Haddock understands certain textual cues inside documentation annotations that tell it how to render the documentation. The cues (or “markup”) have been designed to be simple and mnemonic in ASCII so that the programmer doesn't have to deal with heavyweight annotations when editing documentation comments.
The following characters have special meanings in
documentation comments: \
, /
,
'
, `
,
"
, @
,
<
. To insert a literal occurrence of
one of these special characters, precede it with a backslash
(\
).
Additionally, the character >
has
a special meaning at the beginning of a line, and the
following characters have special meanings at the beginning of
a paragraph:
*
, -
. These characters
can also be escaped using \
.
Furthermore, the character sequence >>>
has a special meaning at the beginning of a line. To
escape it, just prefix the characters in the sequence with a
backslash.
Although Haskell source files may contain any character
from the Unicode character set, the encoding of these characters
as bytes varies between systems, so that only source files
restricted to the ASCII character set are portable. Other
characters may be specified in character and string literals
using Haskell character escapes. To represent such characters
in documentation comments, Haddock supports SGML-style numeric
character references of the forms
&#
D
;
and
&#x
H
;
where D
and H
are decimal and hexadecimal numbers denoting a code position
in Unicode (or ISO 10646). For example, the references
λ
, λ
and λ
all represent the lower-case
letter lambda.
Displayed blocks of code are indicated by surrounding a
paragraph with @...@
or by preceding each
line of a paragraph with >
(we often
call these “bird tracks”). For
example:
-- | This documentation includes two blocks of code: -- -- @ -- f x = x + x -- @ -- -- > g x = x * 42
There is an important difference between the two forms
of code block: in the bird-track form, the text to the right
of the ’>
‚ is interpreted
literally, whereas the @...@
form
interprets markup as normal inside the code block.
Haddock has markup support for examples of interaction with a
read-eval-print loop (REPL). An
example is introduced with
>>>
followed by an expression followed
by zero or more result lines:
-- | Two examples are given below: -- -- >>> fib 10 -- 55 -- -- >>> putStrLn "foo\nbar" -- foo -- bar
Result lines that only contain the string
<BLANKLINE>
are rendered as blank lines in the
generated documentation.
Haddock provides markup for properties:
-- | Addition is commutative: -- -- prop> a + b = b + a
This allows third-party applications to extract and verify them.
Referring to a Haskell identifier, whether it be a type, class, constructor, or function, is done by surrounding it with single quotes:
-- | This module defines the type 'T'.
If there is an entity T
in scope in
the current module, then the documentation will hyperlink the
reference in the text to the definition of
T
(if the output format supports
hyperlinking, of course; in a printed format it might instead
insert a page reference to the definition).
It is also possible to refer to entities that are not in scope in the current module, by giving the full qualified name of the entity:
-- | The identifier 'M.T' is not in scope
If M.T
is not otherwise in scope,
then Haddock will simply emit a link pointing to the entity
T
exported from module M
(without checking to see whether either M
or M.T
exist).
To make life easier for documentation writers, a quoted identifier is only interpreted as such if the quotes surround a lexically valid Haskell identifier. This means, for example, that it normally isn't necessary to escape the single quote when used as an apostrophe:
-- | I don't have to escape my apostrophes; great, isn't it?
Nothing special is needed to hyperlink identifiers which
contain apostrophes themselves: to hyperlink
foo'
one would simply type
'foo''
. To hyperlink identifiers written in
infix form, simply put them in quotes as always:
'`elem`'
.
For compatibility with other systems, the following
alternative form of markup is accepted[3]: `T'
.
Emphasis may be added by surrounding text with
/.../
. Other markup is valid inside emphasis. To have a forward
slash inside of emphasis, just escape it: /fo\/o/
Bold (strong) text is indicated by surrounding it with __...__
.
Other markup is valid inside bold. For example, __/foo/__
will make the emphasised
text foo
bold. You don't have to escape a single underscore if you need it bold:
__This_text_with_underscores_is_bold__
.
Monospaced (or typewriter) text is indicated by
surrounding it with @...@
. Other markup is
valid inside a monospaced span: for example
@'f' a b@
will hyperlink the
identifier f
inside the code fragment.
Linking to a module is done by surrounding the module name with double quotes:
-- | This is a reference to the "Foo" module.
A basic check is done on the syntax of the header name to ensure that it is valid before turning it into a link but unlike with identifiers, whether the module is in scope isn't checked and will always be turned into a link.
A bulleted item is represented by preceding a paragraph
with either “*
” or
“-
”. A sequence of bulleted
paragraphs is rendered as an itemized list in the generated
documentation, eg.:
-- | This is a bulleted list: -- -- * first item -- -- * second item
An enumerated list is similar, except each paragraph
must be preceded by either
“(
”
or
“n
)
”
where n
.n
is any integer. e.g.
-- | This is an enumerated list: -- -- (1) first item -- -- 2. second item
Lists of the same type don't have to be separated by a newline:
-- | This is an enumerated list: -- -- (1) first item -- 2. second item -- -- This is a bulleted list: -- -- * first item -- * second item
You can have more than one line of content in a list element:
-- | -- * first item -- and more content for the first item -- * second item -- and more content for the second item
You can even nest whole paragraphs inside of list elements. The rules are 4 spaces for each indentation level. You're required to use a newline before such nested paragraph:
{-| * Beginning of list This belongs to the list above! > nested > bird > tracks * Next list More of the indented list. * Deeper @ even code blocks work @ * Deeper 1. Even deeper! 2. No newline separation even in indented lists. -}
The indentation of the first list item is honoured. That is, in the following example the items are on the same level. Before Haddock 2.16.1, the second item would have been nested under the first item which was unexpected.
{-| * foo * bar -}
Definition lists are written as follows:
-- | This is a definition list: -- -- [@foo@]: The description of @foo@. -- -- [@bar@]: The description of @bar@.
To produce output something like this:
foo
The description of foo
.
bar
The description of bar
.
Each paragraph should be preceded by the
“definition term” enclosed in square brackets and followed by a colon.
Other markup operators may be used freely within the
definition term. You can escape ]
with a backslash as usual.
Same rules about nesting and no newline separation as for bulleted and numbered lists apply.
A URL can be included in a documentation comment by surrounding it in angle brackets, for example:
<http://example.com>
If the output format supports it, the URL will be turned into a hyperlink when rendered.
If Haddock sees something that looks like a URL (such as something starting with
http://
or ssh://
) where the URL markup is valid,
it will automatically make it a hyperlink.
Haddock supports Markdown syntax for inline links. A link consists of a link text and a URL. The link text is enclosed in square brackets and followed by the URL enclosed in regular parentheses, for example:
[some link](http://example.com)
The link text is used as a descriptive text for the URL, if the output format supports it.
Haddock supports Markdown syntax for inline images. This resembles the syntax for links, but starts with an exclamation mark. An example looks like this:

If the output format supports it, the image will be rendered inside the documentation. The image description is used as relpacement text and/or image title.
Sometimes it is useful to be able to link to a point in
the documentation which doesn't correspond to a particular
entity. For that purpose, we allow anchors to be
included in a documentation comment. The syntax is
#
, where
label
#label
is the name of the anchor.
An anchor is invisible in the generated documentation.
To link to an anchor from elsewhere, use the syntax
"
where module
#label
"module
is the module name
containing the anchor, and label
is
the anchor label. The module does not have to be local, it can
be imported via an interface. Please note that in Haddock
versions 2.13.x and earlier, the syntax was
"
.
It is considered deprecated and will be removed in the future.module
\#label
"
Headings inside of comment documentation are possible be preceding them with
a number of =
s. From 1 to 6 are accepted. Extra =
s will
be treated as belonging to the text of the heading. Note that it's up to the output format to decide
how to render the different levels.
-- | -- = Heading level 1 with some /emphasis/ -- Something underneath the heading. -- -- == /Subheading/ -- More content. -- -- === Subsubheading -- Even more content.
Note that while headings have to start on a new paragraph, we allow paragraph-level content to follow these immediately.
-- | -- = Heading level 1 with some __bold__ -- Something underneath the heading. -- -- == /Subheading/ -- More content. -- -- === Subsubheading -- >>> examples are only allowed at the start of paragraphs
As of 2.15.1, there's experimental (read: subject to
change or get removed) support for collapsible headers: simply
wrap your existing header title in underscores, as per bold
syntax. The collapsible section will stretch until the end of
the comment or until a header of equal or smaller number of
=
s.
-- | -- === __Examples:__ -- >>> Some very long list of examples -- -- ==== This still falls under the collapse -- Some specialised examples -- -- === This is does not go into the collapsable section. -- More content.
Since Haddock 2.16.0, some support for embedding metadata in the comments has started to appear. The use of such data aims to standardise various community conventions in how such information is conveyed and to provide uniform rendering.
@since
annotation can be used to
convey information about when the function was introduced or
when it has changed in the way significant to the user.
@since
is a paragraph-level element.
While multiple such annotations are not an error, only the
one to appear in the comment last will be used.
@since
has to be followed with a version
number, no further description is currently allowed. The
meaning of this feature is subject to change in the future
per user feedback.
-- | -- Some comment -- -- @since 1.2.3
[3]
We chose not to use this as the primary markup for
identifiers because strictly speaking the `
character should not be used as a left quote, it is a grave accent.