Note: If you are viewing this Readme on GitHub, its Blogdown-specific features will not render correctly.
The Blogdown-formatted output is in test/goldens/Readme.html
.
Blogdown is a markup language based on Markdown, designed for writing blog posts.
Blogdown's goals are:
- Clean syntax for common blog post features absent from Markdown.
- Ability to handle untrusted input.
- Simple syntax, with no surprises.
- Debuggability.
- Near compatibility with common Markdown implementations.
Because there is no Markdown standard and existing Markdown implementations
disagree wildly even on simple cases,
Blogdown cannot be 100% compatible with even a majority of Markdown implementations.
While there have been attemps to create a common Markdown standard--most notably CommonMark--they
are necessarily quite complex. The primary cause of this complexity is that Markdown insists on rendering something for every input,
no matter how malformed. Blogdown is considerably simpler, and hopefully easier for authors to debug, because it fails on malformed inputs.
With full compatability out of the window, I have chosen to make some other small improvements on Markdown syntax.
Installation
With Cabal or Stack
The recommended way to install any Haskell project is using Cabal or Stack.
With these tools, you can simply run cabal install
or stack build
respectively.
Because configuring Cabal and Stack can be tricky for Haskell beginners, Blogdown supports installation without any Haskell tooling.
First, install GHC, Parsec and MissingH,
all of which are available through common Linux package managers.
To compile static assets into Haskell files, run ghci Setup.hs
and manually invoke compileStaticFiles
.
Then run ghc -isrc -iassets -o Blogdown src/Blogdown.hs
in the repository's base directory.
Usage
The Blogdown
binary reads from stdin
and writes to stdout
. Typical usage looks like:
cat blogpost.md | ./Blogdown > blogpost.html
Optional Styling and Scripts
It is recommended to include footnotes.css
and footnotes.js
on any pages which make use of Blogdown-generated footnotes,
which improve the appearance of footnotes and allow them to be shown inline.
These can be inlined using the --inline-css
and --inline-js
flags respectively^[inline].
Optional Flags
Blogdown
accepts the following long-style flags:
--allowed-tags
: Specifies a comma-separated list of HTML tags which should be
rendered faithfully. All other tags are escaped, e.g. <a>
becomes &lt;a&gt;
.
If no list is supplied, all tags are escaped.
--allowed-attributes
: Specifies a comma-separated list of HTML attributes which
should be rendered faithfully. Any other attributes will be stripped from tags when
they are rendered. If no list is supplied, all attributes are stripped.
--allow-unsafe-tags
: By default, Blogdown will fail when it encounters a <script>
or a <style>
tag, because there are certain corner cases it cannot parse correctly,
e.g. <style>"</style>"evil()"<style>"</style>
. However, these are unlikely on
non-malicious input, so this flag can be passed to attempt parsing these tags.
--em-dashes
: If this flag is passed, --
will be replaced with "—" in text.
--footnote-backlinks
: If this flag is passed, footnotes will be preceded by a caret linking back to the point where the footnote is referenced.
--footnote-index-from
: The index from which footnotes are numbered. Defaults to 0.
--footnote-prefix
: Defines a prefix for the id
s of footnotes. Recommended if multiple output files are included in a single HTML page, to avoid id
collisions.
--inline-css
: If this flag is passed, the recommended CSS will be inlined at the end of the output document.
--inline-js
: If this flag is passed, the recommended JS will be inlined at the end of the output document.
Syntax
Differences from Markdown
Most of the syntax of Blogdown should be familiar to Markdown users, but some new syntax has been added, and some existing syntax has changed.
New Features
Blogdown adds footnote support to Markdown.
Footnotes can be referenced inline with ^[footnote-name], which will render as a superscript link to a footnote-definition
at the end of the document, which is defined by ~[footnote-name] followed by the footnote contents.
Markdown Incompatibilities
Blogdown does not support the Markdown syntax of underlining text with =
or -
characters to define a header,
as this comes at a large cost in the parser implementation^[underline-parser-complexity].
The #
syntax for headers is supported instead.
It also does not support using multiple trailing spaces to force a breakpoint at the end of a line.
The <br/>
tag is supported instead.
While tables are not a feature of base Markdown, some common Markdown implementations
such as Github Flavored Markdown
support them. Blogdown also supports tables, but its implementation is slightly different
from Github's, requiring |
characters at the start and end of a row and using +
instead of |
in the separator between the (optional) table header and table body.
Since Blogdown introduces new syntax, some valid Markdown will require escaping to render as expected in Blogdown.
Additionally, while most Markdown implementations do not require escaping many special characters when their special meaning would
not be valid, Blogdown always requires they be escaped.
HTML Embedding
Like most Markdown implementations, Blogdown documents can have HTML embedded inside them.
However, Blogdown allows only a limited subset of HTML, specifically
XHTML,
with the exceptions that unknown tags are permitted, and that any tag
is allowed to be self-closing^[arbitrary-tags].
By default Blogdown will not allow <script>
or <style>
tags, since parsing
for these is very complex and not fully implemented. However, when parsing trusted
content it is generally safe to use --allow-unsafe-tags
to attempt to allow
these anyway.
Note also that any content within an HTML node is rendered verbatim, so Blogdown features
cannot be used inside HTML nodes.
HTML Bleaching
If using Blogdown to parse untrusted content (such as comments on a blog),
the --allowed-tags
and --allowed-attributes
flags must be used to restrict what
HTML can be rendered in the output. Generally these flags should be used without
without arguments, bleaching all HTML. If you must allow some HTML, you should only
allow tags and attributes which are necessary for your use case and which you understand
the security implications of.
The --allow-unsafe-tags
must not with untrusted content.
The body of a Blogdown document consists of a sequence of block nodes, which in turn consist of inline nodes.
Block Nodes
Block nodes can contain any sequence of inline nodes, with the exception of code and HTML blocks, whose contents are rendered verbatim.
Block nodes can span multiple lines and are terminated by a blank line, the beginning of another type of block, or the end of the document.
The following block node types are supported:
- Paragraph: The default block type; any content not in another block is part of a paragraph.
Paragraphs must be separated by a blank line.
Can contain arbitrary inline nodes.
- Header: 1-6
#
characters at the beginning of a line begins a header, with the number of #
characters determining the header level.
Can contain arbitrary inline nodes.
- Ordered Lists: A
-
begins an ordered list item, which itself is a block.
Sequential ordered list items or sublists form an ordered list.
List items can contain arbitrary inline nodes.
- Sublists are created when list items begin with one more space than their parent list, and can be ordered or unordered.
For example, this sublist begins with
*
.
- Unordered Lists: A
*
begins an unordered list item, which itself is a block.
Sequential unordered list items or sublists form an unordered list.
List items can contain arbitrary inline nodes.
- Sublists are created when list items begin with one more space than their parent list, and can be ordered or unordered.
For example, this sublist begins with
-
.
- Blockquote: Lines beginning with
>
define a blockquote.
Can contain arbitrary inline nodes.
Note that the first line not beginning with >
will start a new block.
- Code Block: Lines indented with 4+ spaces or a tab define a code block.
Code blocks are rendered verbatim, ignoring special characters.
Note that the first un-indented line will start a new block.
- HTML Block: An HTML tag at the beginning of a line starts an HTML block.
Its contents must be valid HTML, and it is ended by the corresponding closing tag.
HTML blocks are rendered verbatim, unless HTML bleaching is enabled.
- Hard Rule: A line consisting of 3+
-
defines a hard rule.
- Table: A
|
character at the beginning of a line begins a table row, consisting of table cells separated by |
characters.
The cells are themselves blocks, and as such can contain newlines. The rows are terminated by a |
followed by a newline.
By default the table has only a body, but if rows are separated by an alternating string of +
and multiple -
characters,
e.g. +---+---+
, then every row above the separator will be in the header and every row below will be in the body.
Optionally the table may start and end with such a separator as well.
Inline Nodes
Inline nodes can generally contain a sequence of other inline nodes, but cannot contain nodes of the same type.
Despite the name, inline nodes can span multiple lines, e.g. to accommodate line length limits.
The following inline node types are supported:
- Plaintext: The default inline type; any text not in another inline node is plaintext. Rendered verbatim.
- Italic: Surrounding text with
*
italicizes it. Italic nodes can contain any other type of inline node.
- Bold: Surrounding text with
**
bolds it. Bold nodes can contain any other type of inline node.
- Code: Surrounding text with
\`` renders it as
code`. The content is rendered verbatim.
- Link: A link is written as [text](href). The text can contain any other type of inline node.
Note that the (href) portion can be omitted entirely if text is a valid absolute URI (including scheme).
The href portion is the link destination, and is parsed verbatim except that any literal
(
or )
must be escaped.
- Image: A image is written as ![alt](src). The alt and src values are parsed verbatim except that any literal
[
or ]
in the alt value and (
or )
in the src value must be escaped.
- Footnote Reference: Writing ^[footnote-name] defines a footnote reference.
It is rendered as a superscript footnote number^[footnote-numbering], and links to the footnote named footnote-name
if it is present in the footer.
A Blogdown document may optionally include a footer after the body.
The footer consists of a sequence of footnote definitions, each of which begins on a new line with ~[footnote-name] and consists of an arbitrary sequence of blocks.
A footnote definition is only terminated by another footnote definition or the end of the document.
Escaping
Any character (special or not) can be escaped with \\
. For a literal backslash, use \\\\
.
A backslash before a newline acts as a continuation character.
Planned improvements
- Better error messages on parse failures
- Windows support
- Comments
~[inline] Inlining CSS and JS is not recommended if you will be rendering multiple Blogdown documents on a single page, e.g. multiple blog posts on a blog.
Doing so will degrade network and browser performance slightly.
~[underline-parser-complexity] Supporting underlines for headers requires the parser to look-ahead arbitrarily far, resulting in quadratic time complexity.
~[arbitrary-tags] Arbitrary tags are allowed for ease of implementation,
although they are also potentially useful, e.g. for Angular support.
~[footnote-numbering] Footnotes are auto-numbered in order of appearance, starting from 0 by default (this can be changed by passing the --footnote-index-from
flag).