19 minute read Published:

Table of Contents

This is version 2 of Mmark: based on a new markdown implementation and some (small) language changes as well. We think these language changes lead to a more consistent user experience and lead to less confusion.

See changes from v1 if you’re coming from version 1.

Biggest changes:

  • Including files is now done relative to the file being parsed (i.e. the sane way).
  • Block attributes apply to block elements only.
  • Callouts
    • always rendered and require double greater/less-than signs, <<1>>.
    • always require a comment in the code, i.e. //<<1>> will be rendered as a callout, a plain <<1>> will not.
  • Block Tables have been dropped.
  • Example lists (originally copied from Pandoc) have been dropped.
  • Plain citations, i.e. @RFC5412, when the reference was previously seen don’t work anymore, always use the full syntax [@RFC5412].

Why this new version?

It fixes a bunch of long standing bugs and the parser generates an abstract syntax tree (AST). It will be easier to add new renderers with this setup. It is also closer to Common Mark. So we took this opportunity to support RFC 7991 XML (xml2rfc version 3), HTML5 and manual page output. Also with code upstreamed (to gomarkdown), we have less code to maintain.

Because of the abstract syntax tree it will also be easier to write helper tools, like, for instance a tool that checks if all referenced labels in the document are actually defined. Another idea could be to write a “check-the-code” tool that syntax checks all code in code blocks. Eventually these could be build into the mmark binary itself. See some fun ideas here.

Mmark V2 Syntax

This document describes all the extra syntax elements that can be used in Mmark. Mmark’s syntax is based on the “standard” Markdown syntax. A good primer on what blackfriday implements is this article.

Read the above documents if you haven’t already, it helps you understand how markdown looks and feels.

For the rest we build up on and support all syntax it supports. We enable the following extensions by default:

  • Strikethrough, allow strike through text using ~~test~~.
  • Autolink, detect embedded URLs that are not explicitly marked.
  • Footnotes Pandoc style footnotes.
  • HeadingIDs, specify heading IDs with {#id}.
  • AutoHeadingIDs, create the heading ID from the text.
  • DefinitionLists, parse definition lists.
  • MathJax, parse MathJax
  • OrderedListStart, notice start element of ordered list.
  • Attributes, allow block level attributes.
  • Smartypants, expand -- and --- into ndash and mdashes.
  • SuperSubscript, parse super- and subscript: H~2~O is water and 2^10^ is 1024.
  • Tables, parse tables.
  • NonBlockingSpace, convert “backslash space” into a non blocking space.

Mmark adds numerous enhancements to make it suitable for writing (IETF) Internet Drafts and even complete books. It steals borrows syntax elements from pandoc, kramdown, leanpub, asciidoc, PHP markdown extra and Scholarly markdown.

What does Mmark add?

Mmark adds:

Syntax Gotchas

Because markdown is not perfect, there are some gotchas you have to be aware of:

  • Adding a caption under a quote block (Quote: ) needs a newline before it, otherwise the caption text will be detected as being part of the quote.
  • Including files (and code includes) requires an empty line before them, as they are block level elements and we need to trigger that scan from the parser.
  • Including files in lists requires a empty line to be present in the list item; otherwise Mmark will only assume inline elements and not parse the includes (which are block level elements).
  • A bibliography is only added if a {backmatter} has been specified, because we need to add just before that point.
  • Intra-work emphasis is enabled so a string like SSH_MSG_KEXECDH_REPLY is interpreted as SSH<em>MSG</em>.... You need to escape the underscores: SSH\_MSG....

RFC 7991 XML Output

This is the output format used for generating Internet-Drafts and RFCs. The generated XML needs to be processed by another tool (xml2rfc) to generate to official (final) output. The XML from mmark can be used directly to upload to the IETF tools website.

Title Block:
If the document has a title block the front matter is already open. Closing the front matter can only be done by starting the middle matter with {mainmatter}. Any open “matters” are closed when the document ends. Area defaults to “Internet” and Ipr defaults to trust200902.

Not giving a date will output <date/> which mean the current date will be applied when xml2rfc is run.

The abstract can be started by using the special header syntax .# Abstract
Any special header that is not “abstract” or “preface” will be a note: a numberless section. These notes are only allowed in the <front> section of the document. Note [sic] that notes can only contain <t> and not other block level elements, Mmark will filter these out for: blockquote currently (2020 September).
BCP 14/RFC 2119 Keywords:
If an RFC 2119 word is found enclosed in ** it will be rendered as an <bcp14> element: i.e. **MUST** becomes <bcp14>MUST</bcp14>.
Artwork is added by using a (fenced) code block. If the code block has an caption it will be wrapped in a <figure>, this is true for source code as well.
Source code:
If you want to typeset a source code instead of an artwork you must specify a language to the fenced block:
``` go

Will be typesets as source code with the language set to go.

Block Level Attributes:
We use the attributes as specified in RFC 7991, e.g. to specify an empty list style use: {empty="true"} before the list. The renderer for this output format filters unknown attributes away.
Are discarded from the final output, don’t use them.
Images are supported. We convert this to an <artwork> with src set to the image URL of path. I.e. ![alt text](img.svg "title") becomes <artwork src="img.svg" type="svg" name="title"/>. Note the first svg (the alt text) is used as the type= attribute. Also note that an image like this will be wrapped in <t> which is not allowed in RFC 7991 syntax. So to make this fully work you need to the image in a subfigure: !---. See Images in Mmark for more details.
HTML Comments are detected and discarded. These can be useful to make the parser parse certain constructs as a block element without meddling with the output.
The <br> tag is detected and converted into a hard break.
Just type the Unicode characters, the renderer takes care of putting these in between <u> tags.

HTML5 Output

Title Block:
From the title block only the title is used, in the <title> tag.

Manual Page Output

Title Block:
The title block needs a few elements to correctly generate a manual page
  • title, title needs to end in a digit to signal the section, defaults to “1” if nothing is found.
  • area, what is it, e.g. “User Commands”.
  • workgroup, who wrote this e.g. “Mmark Markdown”.
  • date, date of the man page, optional, defaults to “today”.
  • author, to add an Authors section at the end.
See Images in Mmark for details, ascii-art images from a sub-figure are included.
References and citations:
Supported, a “Bibliography” section is added. Note that unlike XML2RFC, references for IDs and RFCs are not automatically added.
Code Block:
Tabs are converted into four spaces.

Block Elements

Title Block

A Title Block contains a document’s meta data; title, authors, date and other elements. The elements that can be specified are copied from the xml2rfc v3 standard. More on these below. The complete title block is specified in TOML. Examples title blocks can be found in the repository of Mmark.

The title block itself needs three or more %’s at the start and end of the block. A minimal title block would look like this:

title = "Foo Bar"

Elements of the Title Block

An I-D needs to have a Title Block with the following items filled out:

  • title - the main title of the document.
  • abbrev - abbreviation of the title.
  • updates/obsoletes - array of integers.
  • seriesInfo, containing:
    • name - RFC, Internet-Draft, DOI, or FYI.
    • value - draft name or RFC number
    • stream - IETF (default), IAB, IRTF or independent.
    • status - standard, informational, experimental, bcp, historic, or full-standard.
  • ipr - usually just set trust200902.
  • area - usually just Internet.
  • workgroup - the workgroup the document is created for.
  • keyword - array with keywords (optional).
  • author(s) - define all the authors.
  • contact(s) - define all the contacts.
  • date - the date for this I-D/RFC.
  • language - the language for this document, this uses localized names for Index, Footnotes and References, etc. Valid values are from BCP47. This defaults to en (English). See the current list.
  • indexInclude - set to true when you want to include an index (defaults to true).

For a manual page the title, area and workgroup are mandatory, if date is not specified, “today” is assumed.

An example would be:

title = "Using Mmark to create I-Ds and RFCs"
abbrev = "mmark2rfc"
updates = [1925, 7511]
ipr= "trust200902"
area = "Internet"
workgroup = ""
keyword = ["markdown", "xml", "mmark"]

status = "informational"
name = "Internet-Draft"
value = "draft-gieben-mmark2rfc-00"
stream = "IETF"

date = 2014-12-10T00:00:00Z

fullname="R. (Miek) Gieben"
organization = "Mmark"
  email = ""
  emails = [""] # for when you need to speficy more than 1 email address

An # acts as a comment in this block. TOML itself is specified here.

If you want to define a contact do the following:

fullname="R. (Miek) Gieben"
  email = "

You can then reference this contact using a citation via the fullname: [@R. (Miek) Gieben]. This also works when referencing an author of the I-D. Note just like authors, defining contacts needs to happen in the titleblock.

To renders contacts just like the authors are rendered, they need to be a put directly after opening a new section in the first paragraph:

# Acknowledgements

[@R. (Miek) Gieben] [@More Folk]

Miek wrote ..., While More wrote ..

Special Sections

Any section that needs special handling, like an abstract or preface can be started with .# Heading. This creates a special section that is usually unnumbered.

Including Files

Including other files can done be with {{filename}}, if the path of filename is not absolute, the filename is taken relative to current file being processed. With <{{filename}} you include a file as a code block. The main difference being it will be returned as a code block. The file’s extension will be used as the language. The syntax is:


And address can be N,M, where N and M are line numbers. If M is not specified, i.e. N, it is taken that we should include the entire file starting from N.

Or you can use regular expression with: /N/,/M/, where N and M are regular expressions that specify from where to where to include lines from file.

Each of these can have an optional prefix="" specifier.


Only includes the lines 3 to (not inclusive) 5 into the current document.

{{filename}}[3,5;prefix="C: "]

will include the same lines and prefix each include line with C: .

Captioning works as well:

Figure: A sample function.

Note that because the extension of the file above is “go”, this include will lead to the following block being parsed:

~~~ go
// test.go data
Figure: A sample function.

Document Divisions

Mmark support three document divisions, front matter, main matter and the back matter. Mmark automatically starts the front matter for you if the document has a title block. Switching divisions can be done with {frontmatter}, {mainmatter} and {backmatter}. This must be the only thing on the line.

Note if there isn’t a {backmatter} the bibliography will not be inserted.


Mmark supports caption below tables, code blocks and block quotes. You can caption each elements with Table: , Figure: and Quote: respectively. The caption extends to the first empty line. Some examples:

Name    | Age
Bob     | 27
Alice   | 23
Table: This is the table caption.

Or for a code block:

 ~~~ go
 func getTrue() bool {
     return true
 Figure: This is a caption for a code block.

And for a quote:

 > Ability is nothing without opportunity.

 Quote:, Napoleon Bonaparte

A caption can potentially contain a “heading ID”: {#id} as the last text in the caption. If this is found that ID is used as the ID for the entire figure:

Name    | Age
Bob     | 27
Alice   | 23
Table: This is the table caption. {#ages}

Colspan is also supported, just repeat the pipe symbol after the cell:

Name    | Age
Bob     ||
Alice   | 23


Any text prefixed with A> will become an aside. This is similar to a block quote, but can be styled differently.

Figures and Subfigures

To group artworks and code blocks into figures, we need an extra syntax element. Scholarly markdown has a neat syntax for this. It uses a special section syntax and all images in that section become subfigures of a larger figure. Disadvantage of this syntax is that it can not be used in lists. We use a fenced code block like syntax: !--- as the opening and closing “tag”. Note: only inline elements are parsed inside a figure block.

Basic usage:

![Alt text](/path/to/img.jpg "Optional title")

if the figure block has a caption that will be used as well:

![Alt text](/path/to/img.jpg "Optional title")
![Alt2 text](/path/to/img2.jpg "Optional title2")
Figure: this is a figure containing subfigures.

Or when just using fenced code blocks:

~~~ ascii-art
| ART |
Figure: Caption for this ascii-art

~~~ c
printf("%s\n", "hello");
Figure: Caption for both figures.

Block Level Attributes

A “Block Level Attribute” is a list of HTML attributes between braces: {...}. It allows you to set classes, an anchor and other types of extra information for the next block level element.

The full syntax is: {#id .class key="value"}. Values may be omitted, i.e., just {.class} is valid.

The following example applies the attributes: title and anchor to the blockquote:

{title="The blockquote" #myid}
> A blockquote with a title

Gets expanded into:

<blockquote anchor="myid" title="The blockquote">
    <t>A blockquote with a title</t>


Text that is separated from the rest of the content with empty lines.


Tables can be entered by using a simple syntax:

Name    | Age
Bob     | 27
Alice   | 23

Table footers are supported as well and can be added with equal signs (=):

Name    | Age
Bob     | 27
Alice   | 23
Total   | 50

The pipe symbol (|) to mark columns does not need to be aligned. Each row must be on a single line.

Headerless tables are also supported, just leave of the first line.


Lists are the normal markdown lists, but we track how they are typeset, for ordered list the delimiter can be either . or ). When a parenthesis is used the type is set to %d). Note that any block level attributes take precedence.

Newlines between list items will create a non-compact list, i.e. compare:

1. Item
2. Item


1. Item

2. Item

This is true for all types of lists.

Inline Elements


Defining indices allows you to create an index. The define an index use the (!item). Sub items can be added as well, with (!item, subitem). To make item primary, use another !: (!!item, subitem). If any index is defined the end of the document contains the list of indices. The -index=false flag suppresses this generation.

An index may apply to an entire section. This can be entered (just like contacts) by having an index (or multiple), and just the index, to be the first paragraph after a new section.


Mmark uses the citation syntax from Pandoc: [@RFC2535], the citation can either be informative (default) or normative, this can be indicated by using the ? or ! modifier: [@!RFC2535] create a normative reference for RFC 2535. To suppress a citation use [@-RFC1000]. It will still add the citation to the references, but does not show up in the document as a citation.

The first seen modifier determines the type (suppressed, normative or informative). Multiple citation can separated with a semicolon: [@RFC1034;@RFC1035].

If you reference an RFC, I-D, BCP or STD or W3C document the reference will be added automatically (no need to muck about with an <reference> block). This is to say:

Any reference starting with RFC, BCP, STD, I-D. or W3C. will be automatically added to the correct reference section.

Referring to a specific RFC of an STD/BCP can be done using the following syntax: [@RFCxxx@STDyy], this expands <xref target="RFCxxx"></xref> of <xref target="STDxx"></xref> where both will be added to the automatically generated references. The word “of” will be translated according to the document language. Note: no attempt is made to validate if the RFC is actually part of the STD or BCP, i.e. [@RFCxxx@RFCyyy] will be happily accepted.

For I-Ds you may want to add a draft sequence number, which can be done as such: [@?I-D.blah#06]. If you reference an I-D without a sequence number it will create a reference to the last I-D in citation index. I.e. a draft named “draft-gieben-pandoc2rfc”, the I-D reference becomes: I-D.gieben-pandoc2rfc. Referencing multiple versions of the same I-D in a document will lead to validation errors when running xml2rfc.

A bibliography section is created by default if a {backmatter} is given, but you can suppress it by using the command line flag -bibliography=false. No {backmatter}, no bibliography.

A non-suppressed reference to the full name of an author or contact will insert the referenced person as a contact. See

Reference Text Suffices

You can specify extra text after the citation using a comma: [@RFC2535, section 5], see This is used in the following manner:

  • [@RFC2535, section 5] -> sectionFormat=“of”
  • [@RFC2525, see, section 5] -> sectionFormat=“comma”
  • [@RFC2525, (see) section 5] -> sectionFormat=“parens”
  • [@RFC2525, 5] -> sectionFormat=“bare”

page, paragraph, etc., might be supported in the future if these pop up in XML2RFC. Translation of these strings is supported for a few languages, zie, sectie 5 (Dutch) is supported for instance.

Also note these strings need to be literary typed as shown here (we may become more lenient in the future).

XML References

Any valid XML reference fragment found anywhere in the document, can be used as a citation reference. The syntax of the XML reference element is defined in RFC 7749. The anchor defined can be used in the citation, which the example below that would be [@pandoc]:

<reference anchor='pandoc' target=''>
        <title>Pandoc, a universal document converter</title>
        <author initials='J.' surname='MacFarlane' fullname='John MacFarlane'>
            <organization>University of California, Berkeley</organization>
        <date year='2006'/>

Note that for citing I-Ds and RFCs you don’t need to include any XML, as Mmark will pull these automatically from their online location: or technically more correct: the xml2rfc post processor will do this.

The newer referencegroup is also supported. No attempt to parse it is made, it’s detected and included in the bibliography.

Cross References

Cross references can use the syntax [](#id), but usually the need for the title within the brackets is not needed, so Mmark has the shorter syntax (#id) to cross reference in the document.


My header {#header}

Lorem ipsum dolor sit amet, at ultricies ...
See Section (#header).

Using Block Level Attributes this also works for tables and figures (including artwork):

 a | b
 d | d



And then reference the same (#myid), the formatter (xml2rfc) will do the right thing.

Cross Reference Text Suffixes

Just like , you can add a suffix text to a reference, to influence how xml2rfc will render it, see it will allow you to set the format attribute. The following is supported:

  • counter

  • title

  • default

  • (#myid, use counter) -> format=“counter”

  • (#myid, use title) -> format=“title”

Translation of these strings is supported for a few languages, (#myid, gebruik titel) (Dutch) is supported for instance.

Also note these strings need to be literary typed as shown here (we may become more lenient in the future).

Super- and Subscript

For superscript use ^ and for subscripts use ~. For example:

H~2~O is a liquid. 2^10^ is 1024.

Inside a super- or subscript you must escape spaces. Thus, if you want the letter P with ‘a cat’ in subscripts, use P~a\ cat~, not P~a cat~.


Callouts are way to reference code from paragraphs following that code. Mmark uses the following syntax for specifying a callout <<N>> where N is integer > 0.

In code blocks you can use the same syntax to create a callout:

    Code  //<<1>>
    More  //<<2>>

As you can see in <<1>> but not in <<2>>. There is no <<3>>.

Using callouts in source code examples will lead to code examples that do not compile. To fix this the callout needs to be placed in a comment, but then your source show useless empty comments. To fix this Mmark will detect (and remove!) the comment from the callout, leaving your example pristine in the document.

Note that callouts in code blocks are only detected if the renderer has been configured to look for them. The default mmark configuration is to detect them after // and # comment starters.

Lone callouts (in code blocks) without them being prefixed with a comment means they are not detected by Mmark.


Phrases that are defined in RFC 2119 (i.e. MUST, SHOULD, etc) are detected when being type set as strong elements: **MUST**, in the RFC 7991 output these will typeset as <bcp14>MUST</bcp14>.

Changes from version 1

These are the changes from Mmark version 1:

  • Citations:
    • Suppressing a citation is done with [@-ref] (it was the reverse -@ in v1), this is more consistent.
    • Multiple citations are allowed in one go, separated with a semicolons: [@ref1; @ref2].
    • A reference text suffix is allowed [@ref, section 23], the separation character is a comma; this mirrors the pandoc syntax.
  • Indices: now just done with (!item), marking one primary will be: (!!item).
  • Code block callouts are now a renderer setting, not a Block Level Attribute. Callout in code are only detected if they are used after a comment.
  • Including files with a prefix is now specified in the address specification: {{myfile}}[prefix="C: "] will use C: as the prefix. No more mucking about with block attribute lists that are hard to discover.
  • There no extended table syntax; if this ever comes back it needs to more robust implementation.
  • Title Block need to be sandwiched between %%%, the prefix % does not work anymore.

Syntax that is not supported anymore:

  • HTML abbreviations.
  • The different list syntaxes have been dropped, use a Block Level Attribute to tweak the output.
  • Tasks lists and example lists.
  • Comment detection, i.e. to support cref: dropped. Comments are copied depending on the output renderer.
  • Parts
  • Extended table syntax.