of unescaped #
characters.
The opening sequence of #
characters must be followed by a
space or by the end of line. The optional closing sequence of #
s must be
preceded by a space and may be followed by spaces only. The opening
#
character may be indented 0-3 spaces. The raw contents of the
heading are stripped of leading and trailing spaces before being parsed
as inline content. The heading level is equal to the number of #
characters in the opening sequence.
Simple headings:
# foo
## foo
### foo
#### foo
##### foo
###### foo
<h1>foo</h1>
<h2>foo</h2>
<h3>foo</h3>
<h4>foo</h4>
<h5>foo</h5>
<h6>foo</h6>
More than six #
characters is not a heading:
At least one space is required between the #
characters and the
heading’s contents, unless the heading is empty. Note that many
implementations currently do not require the space. However, the
space was required by the
original ATX implementation,
and it helps prevent things like the following from being parsed as
headings:
A tab will not work:
This is not a heading, because the first #
is escaped:
Contents are parsed as inlines:
Leading and trailing blanks are ignored in parsing inline content:
One to three spaces indentation are allowed:
Four spaces are too much:
A closing sequence of #
characters is optional:
It need not be the same length as the opening sequence:
# foo ##################################
##### foo ##
<h1>foo</h1>
<h5>foo</h5>
Spaces are allowed after the closing sequence:
A sequence of #
characters with anything but spaces following it
is not a closing sequence, but counts as part of the contents of the
heading:
The closing sequence must be preceded by a space:
Backslash-escaped #
characters do not count as part
of the closing sequence:
### foo \###
## foo #\##
# foo \#
<h3>foo ###</h3>
<h2>foo ###</h2>
<h1>foo #</h1>
ATX headings need not be separated from surrounding content by blank lines, and they can interrupt paragraphs:
ATX headings can be empty:
A setext heading consists of one or more lines of text, each containing at least one non-whitespace character, with no more than 3 spaces indentation, followed by a setext heading underline. The lines of text must be such that, were they not followed by the setext heading underline, they would be interpreted as a paragraph: they cannot be interpretable as a code fence, ATX heading, block quote, thematic break, list item, or HTML block.
A setext heading underline is a sequence of
=
characters or a sequence of -
characters, with no more than 3
spaces indentation and any number of trailing spaces. If a line
containing a single -
can be interpreted as an
empty list items, it should be interpreted this way
and not as a setext heading underline.
The heading is a level 1 heading if =
characters are used in
the setext heading underline, and a level 2 heading if -
characters are used. The contents of the heading are the result
of parsing the preceding lines of text as CommonMark inline
content.
In general, a setext heading need not be preceded or followed by a blank line. However, it cannot interrupt a paragraph, so when a setext heading comes after a paragraph, a blank line is needed between them.
Simple examples:
Foo *bar*
=========
Foo *bar*
---------
<h1>Foo <em>bar</em></h1>
<h2>Foo <em>bar</em></h2>
The content of the header may span more than one line:
The underlining can be any length:
The heading content can be indented up to three spaces, and need not line up with the underlining:
Four spaces indent is too much:
The setext heading underline can be indented up to three spaces, and may have trailing spaces:
Four spaces is too much:
The setext heading underline cannot contain internal spaces:
Trailing spaces in the content line do not cause a line break:
Nor does a backslash at the end:
Since indicators of block structure take precedence over indicators of inline structure, the following are setext headings:
`Foo
----
`
<a title="a lot
---
of dashes"/>
<h2>`Foo</h2>
<p>`</p>
<h2><a title="a lot</h2>
<p>of dashes"/></p>
The setext heading underline cannot be a lazy continuation line in a list item or block quote:
A blank line is needed between a paragraph and a following setext heading, since otherwise the paragraph becomes part of the heading’s content:
But in general a blank line is not required before or after setext headings:
Setext headings cannot be empty:
Setext heading text lines must not be interpretable as block constructs other than paragraphs. So, the line of dashes in these examples gets interpreted as a thematic break:
If you want a heading with > foo
as its literal text, you can
use backslash escapes:
Compatibility note: Most existing Markdown implementations do not allow the text of setext headings to span multiple lines. But there is no consensus about how to interpret
Foo
bar
---
baz
One can find four different interpretations:
We find interpretation 4 most natural, and interpretation 4 increases the expressive power of CommonMark, by allowing multiline headings. Authors who want interpretation 1 can put a blank line after the first paragraph:
Authors who want interpretation 2 can put blank lines around the thematic break,
or use a thematic break that cannot count as a setext heading underline, such as
Authors who want interpretation 3 can use backslash escapes:
An indented code block is composed of one or more indented chunks separated by blank lines. An indented chunk is a sequence of non-blank lines, each indented four or more spaces. The contents of the code block are the literal contents of the lines, including trailing line endings, minus four spaces of indentation. An indented code block has no info string.
An indented code block cannot interrupt a paragraph, so there must be a blank line between a paragraph and a following indented code block. (A blank line is not needed, however, between a code block and a following paragraph.)
a simple
indented code block
<pre><code>a simple
indented code block
</code></pre>
If there is any ambiguity between an interpretation of indentation as a code block and as indicating that material belongs to a list item, the list item interpretation takes precedence:
The contents of a code block are literal text, and do not get parsed as Markdown:
Here we have three chunks separated by blank lines:
Any initial spaces beyond four will be included in the content, even in interior blank lines:
An indented code block cannot interrupt a paragraph. (This allows hanging indents and the like.)
However, any non-blank line with fewer than four leading spaces ends the code block immediately. So a paragraph may occur immediately after indented code:
And indented code can occur immediately before and after other kinds of blocks:
# Heading
foo
Heading
------
foo
----
<h1>Heading</h1>
<pre><code>foo
</code></pre>
<h2>Heading</h2>
<pre><code>foo
</code></pre>
<hr />
The first line can be indented more than four spaces:
Blank lines preceding or following an indented code block are not included in it:
Trailing spaces are included in the code block’s content:
A code fence is a sequence
of at least three consecutive backtick characters (`
) or
tildes (~
). (Tildes and backticks cannot be mixed.)
A fenced code block
begins with a code fence, indented no more than three spaces.
The line with the opening code fence may optionally contain some text following the code fence; this is trimmed of leading and trailing spaces and called the info string. The info string may not contain any backtick characters. (The reason for this restriction is that otherwise some inline code would be incorrectly interpreted as the beginning of a fenced code block.)
The content of the code block consists of all subsequent lines, until a closing code fence of the same type as the code block began with (backticks or tildes), and with at least as many backticks or tildes as the opening code fence. If the leading code fence is indented N spaces, then up to N spaces of indentation are removed from each line of the content (if present). (If a content line is not indented, it is preserved unchanged. If it is indented less than N spaces, all of the indentation is removed.)
The closing code fence may be indented up to three spaces, and may be followed only by spaces, which are ignored. If the end of the containing block (or document) is reached and no closing code fence has been found, the code block contains all of the lines after the opening code fence until the end of the containing block (or document). (An alternative spec would require backtracking in the event that a closing code fence is not found. But this makes parsing much less efficient, and there seems to be no real down side to the behavior described here.)
A fenced code block may interrupt a paragraph, and does not require a blank line either before or after.
The content of a code fence is treated as literal text, not parsed
as inlines. The first word of the info string is typically used to
specify the language of the code sample, and rendered in the class
attribute of the code
tag. However, this spec does not mandate any
particular treatment of the info string.
Here is a simple example with backticks:
With tildes:
The closing code fence must use the same character as the opening fence:
The closing code fence must be at least as long as the opening fence:
Unclosed code blocks are closed by the end of the document (or the enclosing block quote or list item):
> ```
> aaa
bbb
<blockquote>
<pre><code>aaa
</code></pre>
</blockquote>
<p>bbb</p>
A code block can have all empty lines as its content:
A code block can be empty:
Fences can be indented. If the opening fence is indented, content lines will have equivalent opening indentation removed, if present:
Four spaces indentation produces an indented code block:
Closing fences may be indented by 0-3 spaces, and their indentation need not match that of the opening fence:
This is not a closing fence, because it is indented 4 spaces:
Code fences (opening and closing) cannot contain internal spaces:
Fenced code blocks can interrupt paragraphs, and can be followed directly by paragraphs, without a blank line between:
Other blocks can also occur before and after fenced code blocks without an intervening blank line:
foo
---
~~~
bar
~~~
# baz
<h2>foo</h2>
<pre><code>bar
</code></pre>
<h1>baz</h1>
An info string can be provided after the opening code fence.
Opening and closing spaces will be stripped, and the first word, prefixed
with language-
, is used as the value for the class
attribute of the
code
element within the enclosing pre
element.
```ruby
def foo(x)
return 3
end
```
<pre><code class="language-ruby">def foo(x)
return 3
end
</code></pre>
~~~~ ruby startline=3 $%@#$
def foo(x)
return 3
end
~~~~~~~
<pre><code class="language-ruby">def foo(x)
return 3
end
</code></pre>
Info strings for backtick code blocks cannot contain backticks:
Closing code fences cannot have info strings:
An HTML block is a group of lines that is treated as raw HTML (and will not be escaped in HTML output).
There are seven kinds of HTML block, which can be defined by their start and end conditions. The block begins with a line that meets a start condition (after up to three spaces optional indentation). It ends with the first subsequent line that meets a matching end condition, or the last line of the document, if no line is encountered that meets the end condition. If the first line meets both the start condition and the end condition, the block will contain just that line.
Start condition: line begins with the string <script
,
<pre
, or <style
(case-insensitive), followed by whitespace,
the string >
, or the end of the line.
End condition: line contains an end tag
</script>
, </pre>
, or </style>
(case-insensitive; it
need not match the start tag).
Start condition: line begins with the string <!--
.
End condition: line contains the string -->
.
Start condition: line begins with the string <?
.
End condition: line contains the string ?>
.
Start condition: line begins with the string <!
followed by an uppercase ASCII letter.
End condition: line contains the character >
.
Start condition: line begins with the string
<![CDATA[
.
End condition: line contains the string ]]>
.
Start condition: line begins the string <
or </
followed by one of the strings (case-insensitive) address
,
article
, aside
, base
, basefont
, blockquote
, body
,
caption
, center
, col
, colgroup
, dd
, details
, dialog
,
dir
, div
, dl
, dt
, fieldset
, figcaption
, figure
,
footer
, form
, frame
, frameset
, h1
, head
, header
, hr
,
html
, iframe
, legend
, li
, link
, main
, menu
, menuitem
,
meta
, nav
, noframes
, ol
, optgroup
, option
, p
, param
,
section
, source
, summary
, table
, tbody
, td
,
tfoot
, th
, thead
, title
, tr
, track
, ul
, followed
by whitespace, the end of the line, the string >
, or
the string />
.
End condition: line is followed by a blank line.
Start condition: line begins with a complete open tag
or closing tag (with any tag name other than script
,
style
, or pre
) followed only by whitespace
or the end of the line.
End condition: line is followed by a blank line.
All types of HTML blocks except type 7 may interrupt a paragraph. Blocks of type 7 may not interrupt a paragraph. (This restriction is intended to prevent unwanted interpretation of long tags inside a wrapped paragraph as starting HTML blocks.)
Some simple examples follow. Here are some basic HTML blocks of type 6:
<table>
<tr>
<td>
hi
</td>
</tr>
</table>
okay.
<table>
<tr>
<td>
hi
</td>
</tr>
</table>
<p>okay.</p>
A block can also start with a closing tag:
Here we have two HTML blocks with a Markdown paragraph between them:
<DIV CLASS="foo">
*Markdown*
</DIV>
<DIV CLASS="foo">
<p><em>Markdown</em></p>
</DIV>
The tag on the first line can be partial, as long as it is split where there would be whitespace:
An open tag need not be closed:
A partial tag need not even be completed (garbage in, garbage out):
The initial tag doesn’t even need to be a valid tag, as long as it starts like one:
In type 6 blocks, the initial tag need not be on a line by itself:
Everything until the next blank line or end of document gets included in the HTML block. So, in the following example, what looks like a Markdown code block is actually part of the HTML block, which continues until a blank line or the end of the document is reached:
To start an HTML block with a tag that is not in the list of block-level tags in (6), you must put the tag by itself on the first line (and it must be complete):
In type 7 blocks, the tag name can be anything:
These rules are designed to allow us to work with tags that
can function as either block-level or inline-level tags.
The <del>
tag is a nice example. We can surround content with
<del>
tags in three different ways. In this case, we get a raw
HTML block, because the <del>
tag is on a line by itself:
In this case, we get a raw HTML block that just includes
the <del>
tag (because it ends with the following blank
line). So the contents get interpreted as CommonMark:
Finally, in this case, the <del>
tags are interpreted
as raw HTML inside the CommonMark paragraph. (Because
the tag is not on a line by itself, we get inline HTML
rather than an HTML block.)
HTML tags designed to contain literal content
(script
, style
, pre
), comments, processing instructions,
and declarations are treated somewhat differently.
Instead of ending at the first blank line, these blocks
end at the first line containing a corresponding end tag.
As a result, these blocks can contain blank lines:
A pre tag (type 1):
<pre language="haskell"><code>
import Text.HTML.TagSoup
main :: IO ()
main = print $ parseTags tags
</code></pre>
<pre language="haskell"><code>
import Text.HTML.TagSoup
main :: IO ()
main = print $ parseTags tags
</code></pre>
A script tag (type 1):
<script type="text/javascript">
// JavaScript example
document.getElementById("demo").innerHTML = "Hello JavaScript!";
</script>
<script type="text/javascript">
// JavaScript example
document.getElementById("demo").innerHTML = "Hello JavaScript!";
</script>
A style tag (type 1):
<style
type="text/css">
h1 {color:red;}
p {color:blue;}
</style>
<style
type="text/css">
h1 {color:red;}
p {color:blue;}
</style>
If there is no matching end tag, the block will end at the end of the document (or the enclosing block quote or list item):
The end tag can occur on the same line as the start tag:
<style>p{color:red;}</style>
*foo*
<style>p{color:red;}</style>
<p><em>foo</em></p>
Note that anything on the last line after the end tag will be included in the HTML block:
A comment (type 2):
A processing instruction (type 3):
A declaration (type 4):
CDATA (type 5):
<![CDATA[
function matchwo(a,b)
{
if (a < b && a < 0) then {
return 1;
} else {
return 0;
}
}
]]>
<![CDATA[
function matchwo(a,b)
{
if (a < b && a < 0) then {
return 1;
} else {
return 0;
}
}
]]>
The opening tag can be indented 1-3 spaces, but not 4:
<!-- foo -->
<!-- foo -->
<!-- foo -->
<pre><code><!-- foo -->
</code></pre>
An HTML block of types 1–6 can interrupt a paragraph, and need not be preceded by a blank line.
However, a following blank line is needed, except at the end of a document, and except for blocks of types 1–5, above:
HTML blocks of type 7 cannot interrupt a paragraph:
This rule differs from John Gruber’s original Markdown syntax specification, which says:
The only restrictions are that block-level HTML elements — e.g.
<div>
,<table>
,<pre>
,<p>
, etc. — must be separated from surrounding content by blank lines, and the start and end tags of the block should not be indented with tabs or spaces.
In some ways Gruber’s rule is more restrictive than the one given here:
Most Markdown implementations (including some of Gruber’s own) do not respect all of these restrictions.
There is one respect, however, in which Gruber’s rule is more liberal than the one given here, since it allows blank lines to occur inside an HTML block. There are two reasons for disallowing them here. First, it removes the need to parse balanced tags, which is expensive and can require backtracking from the end of the document if no matching end tag is found. Second, it provides a very simple and flexible way of including Markdown content inside HTML tags: simply separate the Markdown from the HTML using blank lines:
Compare:
Some Markdown implementations have adopted a convention of
interpreting content inside tags as text if the open tag has
the attribute markdown=1
. The rule given above seems a simpler and
more elegant way of achieving the same expressive power, which is also
much simpler to parse.
The main potential drawback is that one can no longer paste HTML blocks into Markdown documents with 100% reliability. However, in most cases this will work fine, because the blank lines in HTML are usually followed by HTML block tags. For example:
<table>
<tr>
<td>
Hi
</td>
</tr>
</table>
<table>
<tr>
<td>
Hi
</td>
</tr>
</table>
There are problems, however, if the inner tags are indented and separated by spaces, as then they will be interpreted as an indented code block:
<table>
<tr>
<td>
Hi
</td>
</tr>
</table>
<table>
<tr>
<pre><code><td>
Hi
</td>
</code></pre>
</tr>
</table>
Fortunately, blank lines are usually not necessary and can be
deleted. The exception is inside <pre>
tags, but as described
above, raw HTML blocks starting with <pre>
can contain blank
lines.
A link reference definition
consists of a link label, indented up to three spaces, followed
by a colon (:
), optional whitespace (including up to one
line ending), a link destination,
optional whitespace (including up to one
line ending), and an optional link
title, which if it is present must be separated
from the link destination by whitespace.
No further non-whitespace characters may occur on the line.
A link reference definition does not correspond to a structural element of a document. Instead, it defines a label which can be used in reference links and reference-style images elsewhere in the document. Link reference definitions can come either before or after the links that use them.
[Foo*bar\]]:my_(url) 'title (with parens)'
[Foo*bar\]]
<p><a href="my_(url)" title="title (with parens)">Foo*bar]</a></p>
[Foo bar]:
<my%20url>
'title'
[Foo bar]
<p><a href="my%20url" title="title">Foo bar</a></p>
The title may extend over multiple lines:
[foo]: /url '
title
line1
line2
'
[foo]
<p><a href="/url" title="
title
line1
line2
">foo</a></p>
However, it may not contain a blank line:
[foo]: /url 'title
with blank line'
[foo]
<p>[foo]: /url 'title</p>
<p>with blank line'</p>
<p>[foo]</p>
The title may be omitted:
The link destination may not be omitted:
Both title and destination can contain backslash escapes and literal backslashes:
[foo]: /url\bar\*baz "foo\"bar\baz"
[foo]
<p><a href="/url%5Cbar*baz" title="foo"bar\baz">foo</a></p>
A link can come before its corresponding definition:
If there are several matching definitions, the first one takes precedence:
As noted in the section on Links, matching of labels is case-insensitive (see matches).
Here is a link reference definition with no corresponding link. It contributes nothing to the document.
Here is another one:
This is not a link reference definition, because there are non-whitespace characters after the title:
This is a link reference definition, but it has no title:
This is not a link reference definition, because it is indented four spaces:
[foo]: /url "title"
[foo]
<pre><code>[foo]: /url "title"
</code></pre>
<p>[foo]</p>
This is not a link reference definition, because it occurs inside a code block:
A link reference definition cannot interrupt a paragraph.
However, it can directly follow other block elements, such as headings and thematic breaks, and it need not be followed by a blank line.
# [Foo]
[foo]: /url
> bar
<h1><a href="/url">Foo</a></h1>
<blockquote>
<p>bar</p>
</blockquote>
Several link reference definitions can occur one after another, without intervening blank lines.
[foo]: /foo-url "foo"
[bar]: /bar-url
"bar"
[baz]: /baz-url
[foo],
[bar],
[baz]
<p><a href="/foo-url" title="foo">foo</a>,
<a href="/bar-url" title="bar">bar</a>,
<a href="/baz-url">baz</a></p>
Link reference definitions can occur inside block containers, like lists and block quotations. They affect the entire document, not just the container in which they are defined:
A sequence of non-blank lines that cannot be interpreted as other kinds of blocks forms a paragraph. The contents of the paragraph are the result of parsing the paragraph’s raw content as inlines. The paragraph’s raw content is formed by concatenating the lines and removing initial and final whitespace.
A simple example with two paragraphs:
Paragraphs can contain multiple lines, but no blank lines:
Multiple blank lines between paragraph have no effect:
Leading spaces are skipped:
Lines after the first may be indented any amount, since indented code blocks cannot interrupt paragraphs.
However, the first line may be indented at most three spaces, or an indented code block will be triggered:
Final spaces are stripped before inline parsing, so a paragraph that ends with two or more spaces will not end with a hard line break:
Blank lines between block-level elements are ignored, except for the role they play in determining whether a list is tight or loose.
Blank lines at the beginning and end of the document are also ignored.
A container block is a block that has other blocks as its contents. There are two basic kinds of container blocks: block quotes and list items. Lists are meta-containers for list items.
We define the syntax for container blocks recursively. The general form of the definition is:
If X is a sequence of blocks, then the result of transforming X in such-and-such a way is a container of type Y with these blocks as its content.
So, we explain what counts as a block quote or list item by explaining how these can be generated from their contents. This should suffice to define the syntax, although it does not give a recipe for parsing these constructions. (A recipe is provided below in the section entitled A parsing strategy.)
A block quote marker
consists of 0-3 spaces of initial indent, plus (a) the character >
together
with a following space, or (b) a single character >
not followed by a space.
The following rules define block quotes:
Basic case. If a string of lines Ls constitute a sequence of blocks Bs, then the result of prepending a block quote marker to the beginning of each line in Ls is a block quote containing Bs.
Laziness. If a string of lines Ls constitute a block quote with contents Bs, then the result of deleting the initial block quote marker from one or more lines in which the next non-whitespace character after the block quote marker is paragraph continuation text is a block quote with Bs as its content. Paragraph continuation text is text that will be parsed as part of the content of a paragraph, but does not occur at the beginning of the paragraph.
Consecutiveness. A document cannot contain two block quotes in a row unless there is a blank line between them.
Nothing else counts as a block quote.
Here is a simple example:
The spaces after the >
characters can be omitted:
The >
characters can be indented 1-3 spaces:
Four spaces gives us a code block:
The Laziness clause allows us to omit the >
before a
paragraph continuation line:
A block quote can contain some lazy and some non-lazy continuation lines:
Laziness only applies to lines that would have been continuations of
paragraphs had they been prepended with block quote markers.
For example, the >
cannot be omitted in the second line of
> foo
> ---
without changing the meaning:
Similarly, if we omit the >
in the second line of
> - foo
> - bar
then the block quote ends after the first line:
> - foo
- bar
<blockquote>
<ul>
<li>foo</li>
</ul>
</blockquote>
<ul>
<li>bar</li>
</ul>
For the same reason, we can’t omit the >
in front of
subsequent lines of an indented or fenced code block:
> foo
bar
<blockquote>
<pre><code>foo
</code></pre>
</blockquote>
<pre><code>bar
</code></pre>
> ```
foo
```
<blockquote>
<pre><code></code></pre>
</blockquote>
<p>foo</p>
<pre><code></code></pre>
Note that in the following case, we have a paragraph continuation line:
To see why, note that in
> foo
> - bar
the - bar
is indented too far to start a list, and can’t
be an indented code block because indented code blocks cannot
interrupt paragraphs, so it is a [paragraph continuation line].
A block quote can be empty:
A block quote can have initial or final blank lines:
A blank line always separates block quotes:
> foo
> bar
<blockquote>
<p>foo</p>
</blockquote>
<blockquote>
<p>bar</p>
</blockquote>
(Most current Markdown implementations, including John Gruber’s
original Markdown.pl
, will parse this example as a single block quote
with two paragraphs. But it seems better to allow the author to decide
whether two block quotes or one are wanted.)
Consecutiveness means that if we put these block quotes together, we get a single block quote:
To get a block quote with two paragraphs, use:
Block quotes can interrupt paragraphs:
In general, blank lines are not needed before or after block quotes:
> aaa
***
> bbb
<blockquote>
<p>aaa</p>
</blockquote>
<hr />
<blockquote>
<p>bbb</p>
</blockquote>
However, because of laziness, a blank line is needed between a block quote and a following paragraph:
It is a consequence of the Laziness rule that any number
of initial >
s may be omitted on a continuation line of a
nested block quote:
> > > foo
bar
<blockquote>
<blockquote>
<blockquote>
<p>foo
bar</p>
</blockquote>
</blockquote>
</blockquote>
>>> foo
> bar
>>baz
<blockquote>
<blockquote>
<blockquote>
<p>foo
bar
baz</p>
</blockquote>
</blockquote>
</blockquote>
When including an indented code block in a block quote,
remember that the block quote marker includes
both the >
and a following space. So five spaces are needed after
the >
:
> code
> not code
<blockquote>
<pre><code>code
</code></pre>
</blockquote>
<blockquote>
<p>not code</p>
</blockquote>
A list marker is a bullet list marker or an ordered list marker.
A bullet list marker
is a -
, +
, or *
character.
An ordered list marker
is a sequence of 1–9 arabic digits (0-9
), followed by either a
.
character or a )
character. (The reason for the length
limit is that with 10 digits we start seeing integer overflows
in some browsers.)
The following rules define list items:
For example, let Ls be the lines
A paragraph
with two lines.
indented code
> A block quote.
<p>A paragraph
with two lines.</p>
<pre><code>indented code
</code></pre>
<blockquote>
<p>A block quote.</p>
</blockquote>
And let M be the marker 1.
, and N = 2. Then rule #1 says
that the following is an ordered list item with start number 1,
and the same contents as Ls:
1. A paragraph
with two lines.
indented code
> A block quote.
<ol>
<li>
<p>A paragraph
with two lines.</p>
<pre><code>indented code
</code></pre>
<blockquote>
<p>A block quote.</p>
</blockquote>
</li>
</ol>
The most important thing to notice is that the position of the text after the list marker determines how much indentation is needed in subsequent blocks in the list item. If the list marker takes up two spaces, and there are three spaces between the list marker and the next non-whitespace character, then blocks must be indented five spaces in order to fall under the list item.
Here are some examples showing how far content must be indented to be put under the list item:
It is tempting to think of this in terms of columns: the continuation blocks must be indented at least to the column of the first non-whitespace character after the list marker. However, that is not quite right. The spaces after the list marker determine how much relative indentation is needed. Which column this indentation reaches will depend on how the list item is embedded in other constructions, as shown by this example:
> > 1. one
>>
>> two
<blockquote>
<blockquote>
<ol>
<li>
<p>one</p>
<p>two</p>
</li>
</ol>
</blockquote>
</blockquote>
Here two
occurs in the same column as the list marker 1.
,
but is actually contained in the list item, because there is
sufficient indentation after the last containing blockquote marker.
The converse is also possible. In the following example, the word two
occurs far to the right of the initial text of the list item, one
, but
it is not considered part of the list item, because it is not indented
far enough past the blockquote marker:
>>- one
>>
> > two
<blockquote>
<blockquote>
<ul>
<li>one</li>
</ul>
<p>two</p>
</blockquote>
</blockquote>
Note that at least one space is needed between the list marker and any following content, so these are not list items:
A list item may not contain blocks that are separated by more than one blank line. Thus, two blank lines will end a list, unless the two blanks are contained in a fenced code block.
- foo
bar
- foo
bar
- ```
foo
bar
```
- baz
+ ```
foo
bar
```
<ul>
<li>
<p>foo</p>
<p>bar</p>
</li>
<li>
<p>foo</p>
</li>
</ul>
<p>bar</p>
<ul>
<li>
<pre><code>foo
bar
</code></pre>
</li>
<li>
<p>baz</p>
<ul>
<li>
<pre><code>foo
bar
</code></pre>
</li>
</ul>
</li>
</ul>
A list item may contain any kind of block:
1. foo
```
bar
```
baz
> bam
<ol>
<li>
<p>foo</p>
<pre><code>bar
</code></pre>
<p>baz</p>
<blockquote>
<p>bam</p>
</blockquote>
</li>
</ol>
A list item that contains an indented code block will preserve empty lines within the code block verbatim, unless there are two or more empty lines in a row (since as described above, two blank lines end the list):
- Foo
bar
baz
<ul>
<li>
<p>Foo</p>
<pre><code>bar
baz
</code></pre>
</li>
</ul>
- Foo
bar
baz
<ul>
<li>
<p>Foo</p>
<pre><code>bar
</code></pre>
</li>
</ul>
<pre><code> baz
</code></pre>
Note that ordered list start numbers must be nine digits or less:
A start number may begin with 0s:
A start number may not be negative:
An indented code block will have to be indented four spaces beyond the edge of the region where text will be included in the list item. In the following case that is 6 spaces:
And in this case it is 11 spaces:
10. foo
bar
<ol start="10">
<li>
<p>foo</p>
<pre><code>bar
</code></pre>
</li>
</ol>
If the first block in the list item is an indented code block, then by rule #2, the contents must be indented one space after the list marker:
indented code
paragraph
more code
<pre><code>indented code
</code></pre>
<p>paragraph</p>
<pre><code>more code
</code></pre>
1. indented code
paragraph
more code
<ol>
<li>
<pre><code>indented code
</code></pre>
<p>paragraph</p>
<pre><code>more code
</code></pre>
</li>
</ol>
Note that an additional space indent is interpreted as space inside the code block:
1. indented code
paragraph
more code
<ol>
<li>
<pre><code> indented code
</code></pre>
<p>paragraph</p>
<pre><code>more code
</code></pre>
</li>
</ol>
Note that rules #1 and #2 only apply to two cases: (a) cases in which the lines to be included in a list item begin with a non-whitespace character, and (b) cases in which they begin with an indented code block. In a case like the following, where the first block begins with a three-space indent, the rules do not allow us to form a list item by indenting the whole thing and prepending a list marker:
This is not a significant restriction, because when a block begins with 1-3 spaces indent, the indentation can always be removed without a change in interpretation, allowing rule #1 to be applied. So, in the above case:
Here are some list items that start with a blank line but are not empty:
-
foo
-
```
bar
```
-
baz
<ul>
<li>foo</li>
<li>
<pre><code>bar
</code></pre>
</li>
<li>
<pre><code>baz
</code></pre>
</li>
</ul>
A list item can begin with at most one blank line.
In the following example, foo
is not part of the list
item:
Here is an empty bullet list item:
It does not matter whether there are spaces following the list marker:
Here is an empty ordered list item:
A list may start or end with an empty list item:
Indented one space:
1. A paragraph
with two lines.
indented code
> A block quote.
<ol>
<li>
<p>A paragraph
with two lines.</p>
<pre><code>indented code
</code></pre>
<blockquote>
<p>A block quote.</p>
</blockquote>
</li>
</ol>
Indented two spaces:
1. A paragraph
with two lines.
indented code
> A block quote.
<ol>
<li>
<p>A paragraph
with two lines.</p>
<pre><code>indented code
</code></pre>
<blockquote>
<p>A block quote.</p>
</blockquote>
</li>
</ol>
Indented three spaces:
1. A paragraph
with two lines.
indented code
> A block quote.
<ol>
<li>
<p>A paragraph
with two lines.</p>
<pre><code>indented code
</code></pre>
<blockquote>
<p>A block quote.</p>
</blockquote>
</li>
</ol>
Four spaces indent gives a code block:
1. A paragraph
with two lines.
indented code
> A block quote.
<pre><code>1. A paragraph
with two lines.
indented code
> A block quote.
</code></pre>
Here is an example with lazy continuation lines:
1. A paragraph
with two lines.
indented code
> A block quote.
<ol>
<li>
<p>A paragraph
with two lines.</p>
<pre><code>indented code
</code></pre>
<blockquote>
<p>A block quote.</p>
</blockquote>
</li>
</ol>
Indentation can be partially deleted:
1. A paragraph
with two lines.
<ol>
<li>A paragraph
with two lines.</li>
</ol>
These examples show how laziness can work in nested structures:
> 1. > Blockquote
continued here.
<blockquote>
<ol>
<li>
<blockquote>
<p>Blockquote
continued here.</p>
</blockquote>
</li>
</ol>
</blockquote>
> 1. > Blockquote
> continued here.
<blockquote>
<ol>
<li>
<blockquote>
<p>Blockquote
continued here.</p>
</blockquote>
</li>
</ol>
</blockquote>
The rules for sublists follow from the general rules above. A sublist must be indented the same number of spaces a paragraph would need to be in order to be included in the list item.
So, in this case we need two spaces indent:
- foo
- bar
- baz
<ul>
<li>foo
<ul>
<li>bar
<ul>
<li>baz</li>
</ul>
</li>
</ul>
</li>
</ul>
One is not enough:
Here we need four, because the list marker is wider:
Three is not enough:
A list may be the first block in a list item:
1. - 2. foo
<ol>
<li>
<ul>
<li>
<ol start="2">
<li>foo</li>
</ol>
</li>
</ul>
</li>
</ol>
A list item can contain a heading:
- # Foo
- Bar
---
baz
<ul>
<li>
<h1>Foo</h1>
</li>
<li>
<h2>Bar</h2>
baz</li>
</ul>
John Gruber’s Markdown spec says the following about list items:
“List markers typically start at the left margin, but may be indented by up to three spaces. List markers must be followed by one or more spaces or a tab.”
“To make lists look nice, you can wrap items with hanging indents…. But if you don’t want to, you don’t have to.”
“List items may consist of multiple paragraphs. Each subsequent paragraph in a list item must be indented by either 4 spaces or one tab.”
“It looks nice if you indent every line of the subsequent paragraphs, but here again, Markdown will allow you to be lazy.”
“To put a blockquote within a list item, the blockquote’s >
delimiters need to be indented.”
“To put a code block within a list item, the code block needs to be indented twice — 8 spaces or two tabs.”
These rules specify that a paragraph under a list item must be indented four spaces (presumably, from the left margin, rather than the start of the list marker, but this is not said), and that code under a list item must be indented eight spaces instead of the usual four. They also say that a block quote must be indented, but not by how much; however, the example given has four spaces indentation. Although nothing is said about other kinds of block-level content, it is certainly reasonable to infer that all block elements under a list item, including other lists, must be indented four spaces. This principle has been called the four-space rule.
The four-space rule is clear and principled, and if the reference
implementation Markdown.pl
had followed it, it probably would have
become the standard. However, Markdown.pl
allowed paragraphs and
sublists to start with only two spaces indentation, at least on the
outer level. Worse, its behavior was inconsistent: a sublist of an
outer-level list needed two spaces indentation, but a sublist of this
sublist needed three spaces. It is not surprising, then, that different
implementations of Markdown have developed very different rules for
determining what comes under a list item. (Pandoc and python-Markdown,
for example, stuck with Gruber’s syntax description and the four-space
rule, while discount, redcarpet, marked, PHP Markdown, and others
followed Markdown.pl
’s behavior more closely.)
Unfortunately, given the divergences between implementations, there
is no way to give a spec for list items that will be guaranteed not
to break any existing documents. However, the spec given here should
correctly handle lists formatted with either the four-space rule or
the more forgiving Markdown.pl
behavior, provided they are laid out
in a way that is natural for a human to read.
The strategy here is to let the width and indentation of the list marker determine the indentation necessary for blocks to fall under the list item, rather than having a fixed and arbitrary number. The writer can think of the body of the list item as a unit which gets indented to the right enough to fit the list marker (and any indentation on the list marker). (The laziness rule, #5, then allows continuation lines to be unindented if needed.)
This rule is superior, we claim, to any rule requiring a fixed level of indentation from the margin. The four-space rule is clear but unnatural. It is quite unintuitive that
- foo
bar
- baz
should be parsed as two lists with an intervening paragraph,
<ul>
<li>foo</li>
</ul>
<p>bar</p>
<ul>
<li>baz</li>
</ul>
as the four-space rule demands, rather than a single list,
<ul>
<li>
<p>foo</p>
<p>bar</p>
<ul>
<li>baz</li>
</ul>
</li>
</ul>
The choice of four spaces is arbitrary. It can be learned, but it is not likely to be guessed, and it trips up beginners regularly.
Would it help to adopt a two-space rule? The problem is that such
a rule, together with the rule allowing 1–3 spaces indentation of the
initial list marker, allows text that is indented less than the
original list marker to be included in the list item. For example,
Markdown.pl
parses
- one
two
as a single list item, with two
a continuation paragraph:
<ul>
<li>
<p>one</p>
<p>two</p>
</li>
</ul>
and similarly
> - one
>
> two
as
<blockquote>
<ul>
<li>
<p>one</p>
<p>two</p>
</li>
</ul>
</blockquote>
This is extremely unintuitive.
Rather than requiring a fixed indent from the margin, we could require
a fixed indent (say, two spaces, or even one space) from the list marker (which
may itself be indented). This proposal would remove the last anomaly
discussed. Unlike the spec presented above, it would count the following
as a list item with a subparagraph, even though the paragraph bar
is not indented as far as the first paragraph foo
:
10. foo
bar
Arguably this text does read like a list item with bar
as a subparagraph,
which may count in favor of the proposal. However, on this proposal indented
code would have to be indented six spaces after the list marker. And this
would break a lot of existing Markdown, which has the pattern:
1. foo
indented code
where the code is indented eight spaces. The spec above, by contrast, will
parse this text as expected, since the code block’s indentation is measured
from the beginning of foo
.
The one case that needs special treatment is a list item that starts with indented code. How much indentation is required in that case, since we don’t have a “first paragraph” to measure from? Rule #2 simply stipulates that in such cases, we require one space indentation from the list marker (and then the normal four spaces for the indented code). This will match the four-space rule in cases where the list marker plus its initial indentation takes four spaces (a common case), but diverge in other cases.
A list is a sequence of one or more list items of the same type. The list items may be separated by single blank lines, but two blank lines end all containing lists.
Two list items are of the same type
if they begin with a list marker of the same type.
Two list markers are of the
same type if (a) they are bullet list markers using the same character
(-
, +
, or *
) or (b) they are ordered list numbers with the same
delimiter (either .
or )
).
A list is an ordered list if its constituent list items begin with ordered list markers, and a bullet list if its constituent list items begin with bullet list markers.
The start number of an ordered list is determined by the list number of its initial list item. The numbers of subsequent list items are disregarded.
A list is loose if any of its constituent
list items are separated by blank lines, or if any of its constituent
list items directly contain two block-level elements with a blank line
between them. Otherwise a list is tight.
(The difference in HTML output is that paragraphs in a loose list are
wrapped in <p>
tags, while paragraphs in a tight list are not.)
Changing the bullet or ordered list delimiter starts a new list:
- foo
- bar
+ baz
<ul>
<li>foo</li>
<li>bar</li>
</ul>
<ul>
<li>baz</li>
</ul>
1. foo
2. bar
3) baz
<ol>
<li>foo</li>
<li>bar</li>
</ol>
<ol start="3">
<li>baz</li>
</ol>
In CommonMark, a list can interrupt a paragraph. That is, no blank line is needed to separate a paragraph from a following list:
Markdown.pl
does not allow this, through fear of triggering a list
via a numeral in a hard-wrapped line:
The number of windows in my house is
14. The number of doors is 6.
<p>The number of windows in my house is</p>
<ol start="14">
<li>The number of doors is 6.</li>
</ol>
Oddly, Markdown.pl
does allow a blockquote to interrupt a paragraph,
even though the same considerations might apply. We think that the two
cases should be treated the same. Here are two reasons for allowing
lists to interrupt paragraphs:
First, it is natural and not uncommon for people to start lists without blank lines:
I need to buy
- new shoes
- a coat
- a plane ticket
Second, we are attracted to a
principle of uniformity: if a chunk of text has a certain meaning, it will continue to have the same meaning when put into a container block (such as a list item or blockquote).
(Indeed, the spec for list items and block quotes presupposes this principle.) This principle implies that if
* I need to buy
- new shoes
- a coat
- a plane ticket
is a list item containing a paragraph followed by a nested sublist,
as all Markdown implementations agree it is (though the paragraph
may be rendered without <p>
tags, since the list is “tight”),
then
I need to buy
- new shoes
- a coat
- a plane ticket
by itself should be a paragraph followed by a nested sublist.
Our adherence to the principle of uniformity thus inclines us to think that there are two coherent packages:
Require blank lines before all lists and blockquotes, including lists that occur as sublists inside other list items.
Require blank lines in none of these places.
reStructuredText takes the first approach, for which there is much to be said. But the second seems more consistent with established practice with Markdown.
There can be blank lines between items, but two blank lines end a list:
- foo
- bar
- baz
<ul>
<li>
<p>foo</p>
</li>
<li>
<p>bar</p>
</li>
</ul>
<ul>
<li>baz</li>
</ul>
As illustrated above in the section on list items, two blank lines between blocks within a list item will also end a list:
Indeed, two blank lines will end all containing lists:
- foo
- bar
- baz
bim
<ul>
<li>foo
<ul>
<li>bar
<ul>
<li>baz</li>
</ul>
</li>
</ul>
</li>
</ul>
<pre><code> bim
</code></pre>
Thus, two blank lines can be used to separate consecutive lists of the same type, or to separate a list from an indented code block that would otherwise be parsed as a subparagraph of the final list item:
- foo
- bar
- baz
- bim
<ul>
<li>foo</li>
<li>bar</li>
</ul>
<ul>
<li>baz</li>
<li>bim</li>
</ul>
- foo
notcode
- foo
code
<ul>
<li>
<p>foo</p>
<p>notcode</p>
</li>
<li>
<p>foo</p>
</li>
</ul>
<pre><code>code
</code></pre>
List items need not be indented to the same level. The following list items will be treated as items at the same list level, since none is indented enough to belong to the previous list item:
- a
- b
- c
- d
- e
- f
- g
- h
- i
<ul>
<li>a</li>
<li>b</li>
<li>c</li>
<li>d</li>
<li>e</li>
<li>f</li>
<li>g</li>
<li>h</li>
<li>i</li>
</ul>
1. a
2. b
3. c
<ol>
<li>
<p>a</p>
</li>
<li>
<p>b</p>
</li>
<li>
<p>c</p>
</li>
</ol>
This is a loose list, because there is a blank line between two of the list items:
- a
- b
- c
<ul>
<li>
<p>a</p>
</li>
<li>
<p>b</p>
</li>
<li>
<p>c</p>
</li>
</ul>
So is this, with a empty second item:
These are loose lists, even though there is no space between the items, because one of the items directly contains two block-level elements with a blank line between them:
- a
- b
c
- d
<ul>
<li>
<p>a</p>
</li>
<li>
<p>b</p>
<p>c</p>
</li>
<li>
<p>d</p>
</li>
</ul>
- a
- b
[ref]: /url
- d
<ul>
<li>
<p>a</p>
</li>
<li>
<p>b</p>
</li>
<li>
<p>d</p>
</li>
</ul>
This is a tight list, because the blank lines are in a code block:
- a
- ```
b
```
- c
<ul>
<li>a</li>
<li>
<pre><code>b
</code></pre>
</li>
<li>c</li>
</ul>
This is a tight list, because the blank line is between two paragraphs of a sublist. So the sublist is loose while the outer list is tight:
- a
- b
c
- d
<ul>
<li>a
<ul>
<li>
<p>b</p>
<p>c</p>
</li>
</ul>
</li>
<li>d</li>
</ul>
This is a tight list, because the blank line is inside the block quote:
* a
> b
>
* c
<ul>
<li>a
<blockquote>
<p>b</p>
</blockquote>
</li>
<li>c</li>
</ul>
This list is tight, because the consecutive block elements are not separated by blank lines:
- a
> b
```
c
```
- d
<ul>
<li>a
<blockquote>
<p>b</p>
</blockquote>
<pre><code>c
</code></pre>
</li>
<li>d</li>
</ul>
A single-paragraph list is tight:
This list is loose, because of the blank line between the two block elements in the list item:
1. ```
foo
```
bar
<ol>
<li>
<pre><code>foo
</code></pre>
<p>bar</p>
</li>
</ol>
Here the outer list is loose, the inner list tight:
* foo
* bar
baz
<ul>
<li>
<p>foo</p>
<ul>
<li>bar</li>
</ul>
<p>baz</p>
</li>
</ul>
- a
- b
- c
- d
- e
- f
<ul>
<li>
<p>a</p>
<ul>
<li>b</li>
<li>c</li>
</ul>
</li>
<li>
<p>d</p>
<ul>
<li>e</li>
<li>f</li>
</ul>
</li>
</ul>
Inlines are parsed sequentially from the beginning of the character stream to the end (left to right, in left-to-right languages). Thus, for example, in
hi
is parsed as code, leaving the backtick at the end as a literal
backtick.
Any ASCII punctuation character may be backslash-escaped:
\!\"\#\$\%\&\'\(\)\*\+\,\-\.\/\:\;\<\=\>\?\@\[\\\]\^\_\`\{\|\}\~
<p>!"#$%&'()*+,-./:;<=>?@[\]^_`{|}~</p>
Backslashes before other characters are treated as literal backslashes:
Escaped characters are treated as regular characters and do not have their usual Markdown meanings:
\*not emphasized*
\<br/> not a tag
\[not a link](/foo)
\`not code`
1\. not a list
\* not a list
\# not a heading
\[foo]: /url "not a reference"
<p>*not emphasized*
<br/> not a tag
[not a link](/foo)
`not code`
1. not a list
* not a list
# not a heading
[foo]: /url "not a reference"</p>
If a backslash is itself escaped, the following character is not:
A backslash at the end of the line is a hard line break:
Backslash escapes do not work in code blocks, code spans, autolinks, or raw HTML:
<http://example.com?find=\*>
<p><a href="http://example.com?find=%5C*">http://example.com?find=\*</a></p>
But they work in all other contexts, including URLs and link titles, link references, and info strings in fenced code blocks:
All valid HTML entity references and numeric character references, except those occuring in code blocks and code spans, are recognized as such and treated as equivalent to the corresponding Unicode characters. Conforming CommonMark parsers need not store information about whether a particular character was represented in the source using a Unicode character or an entity reference.
Entity references consist of &
+ any of the valid
HTML5 entity names + ;
. The
document https://html.spec.whatwg.org/multipage/entities.json
is used as an authoritative source for the valid entity
references and their corresponding code points.
& © Æ Ď
¾ ℋ ⅆ
∲ ≧̸
<p> & © Æ Ď
¾ ℋ ⅆ
∲ ≧̸</p>
Decimal numeric character
consist of &#
+ a string of 1–8 arabic digits + ;
. A
numeric character reference is parsed as the corresponding
Unicode character. Invalid Unicode code points will be replaced by
the REPLACEMENT CHARACTER (U+FFFD
). For security reasons,
the code point U+0000
will also be replaced by U+FFFD
.
Hexadecimal numeric character consist of &#
+
either X
or x
+ a string of 1-8 hexadecimal digits + ;
.
They too are parsed as the corresponding Unicode character (this
time specified with a hexadecimal numeral instead of decimal).
Here are some nonentities:
  &x; &#; &#x;
&ThisIsNotDefined; &hi?;
<p>&nbsp &x; &#; &#x;
&ThisIsNotDefined; &hi?;</p>
Although HTML5 does accept some entity references
without a trailing semicolon (such as ©
), these are not
recognized here, because it makes the grammar too ambiguous:
Strings that are not on the list of HTML5 named entities are not recognized as entity references either:
Entity and numeric character references are recognized in any context besides code spans or code blocks, including URLs, link titles, and fenced code block info strings:
[foo](/föö "föö")
<p><a href="/f%C3%B6%C3%B6" title="föö">foo</a></p>
[foo]
[foo]: /föö "föö"
<p><a href="/f%C3%B6%C3%B6" title="föö">foo</a></p>
Entity and numeric character references are treated as literal text in code spans and code blocks:
A backtick string
is a string of one or more backtick characters (`
) that is neither
preceded nor followed by a backtick.
A code span begins with a backtick string and ends with a backtick string of equal length. The contents of the code span are the characters between the two backtick strings, with leading and trailing spaces and line endings removed, and whitespace collapsed to single spaces.
This is a simple code span:
Here two backticks are used, because the code contains a backtick. This example also illustrates stripping of leading and trailing spaces:
This example shows the motivation for stripping leading and trailing spaces:
Line endings are treated like spaces:
Interior spaces and line endings are collapsed into single spaces, just as they would be by a browser:
Q: Why not just leave the spaces, since browsers will collapse them anyway? A: Because we might be targeting a non-HTML format, and we shouldn’t rely on HTML-specific rendering assumptions.
(Existing implementations differ in their treatment of internal
spaces and line endings. Some, including Markdown.pl
and
showdown
, convert an internal line ending into a
<br />
tag. But this makes things difficult for those who like to
hard-wrap their paragraphs, since a line break in the midst of a code
span will cause an unintended line break in the output. Others just
leave internal spaces as they are, which is fine if only HTML is being
targeted.)
Note that backslash escapes do not work in code spans. All backslashes are treated literally:
Backslash escapes are never needed, because one can always choose a string of n backtick characters as delimiters, where the code does not contain any strings of exactly n backtick characters.
Code span backticks have higher precedence than any other inline
constructs except HTML tags and autolinks. Thus, for example, this is
not parsed as emphasized text, since the second *
is part of a code
span:
And this is not parsed as a link:
Code spans, HTML tags, and autolinks have the same precedence. Thus, this is code:
But this is an HTML tag:
And this is code:
But this is an autolink:
<http://foo.bar.`baz>`
<p><a href="http://foo.bar.%60baz">http://foo.bar.`baz</a>`</p>
When a backtick string is not closed by a matching backtick string, we just have literal backticks:
John Gruber’s original Markdown syntax description says:
Markdown treats asterisks (
*
) and underscores (_
) as indicators of emphasis. Text wrapped with one*
or_
will be wrapped with an HTML<em>
tag; double*
’s or_
’s will be wrapped with an HTML<strong>
tag.
This is enough for most users, but these rules leave much undecided,
especially when it comes to nested emphasis. The original
Markdown.pl
test suite makes it clear that triple ***
and
___
delimiters can be used for strong emphasis, and most
implementations have also allowed the following patterns:
***strong emph***
***strong** in emph*
***emph* in strong**
**in strong *emph***
*in emph **strong***
The following patterns are less widely supported, but the intent is clear and they are useful (especially in contexts like bibliography entries):
*emph *with emph* in it*
**strong **with strong** in it**
Many implementations have also restricted intraword emphasis to
the *
forms, to avoid unwanted emphasis in words containing
internal underscores. (It is best practice to put these in code
spans, but users often do not.)
internal emphasis: foo*bar*baz
no emphasis: foo_bar_baz
The rules given below capture all of these patterns, while allowing for efficient parsing strategies that do not backtrack.
First, some definitions. A delimiter run is either
a sequence of one or more *
characters that is not preceded or
followed by a *
character, or a sequence of one or more _
characters that is not preceded or followed by a _
character.
A left-flanking delimiter run is a delimiter run that is (a) not followed by Unicode whitespace, and (b) either not followed by a punctuation character, or preceded by Unicode whitespace or a punctuation character. For purposes of this definition, the beginning and the end of the line count as Unicode whitespace.
A right-flanking delimiter run is a delimiter run that is (a) not preceded by Unicode whitespace, and (b) either not preceded by a punctuation character, or followed by Unicode whitespace or a punctuation character. For purposes of this definition, the beginning and the end of the line count as Unicode whitespace.
Here are some examples of delimiter runs.
left-flanking but not right-flanking:
***abc
_abc
**"abc"
_"abc"
right-flanking but not left-flanking:
abc***
abc_
"abc"**
"abc"_
Both left and right-flanking:
abc***def
"abc"_"def"
Neither left nor right-flanking:
abc *** def
a _ b
(The idea of distinguishing left-flanking and right-flanking delimiter runs based on the character before and the character after comes from Roopesh Chander’s vfmd. vfmd uses the terminology “emphasis indicator string” instead of “delimiter run,” and its rules for distinguishing left- and right-flanking runs are a bit more complex than the ones given here.)
The following rules define emphasis and strong emphasis:
A single *
character can open emphasis
iff (if and only if) it is part of a left-flanking delimiter run.
A single _
character can open emphasis iff
it is part of a left-flanking delimiter run
and either (a) not part of a right-flanking delimiter run
or (b) part of a right-flanking delimiter run
preceded by punctuation.
A single *
character can close emphasis
iff it is part of a right-flanking delimiter run.
A single _
character can close emphasis iff
it is part of a right-flanking delimiter run
and either (a) not part of a left-flanking delimiter run
or (b) part of a left-flanking delimiter run
followed by punctuation.
A double **
can open strong emphasis
iff it is part of a left-flanking delimiter run.
A double __
can open strong emphasis iff
it is part of a left-flanking delimiter run
and either (a) not part of a right-flanking delimiter run
or (b) part of a right-flanking delimiter run
preceded by punctuation.
A double **
can close strong emphasis
iff it is part of a right-flanking delimiter run.
A double __
can close strong emphasis
it is part of a right-flanking delimiter run
and either (a) not part of a left-flanking delimiter run
or (b) part of a left-flanking delimiter run
followed by punctuation.
Emphasis begins with a delimiter that can open emphasis and ends
with a delimiter that can close emphasis, and that uses the same
character (_
or *
) as the opening delimiter. There must
be a nonempty sequence of inlines between the open delimiter
and the closing delimiter; these form the contents of the emphasis
inline.
Strong emphasis begins with a delimiter that
can open strong emphasis and ends with a delimiter that
can close strong emphasis, and that uses the same character
(_
or *
) as the opening delimiter.
There must be a nonempty sequence of inlines between the open
delimiter and the closing delimiter; these form the contents of
the strong emphasis inline.
A literal *
character cannot occur at the beginning or end of
*
-delimited emphasis or **
-delimited strong emphasis, unless it
is backslash-escaped.
A literal _
character cannot occur at the beginning or end of
_
-delimited emphasis or __
-delimited strong emphasis, unless it
is backslash-escaped.
Where rules 1–12 above are compatible with multiple parsings, the following principles resolve ambiguity:
The number of nestings should be minimized. Thus, for example,
an interpretation <strong>...</strong>
is always preferred to
<em><em>...</em></em>
.
An interpretation <strong><em>...</em></strong>
is always
preferred to <em><strong>..</strong></em>
.
When two potential emphasis or strong emphasis spans overlap,
so that the second begins before the first ends and ends after
the first ends, the first takes precedence. Thus, for example,
*foo _bar* baz_
is parsed as <em>foo _bar</em> baz_
rather
than *foo <em>bar* baz</em>
. For the same reason,
**foo*bar**
is parsed as <em><em>foo</em>bar</em>*
rather than <strong>foo*bar</strong>
.
When there are two potential emphasis or strong emphasis spans
with the same closing delimiter, the shorter one (the one that
opens later) takes precedence. Thus, for example,
**foo **bar baz**
is parsed as **foo <strong>bar baz</strong>
rather than <strong>foo **bar baz</strong>
.
Inline code spans, links, images, and HTML tags group more tightly
than emphasis. So, when there is a choice between an interpretation
that contains one of these elements and one that does not, the
former always wins. Thus, for example, *[foo*](bar)
is
parsed as *<a href="bar">foo*</a>
rather than as
<em>[foo</em>](bar)
.
These rules can be illustrated through a series of examples.
Rule 1:
This is not emphasis, because the opening *
is followed by
whitespace, and hence not part of a left-flanking delimiter run:
This is not emphasis, because the opening *
is preceded
by an alphanumeric and followed by punctuation, and hence
not part of a left-flanking delimiter run:
Unicode nonbreaking spaces count as whitespace, too:
Intraword emphasis with *
is permitted:
Rule 2:
This is not emphasis, because the opening _
is followed by
whitespace:
This is not emphasis, because the opening _
is preceded
by an alphanumeric and followed by punctuation:
Emphasis with _
is not allowed inside words:
Here _
does not generate emphasis, because the first delimiter run
is right-flanking and the second left-flanking:
This is emphasis, even though the opening delimiter is both left- and right-flanking, because it is preceded by punctuation:
Rule 3:
This is not emphasis, because the closing delimiter does not match the opening delimiter:
This is not emphasis, because the closing *
is preceded by
whitespace:
A newline also counts as whitespace:
This is not emphasis, because the second *
is
preceded by punctuation and followed by an alphanumeric
(hence it is not part of a right-flanking delimiter run:
The point of this restriction is more easily appreciated with this example:
Intraword emphasis with *
is allowed:
Rule 4:
This is not emphasis, because the closing _
is preceded by
whitespace:
This is not emphasis, because the second _
is
preceded by punctuation and followed by an alphanumeric:
This is emphasis within emphasis:
Intraword emphasis is disallowed for _
:
This is emphasis, even though the closing delimiter is both left- and right-flanking, because it is followed by punctuation:
Rule 5:
This is not strong emphasis, because the opening delimiter is followed by whitespace:
This is not strong emphasis, because the opening **
is preceded
by an alphanumeric and followed by punctuation, and hence
not part of a left-flanking delimiter run:
Intraword strong emphasis with **
is permitted:
Rule 6:
This is not strong emphasis, because the opening delimiter is followed by whitespace:
A newline counts as whitespace:
This is not strong emphasis, because the opening __
is preceded
by an alphanumeric and followed by punctuation:
Intraword strong emphasis is forbidden with __
:
This is strong emphasis, even though the opening delimiter is both left- and right-flanking, because it is preceded by punctuation:
Rule 7:
This is not strong emphasis, because the closing delimiter is preceded by whitespace:
(Nor can it be interpreted as an emphasized *foo bar *
, because of
Rule 11.)
This is not strong emphasis, because the second **
is
preceded by punctuation and followed by an alphanumeric:
The point of this restriction is more easily appreciated with these examples:
**Gomphocarpus (*Gomphocarpus physocarpus*, syn.
*Asclepias physocarpa*)**
<p><strong>Gomphocarpus (<em>Gomphocarpus physocarpus</em>, syn.
<em>Asclepias physocarpa</em>)</strong></p>
Intraword emphasis:
Rule 8:
This is not strong emphasis, because the closing delimiter is preceded by whitespace:
This is not strong emphasis, because the second __
is
preceded by punctuation and followed by an alphanumeric:
The point of this restriction is more easily appreciated with this example:
Intraword strong emphasis is forbidden with __
:
This is strong emphasis, even though the closing delimiter is both left- and right-flanking, because it is followed by punctuation:
Rule 9:
Any nonempty sequence of inline elements can be the contents of an emphasized span.
In particular, emphasis and strong emphasis can be nested inside emphasis:
But note:
The difference is that in the preceding case, the internal delimiters can close emphasis, while in the cases with spaces, they cannot.
Note, however, that in the following case we get no strong
emphasis, because the opening delimiter is closed by the first
*
before bar
:
Indefinite levels of nesting are possible:
*foo **bar *baz* bim** bop*
<p><em>foo <strong>bar <em>baz</em> bim</strong> bop</em></p>
There can be no empty emphasis or strong emphasis:
**** is not an empty strong emphasis
<p>**** is not an empty strong emphasis</p>
Rule 10:
Any nonempty sequence of inline elements can be the contents of an strongly emphasized span.
In particular, emphasis and strong emphasis can be nested inside strong emphasis:
But note:
The difference is that in the preceding case, the internal delimiters can close emphasis, while in the cases with spaces, they cannot.
Indefinite levels of nesting are possible:
**foo *bar **baz**
bim* bop**
<p><strong>foo <em>bar <strong>baz</strong>
bim</em> bop</strong></p>
**foo [*bar*](/url)**
<p><strong>foo <a href="/url"><em>bar</em></a></strong></p>
There can be no empty emphasis or strong emphasis:
____ is not an empty strong emphasis
<p>____ is not an empty strong emphasis</p>
Rule 11:
Note that when delimiters do not match evenly, Rule 11 determines
that the excess literal *
characters will appear outside of the
emphasis, rather than inside it:
Rule 12:
Note that when delimiters do not match evenly, Rule 12 determines
that the excess literal _
characters will appear outside of the
emphasis, rather than inside it:
Rule 13 implies that if you want emphasis nested directly inside emphasis, you must use different delimiters:
However, strong emphasis within strong emphasis is possible without switching delimiters:
Rule 13 can be applied to arbitrarily long sequences of delimiters:
Rule 14:
Rule 15:
*foo __bar *baz bim__ bam*
<p><em>foo <strong>bar *baz bim</strong> bam</em></p>
Rule 16:
Rule 17:
**a<http://foo.bar/?q=**>
<p>**a<a href="http://foo.bar/?q=**">http://foo.bar/?q=**</a></p>
__a<http://foo.bar/?q=__>
<p>__a<a href="http://foo.bar/?q=__">http://foo.bar/?q=__</a></p>
A link contains link text (the visible text), a link destination (the URI that is the link destination), and optionally a link title. There are two basic kinds of links in Markdown. In inline links the destination and title are given immediately after the link text. In reference links the destination and title are defined elsewhere in the document.
A link text consists of a sequence of zero or more
inline elements enclosed by square brackets ([
and ]
). The
following rules apply:
Links may not contain other links, at any level of nesting. If multiple otherwise valid link definitions appear nested inside each other, the inner-most definition is used.
Brackets are allowed in the link text only if (a) they
are backslash-escaped or (b) they appear as a matched pair of brackets,
with an open bracket [
, a sequence of zero or more inlines, and
a close bracket ]
.
Backtick code spans, autolinks, and raw HTML tags bind more tightly
than the brackets in link text. Thus, for example,
[foo`]`
could not be a link text, since the second ]
is part of a code span.
The brackets in link text bind more tightly than markers for
emphasis and strong emphasis. Thus, for example, *[foo*](url)
is a link.
A link destination consists of either
a sequence of zero or more characters between an opening <
and a
closing >
that contains no spaces, line breaks, or unescaped
<
or >
characters, or
a nonempty sequence of characters that does not include ASCII space or control characters, and includes parentheses only if (a) they are backslash-escaped or (b) they are part of a balanced pair of unescaped parentheses that is not itself inside a balanced pair of unescaped parentheses.
A link title consists of either
a sequence of zero or more characters between straight double-quote
characters ("
), including a "
character only if it is
backslash-escaped, or
a sequence of zero or more characters between straight single-quote
characters ('
), including a '
character only if it is
backslash-escaped, or
a sequence of zero or more characters between matching parentheses
((...)
), including a )
character only if it is backslash-escaped.
Although link titles may span multiple lines, they may not contain a blank line.
An inline link consists of a link text followed immediately
by a left parenthesis (
, optional whitespace, an optional
link destination, an optional link title separated from the link
destination by whitespace, optional whitespace, and a right
parenthesis )
. The link’s text consists of the inlines contained
in the link text (excluding the enclosing square brackets).
The link’s URI consists of the link destination, excluding enclosing
<...>
if present, with backslash-escapes in effect as described
above. The link’s title consists of the link title, excluding its
enclosing delimiters, with backslash-escapes in effect as described
above.
Here is a simple inline link:
The title may be omitted:
Both the title and the destination may be omitted:
The destination cannot contain spaces or line breaks, even if enclosed in pointy brackets:
Parentheses inside the link destination may be escaped:
One level of balanced parentheses is allowed without escaping:
However, if you have parentheses within parentheses, you need to escape
or use the <...>
form:
Parentheses and other symbols can also be escaped, as usual in Markdown:
A link can contain fragment identifiers and queries:
[link](#fragment)
[link](http://example.com#fragment)
[link](http://example.com?foo=3#frag)
<p><a href="#fragment">link</a></p>
<p><a href="http://example.com#fragment">link</a></p>
<p><a href="http://example.com?foo=3#frag">link</a></p>
Note that a backslash before a non-escapable character is just a backslash:
URL-escaping should be left alone inside the destination, as all URL-escaped characters are also valid URL characters. Entity and numerical character references in the destination will be parsed into the corresponding Unicode code points, as usual. These may be optionally URL-escaped when written as HTML, but this spec does not enforce any particular policy for rendering URLs in HTML or other formats. Renderers may make different decisions about how to escape or normalize URLs in the output.
Note that, because titles can often be parsed as destinations, if you try to omit the destination and keep the title, you’ll get unexpected results:
Titles may be in single quotes, double quotes, or parentheses:
[link](/url "title")
[link](/url 'title')
[link](/url (title))
<p><a href="/url" title="title">link</a>
<a href="/url" title="title">link</a>
<a href="/url" title="title">link</a></p>
Backslash escapes and entity and numeric character references may be used in titles:
[link](/url "title \""")
<p><a href="/url" title="title """>link</a></p>
Nested balanced quotes are not allowed without escaping:
[link](/url "title "and" title")
<p>[link](/url "title "and" title")</p>
But it is easy to work around this by using a different quote type:
[link](/url 'title "and" title')
<p><a href="/url" title="title "and" title">link</a></p>
(Note: Markdown.pl
did allow double quotes inside a double-quoted
title, and its test suite included a test demonstrating this.
But it is hard to see a good rationale for the extra complexity this
brings, since there are already many ways—backslash escaping,
entity and numeric character references, or using a different
quote type for the enclosing title—to write titles containing
double quotes. Markdown.pl
’s handling of titles has a number
of other strange features. For example, it allows single-quoted
titles in inline links, but not reference links. And, in
reference links but not inline links, it allows a title to begin
with "
and end with )
. Markdown.pl
1.0.1 even allows
titles with no closing quotation mark, though 1.0.2b8 does not.
It seems preferable to adopt a simple, rational rule that works
the same way in inline links and link reference definitions.)
Whitespace is allowed around the destination and title:
But it is not allowed between the link text and the following parenthesis:
The link text may contain balanced brackets, but not unbalanced ones, unless they are escaped:
The link text may contain inline content:
[link *foo **bar** `#`*](/uri)
<p><a href="/uri">link <em>foo <strong>bar</strong> <code>#</code></em></a></p>
[![moon](moon.jpg)](/uri)
<p><a href="/uri"><img src="moon.jpg" alt="moon" /></a></p>
However, links may not contain other links, at any level of nesting.
[foo *[bar [baz](/uri)](/uri)*](/uri)
<p>[foo <em>[bar <a href="/uri">baz</a>](/uri)</em>](/uri)</p>
These cases illustrate the precedence of link text grouping over emphasis grouping:
Note that brackets that aren’t part of links do not take precedence:
These cases illustrate the precedence of HTML tags, code spans, and autolinks over link grouping:
[foo<http://example.com/?search=](uri)>
<p>[foo<a href="http://example.com/?search=%5D(uri)">http://example.com/?search=](uri)</a></p>
There are three kinds of reference links: full, collapsed, and shortcut.
A full reference link consists of a link text immediately followed by a link label that matches a link reference definition elsewhere in the document.
A link label begins with a left bracket ([
) and ends
with the first right bracket (]
) that is not backslash-escaped.
Between these brackets there must be at least one non-whitespace character.
Unescaped square bracket characters are not allowed in
link labels. A link label can have at most 999
characters inside the square brackets.
One label matches another just in case their normalized forms are equal. To normalize a label, perform the Unicode case fold and collapse consecutive internal whitespace to a single space. If there are multiple matching reference link definitions, the one that comes first in the document is used. (It is desirable in such cases to emit a warning.)
The contents of the first link label are parsed as inlines, which are used as the link’s text. The link’s URI and title are provided by the matching link reference definition.
Here is a simple example:
The rules for the link text are the same as with inline links. Thus:
The link text may contain balanced brackets, but not unbalanced ones, unless they are escaped:
[link [foo [bar]]][ref]
[ref]: /uri
<p><a href="/uri">link [foo [bar]]</a></p>
The link text may contain inline content:
[link *foo **bar** `#`*][ref]
[ref]: /uri
<p><a href="/uri">link <em>foo <strong>bar</strong> <code>#</code></em></a></p>
[![moon](moon.jpg)][ref]
[ref]: /uri
<p><a href="/uri"><img src="moon.jpg" alt="moon" /></a></p>
However, links may not contain other links, at any level of nesting.
[foo [bar](/uri)][ref]
[ref]: /uri
<p>[foo <a href="/uri">bar</a>]<a href="/uri">ref</a></p>
[foo *bar [baz][ref]*][ref]
[ref]: /uri
<p>[foo <em>bar <a href="/uri">baz</a></em>]<a href="/uri">ref</a></p>
(In the examples above, we have two shortcut reference links instead of one full reference link.)
The following cases illustrate the precedence of link text grouping over emphasis grouping:
These cases illustrate the precedence of HTML tags, code spans, and autolinks over link grouping:
[foo<http://example.com/?search=][ref]>
[ref]: /uri
<p>[foo<a href="http://example.com/?search=%5D%5Bref%5D">http://example.com/?search=][ref]</a></p>
Matching is case-insensitive:
Unicode case fold is used:
[Толпой][Толпой] is a Russian word.
[ТОЛПОЙ]: /url
<p><a href="/url">Толпой</a> is a Russian word.</p>
Consecutive internal whitespace is treated as one space for purposes of determining matching:
No whitespace is allowed between the link text and the link label:
[foo] [bar]
[bar]: /url "title"
<p>[foo] <a href="/url" title="title">bar</a></p>
[foo]
[bar]
[bar]: /url "title"
<p>[foo]
<a href="/url" title="title">bar</a></p>
This is a departure from John Gruber’s original Markdown syntax description, which explicitly allows whitespace between the link text and the link label. It brings reference links in line with inline links, which (according to both original Markdown and this spec) cannot have whitespace after the link text. More importantly, it prevents inadvertent capture of consecutive shortcut reference links. If whitespace is allowed between the link text and the link label, then in the following we will have a single reference link, not two shortcut reference links, as intended:
[foo]
[bar]
[foo]: /url1
[bar]: /url2
(Note that shortcut reference links were introduced by Gruber
himself in a beta version of Markdown.pl
, but never included
in the official syntax description. Without shortcut reference
links, it is harmless to allow space between the link text and
link label; but once shortcut references are introduced, it is
too dangerous to allow this, as it frequently leads to
unintended results.)
When there are multiple matching link reference definitions, the first is used:
Note that matching is performed on normalized strings, not parsed inline content. So the following does not match, even though the labels define equivalent inline content:
Link labels cannot contain brackets, unless they are backslash-escaped:
[foo][ref[bar]]
[ref[bar]]: /uri
<p>[foo][ref[bar]]</p>
<p>[ref[bar]]: /uri</p>
Note that in this example ]
is not backslash-escaped:
A link label must contain at least one non-whitespace character:
A collapsed reference link
consists of a link label that matches a
link reference definition elsewhere in the
document, followed by the string []
.
The contents of the first link label are parsed as inlines,
which are used as the link’s text. The link’s URI and title are
provided by the matching reference link definition. Thus,
[foo][]
is equivalent to [foo][foo]
.
[*foo* bar][]
[*foo* bar]: /url "title"
<p><a href="/url" title="title"><em>foo</em> bar</a></p>
The link labels are case-insensitive:
As with full reference links, whitespace is not allowed between the two sets of brackets:
A shortcut reference link
consists of a link label that matches a
link reference definition elsewhere in the
document and is not followed by []
or a link label.
The contents of the first link label are parsed as inlines,
which are used as the link’s text. the link’s URI and title
are provided by the matching link reference definition.
Thus, [foo]
is equivalent to [foo][]
.
[*foo* bar]
[*foo* bar]: /url "title"
<p><a href="/url" title="title"><em>foo</em> bar</a></p>
[[*foo* bar]]
[*foo* bar]: /url "title"
<p>[<a href="/url" title="title"><em>foo</em> bar</a>]</p>
The link labels are case-insensitive:
A space after the link text should be preserved:
If you just want bracketed text, you can backslash-escape the opening bracket to avoid links:
Note that this is a link, because a link label ends with the first following closing bracket:
Full references take precedence over shortcut references:
In the following case [bar][baz]
is parsed as a reference,
[foo]
as normal text:
Here, though, [foo][bar]
is parsed as a reference, since
[bar]
is defined:
[foo][bar][baz]
[baz]: /url1
[bar]: /url2
<p><a href="/url2">foo</a><a href="/url1">baz</a></p>
Here [foo]
is not parsed as a shortcut reference, because it
is followed by a link label (even though [bar]
is not defined):
Syntax for images is like the syntax for links, with one
difference. Instead of link text, we have an
image description. The rules for this are the
same as for link text, except that (a) an
image description starts with ![
rather than [
, and
(b) an image description may contain links.
An image description has inline elements
as its contents. When an image is rendered to HTML,
this is standardly used as the image’s alt
attribute.
![foo *bar*]
[foo *bar*]: train.jpg "train & tracks"
<p><img src="train.jpg" alt="foo bar" title="train & tracks" /></p>
Though this spec is concerned with parsing, not rendering, it is
recommended that in rendering to HTML, only the plain string content
of the image description be used. Note that in
the above example, the alt attribute’s value is foo bar
, not foo [bar](/url)
or foo <a href="/url">bar</a>
. Only the plain string
content is rendered, without formatting.
![foo *bar*][]
[foo *bar*]: train.jpg "train & tracks"
<p><img src="train.jpg" alt="foo bar" title="train & tracks" /></p>
![foo *bar*][foobar]
[FOOBAR]: train.jpg "train & tracks"
<p><img src="train.jpg" alt="foo bar" title="train & tracks" /></p>
My ![foo bar](/path/to/train.jpg "title" )
<p>My <img src="/path/to/train.jpg" alt="foo bar" title="title" /></p>
Reference-style:
Collapsed:
![foo][]
[foo]: /url "title"
<p><img src="/url" alt="foo" title="title" /></p>
![*foo* bar][]
[*foo* bar]: /url "title"
<p><img src="/url" alt="foo bar" title="title" /></p>
The labels are case-insensitive:
![Foo][]
[foo]: /url "title"
<p><img src="/url" alt="Foo" title="title" /></p>
As with reference links, whitespace is not allowed between the two sets of brackets:
![foo]
[]
[foo]: /url "title"
<p><img src="/url" alt="foo" title="title" />
[]</p>
Shortcut:
![*foo* bar]
[*foo* bar]: /url "title"
<p><img src="/url" alt="foo bar" title="title" /></p>
Note that link labels cannot contain unescaped brackets:
![[foo]]
[[foo]]: /url "title"
<p>![[foo]]</p>
<p>[[foo]]: /url "title"</p>
The link labels are case-insensitive:
If you just want bracketed text, you can backslash-escape the
opening !
and [
:
If you want a link after a literal !
, backslash-escape the
!
:
Autolinks are absolute URIs and email addresses inside
<
and >
. They are parsed as links, with the URL or email address
as the link label.
A URI autolink consists of <
, followed by an
absolute URI not containing <
, followed by >
. It is parsed as
a link to the URI, with the URI as the link’s label.
An absolute URI,
for these purposes, consists of a scheme followed by a colon (:
)
followed by zero or more characters other than ASCII
whitespace and control characters, <
, and >
. If
the URI includes these characters, they must be percent-encoded
(e.g. %20
for a space).
For purposes of this spec, a scheme is any sequence of 2–32 characters beginning with an ASCII letter and followed by any combination of ASCII letters, digits, or the symbols plus (”+”), period (”.”), or hyphen (”-”).
Here are some valid autolinks:
<http://foo.bar.baz>
<p><a href="http://foo.bar.baz">http://foo.bar.baz</a></p>
<http://foo.bar.baz/test?q=hello&id=22&boolean>
<p><a href="http://foo.bar.baz/test?q=hello&id=22&boolean">http://foo.bar.baz/test?q=hello&id=22&boolean</a></p>
<irc://foo.bar:2233/baz>
<p><a href="irc://foo.bar:2233/baz">irc://foo.bar:2233/baz</a></p>
Uppercase is also fine:
<MAILTO:FOO@BAR.BAZ>
<p><a href="MAILTO:FOO@BAR.BAZ">MAILTO:FOO@BAR.BAZ</a></p>
Note that many strings that count as absolute URIs for purposes of this spec are not valid URIs, because their schemes are not registered or because of other problems with their syntax:
<made-up-scheme://foo,bar>
<p><a href="made-up-scheme://foo,bar">made-up-scheme://foo,bar</a></p>
<localhost:5001/foo>
<p><a href="localhost:5001/foo">localhost:5001/foo</a></p>
Spaces are not allowed in autolinks:
Backslash-escapes do not work inside autolinks:
<http://example.com/\[\>
<p><a href="http://example.com/%5C%5B%5C">http://example.com/\[\</a></p>
An email autolink
consists of <
, followed by an email address,
followed by >
. The link’s label is the email address,
and the URL is mailto:
followed by the email address.
An email address, for these purposes, is anything that matches the non-normative regex from the HTML5 spec:
/^[a-zA-Z0-9.!#$%&'*+/=?^_`{|}~-]+@[a-zA-Z0-9](?:[a-zA-Z0-9-]{0,61}[a-zA-Z0-9])?
(?:\.[a-zA-Z0-9](?:[a-zA-Z0-9-]{0,61}[a-zA-Z0-9])?)*$/
Examples of email autolinks:
<foo@bar.example.com>
<p><a href="mailto:foo@bar.example.com">foo@bar.example.com</a></p>
<foo+special@Bar.baz-bar0.com>
<p><a href="mailto:foo+special@Bar.baz-bar0.com">foo+special@Bar.baz-bar0.com</a></p>
Backslash-escapes do not work inside email autolinks:
These are not autolinks:
Text between <
and >
that looks like an HTML tag is parsed as a
raw HTML tag and will be rendered in HTML without escaping.
Tag and attribute names are not limited to current HTML tags,
so custom tags (and even, say, DocBook tags) may be used.
Here is the grammar for tags:
A tag name consists of an ASCII letter
followed by zero or more ASCII letters, digits, or
hyphens (-
).
An attribute consists of whitespace, an attribute name, and an optional attribute value specification.
An attribute name
consists of an ASCII letter, _
, or :
, followed by zero or more ASCII
letters, digits, _
, .
, :
, or -
. (Note: This is the XML
specification restricted to ASCII. HTML5 is laxer.)
An attribute value specification
consists of optional whitespace,
a =
character, optional whitespace, and an attribute
value.
An attribute value consists of an unquoted attribute value, a single-quoted attribute value, or a double-quoted attribute value.
An unquoted attribute value
is a nonempty string of characters not
including spaces, "
, '
, =
, <
, >
, or `
.
A single-quoted attribute value
consists of '
, zero or more
characters not including '
, and a final '
.
A double-quoted attribute value
consists of "
, zero or more
characters not including "
, and a final "
.
An open tag consists of a <
character, a tag name,
zero or more attributes, optional whitespace, an optional /
character, and a >
character.
A closing tag consists of the string </
, a
tag name, optional whitespace, and the character >
.
An HTML comment consists of <!--
+ text + -->
,
where text does not start with >
or ->
, does not end with -
,
and does not contain --
. (See the
HTML5 spec.)
A processing instruction
consists of the string <?
, a string
of characters not including the string ?>
, and the string
?>
.
A declaration consists of the
string <!
, a name consisting of one or more uppercase ASCII letters,
whitespace, a string of characters not including the
character >
, and the character >
.
A CDATA section consists of
the string <![CDATA[
, a string of characters not including the string
]]>
, and the string ]]>
.
An HTML tag consists of an open tag, a closing tag, an HTML comment, a processing instruction, a declaration, or a CDATA section.
Here are some simple open tags:
Empty elements:
Whitespace is allowed:
With attributes:
<a foo="bar" bam = 'baz <em>"</em>'
_boolean zoop:33=zoop:33 />
<p><a foo="bar" bam = 'baz <em>"</em>'
_boolean zoop:33=zoop:33 /></p>
Custom tag names can be used:
Foo <responsive-image src="foo.jpg" />
<p>Foo <responsive-image src="foo.jpg" /></p>
Illegal tag names, not parsed as HTML:
Illegal attribute names:
Illegal attribute values:
Illegal whitespace:
Missing whitespace:
Closing tags:
Illegal attributes in closing tag:
Comments:
foo <!-- this is a
comment - with hyphen -->
<p>foo <!-- this is a
comment - with hyphen --></p>
foo <!-- not a comment -- two hyphens -->
<p>foo <!-- not a comment -- two hyphens --></p>
Not comments:
foo <!--> foo -->
foo <!-- foo--->
<p>foo <!--> foo --></p>
<p>foo <!-- foo---></p>
Processing instructions:
Declarations:
CDATA sections:
Entity and numeric character references are preserved in HTML attributes:
Backslash escapes do not work in HTML attributes:
A line break (not in a code span or HTML tag) that is preceded
by two or more spaces and does not occur at the end of a block
is parsed as a hard line break (rendered
in HTML as a <br />
tag):
For a more visible alternative, a backslash before the line ending may be used instead of two spaces:
More than two spaces can be used:
Leading spaces at the beginning of the next line are ignored:
Line breaks can occur inside emphasis, links, and other constructs that allow inline content:
Line breaks do not occur inside code spans
or HTML tags:
Hard line breaks are for separating inline content within a block. Neither syntax for hard line breaks works at the end of a paragraph or other block element:
A regular line break (not in a code span or HTML tag) that is not preceded by two or more spaces or a backslash is parsed as a softbreak. (A softbreak may be rendered in HTML either as a line ending or as a space. The result will be the same in browsers. In the examples here, a line ending will be used.)
Spaces at the end of the line and beginning of the next line are removed:
A conforming parser may render a soft line break in HTML either as a line break or as a space.
A renderer may also provide an option to render soft line breaks as hard line breaks.
Any characters not given an interpretation by the above rules will be parsed as plain textual content.
Internal spaces are preserved verbatim:
In this appendix we describe some features of the parsing strategy used in the CommonMark reference implementations.
Parsing has two phases:
In the first phase, lines of input are consumed and the block structure of the document—its division into paragraphs, block quotes, list items, and so on—is constructed. Text is assigned to these blocks but not parsed. Link reference definitions are parsed and a map of links is constructed.
In the second phase, the raw text contents of paragraphs and headings are parsed into sequences of Markdown inline elements (strings, code spans, links, emphasis, and so on), using the map of link references constructed in phase 1.
At each point in processing, the document is represented as a tree of
blocks. The root of the tree is a document
block. The document
may have any number of other blocks as children. These children
may, in turn, have other blocks as children. The last child of a block
is normally considered open, meaning that subsequent lines of input
can alter its contents. (Blocks that are not open are closed.)
Here, for example, is a possible document tree, with the open blocks
marked by arrows:
-> document
-> block_quote
paragraph
"Lorem ipsum dolor\nsit amet."
-> list (type=bullet tight=true bullet_char=-)
list_item
paragraph
"Qui *quodsi iracundia*"
-> list_item
-> paragraph
"aliquando id"
Each line that is processed has an effect on this tree. The line is analyzed and, depending on its contents, the document may be altered in one or more of the following ways:
Once a line has been incorporated into the tree in this way, it can be discarded, so input can be read in a stream.
For each line, we follow this procedure:
First we iterate through the open blocks, starting with the
root document, and descending through last children down to the last
open block. Each block imposes a condition that the line must satisfy
if the block is to remain open. For example, a block quote requires a
>
character. A paragraph requires a non-blank line.
In this phase we may match all or just some of the open
blocks. But we cannot close unmatched blocks yet, because we may have a
lazy continuation line.
Next, after consuming the continuation markers for existing
blocks, we look for new block starts (e.g. >
for a block quote.
If we encounter a new block start, we close any blocks unmatched
in step 1 before creating the new block as a child of the last
matched block.
Finally, we look at the remainder of the line (after block
markers like >
, list markers, and indentation have been consumed).
This is text that can be incorporated into the last open
block (a paragraph, code block, heading, or raw HTML).
Setext headings are formed when we see a line of a paragraph that is a setext heading underline.
Reference link definitions are detected when a paragraph is closed; the accumulated text lines are parsed to see if they begin with one or more reference link definitions. Any remainder becomes a normal paragraph.
We can see how this works by considering how the tree above is generated by four lines of Markdown:
> Lorem ipsum dolor
sit amet.
> - Qui *quodsi iracundia*
> - aliquando id
At the outset, our document model is just
-> document
The first line of our text,
> Lorem ipsum dolor
causes a block_quote
block to be created as a child of our
open document
block, and a paragraph
block as a child of
the block_quote
. Then the text is added to the last open
block, the paragraph
:
-> document
-> block_quote
-> paragraph
"Lorem ipsum dolor"
The next line,
sit amet.
is a “lazy continuation” of the open paragraph
, so it gets added
to the paragraph’s text:
-> document
-> block_quote
-> paragraph
"Lorem ipsum dolor\nsit amet."
The third line,
> - Qui *quodsi iracundia*
causes the paragraph
block to be closed, and a new list
block
opened as a child of the block_quote
. A list_item
is also
added as a child of the list
, and a paragraph
as a child of
the list_item
. The text is then added to the new paragraph
:
-> document
-> block_quote
paragraph
"Lorem ipsum dolor\nsit amet."
-> list (type=bullet tight=true bullet_char=-)
-> list_item
-> paragraph
"Qui *quodsi iracundia*"
The fourth line,
> - aliquando id
causes the list_item
(and its child the paragraph
) to be closed,
and a new list_item
opened up as child of the list
. A paragraph
is added as a child of the new list_item
, to contain the text.
We thus obtain the final tree:
-> document
-> block_quote
paragraph
"Lorem ipsum dolor\nsit amet."
-> list (type=bullet tight=true bullet_char=-)
list_item
paragraph
"Qui *quodsi iracundia*"
-> list_item
-> paragraph
"aliquando id"
Once all of the input has been parsed, all open blocks are closed.
We then “walk the tree,” visiting every node, and parse raw string contents of paragraphs and headings as inlines. At this point we have seen all the link reference definitions, so we can resolve reference links as we go.
document
block_quote
paragraph
str "Lorem ipsum dolor"
softbreak
str "sit amet."
list (type=bullet tight=true bullet_char=-)
list_item
paragraph
str "Qui "
emph
str "quodsi iracundia"
list_item
paragraph
str "aliquando id"
Notice how the line ending in the first paragraph has
been parsed as a softbreak
, and the asterisks in the first list item
have become an emph
.
By far the trickiest part of inline parsing is handling emphasis, strong emphasis, links, and images. This is done using the following algorithm.
When we’re parsing inlines and we hit either
*
or _
characters, or[
or ![
we insert a text node with these symbols as its literal content, and we add a pointer to this text node to the delimiter stack.
The delimiter stack is a doubly linked list. Each element contains a pointer to a text node, plus information about
[
, ![
, *
, _
)When we hit a ]
character, we call the look for link or image
procedure (see below).
When we hit the end of the input, we call the process emphasis
procedure (see below), with stack_bottom
= NULL.
Starting at the top of the delimiter stack, we look backwards
through the stack for an opening [
or ![
delimiter.
If we don’t find one, we return a literal text node ]
.
If we do find one, but it’s not active, we remove the inactive
delimiter from the stack, and return a literal text node ]
.
If we find one and it’s active, then we parse ahead to see if we have an inline link/image, reference link/image, compact reference link/image, or shortcut reference link/image.
If we don’t, then we remove the opening delimiter from the
delimiter stack and return a literal text node ]
.
If we do, then
We return a link or image node whose children are the inlines after the text node pointed to by the opening delimiter.
We run process emphasis on these inlines, with the [
opener
as stack_bottom
.
We remove the opening delimiter.
If we have a link (and not an image), we also set all
[
delimiters before the opening delimiter to inactive. (This
will prevent us from getting links within links.)
Parameter stack_bottom
sets a lower bound to how far we
descend in the delimiter stack. If it is NULL, we can
go all the way to the bottom. Otherwise, we stop before
visiting stack_bottom
.
Let current_position
point to the element on the delimiter stack
just above stack_bottom
(or the first element if stack_bottom
is NULL).
We keep track of the openers_bottom
for each delimiter
type (*
, _
). Initialize this to stack_bottom
.
Then we repeat the following until we run out of potential closers:
Move current_position
forward in the delimiter stack (if needed)
until we find the first potential closer with delimiter *
or _
.
(This will be the potential closer closest
to the beginning of the input – the first one in parse order.)
Now, look back in the stack (staying above stack_bottom
and
the openers_bottom
for this delimiter type) for the
first matching potential opener (“matching” means same delimiter).
If one is found:
Figure out whether we have emphasis or strong emphasis: if both closer and opener spans have length >= 2, we have strong, otherwise regular.
Insert an emph or strong emph node accordingly, after the text node corresponding to the opener.
Remove any delimiters between the opener and closer from the delimiter stack.
Remove 1 (for regular emph) or 2 (for strong emph) delimiters
from the opening and closing text nodes. If they become empty
as a result, remove them and remove the corresponding element
of the delimiter stack. If the closing node is removed, reset
current_position
to the next element in the stack.
If none in found:
Set openers_bottom
to the element before current_position
.
(We know that there are no openers for this kind of closer up to and
including this point, so this puts a lower bound on future searches.)
If the closer at current_position
is not a potential opener,
remove it from the delimiter stack (since we know it can’t
be a closer either).
Advance current_position
to the next element in the stack.
After we’re done, we remove all delimiters above stack_bottom
from the
delimiter stack.