Introduction to xml2rfc Version 3

Author: Henrik Levkowetz
Date: 20 Feb 2019


1   Abstract

This document is intended to give an overview of the changes to xml2rfc resulting from the introduction of new renderers to support vocabulary version 3 of the RFC and Internet-Draft XML markup language.

2   Introduction

In 1999, Marshall Rose presented an XML markup language for RFCs and Internet-Drafts in RFC 2629. He also at the same time provided a tool to convert the XML markup to the text format used for RFC and draft publication.

In December 2016, RFC 7991 presented a revised and expanded markup vocabulary based on the original RFC 2629 vocabulary. Given that the RFC 2629 vocabulary had received a number of extensions over the years, the de facto vocabulary implemented by the xml2rfc tool in 2016 was called vocabulary 'v2' for short, and the vocabulary in RFC 7991 was called 'v3'.

Recent versions of xml2rfc, starting with version 2.7.0, have gradually introduced features to support the v3 vocabulary. This document aims at giving an overview of how the new features have been assembled and made available in xml2rfc.

3   The v3 XML Schema

The v3 XML schema is a superset of the v2 schema. This means that any v2 input document is a valid document according to the v3 schema.

The original v3 specification, RFC 7991, did however specify a number of elements as being deprecated, and RFC 7998 went on to specify that the preptool, which is run before any v3 renderer, will limit elements appearing in the new RFC publication format to not include any deprecated elements.

This means that while any v2 document is valid input to the xml2rfc tool, any deprecated XML elements will be transformed to v3-specific constructs before being processed by the preptool and the v3 renderers. More about this under Syntax Changes in v3 XML Sources below.

This distinction should be clear when looking at the xml2rfc processing flow.

When speaking about v3 XML in this remainder of this document, what is meant is XML that does not use any of the deprecated v2 elements.

4   Processing Flow

Conceptually, the following diagram shows the processing flow through both legacy (v2) and new v3 processing blocks. Noteworthy is that the --v3 switch controls the path taken to reach the output renderers; it does not restrict the input file format.


4.1   Diagram Notes

The diagram above is a simplified presentation. For some of the details in the diagram, additional explanations are in order:

  • The --v3 switch is implied when other switches specific to the v3 processing path are given, for example --prep, --v2v3, and --pdf.
  • The v3 processing path is also implicitly chosen when the input XML document is declared to have version 3 by the version="3" attribute on the <xml> element.
  • The original implementation of xml2rfc permitted the generation of multiple output format files with in the same xml2rfc invocation, by specifying multiple format switches. This continues to be permitted, but the --v3 modifier switches all output from the legacy formats to the v3 formats. Output of v2 and v3 format files in the same invocation is not supported.

4.2   Format Switches

The full set of available output format switches are at this writing:

Usage: xml2rfc SOURCE [OPTIONS] ...
Example: xml2rfc draft.xml -o Draft-1.0 --text --html

  -h, --help                     show this help message and exit

    Any or all of the following output formats may be specified. The
    default is --text. The destination filename will be based on the input
    filename, unless an argument is given to --basename.

    --text                       outputs to a text file with proper page
    --html                       outputs to an html file
    --nroff                      outputs to an nroff file
    --pdf                        (unavailable due to missing external library)
    --raw                        outputs to a text file, unpaginated
    --expand                     outputs to an XML file with all references
    --v2v3                       convert vocabulary version 2 XML to version 3
    --preptool                   run preptool on the input
    --info                       generate a JSON file with anchor to section
                                 lookup information

with the following modifiers:

Format Options:
  --v3                         with --text and --html: use the v3 formatter,
                               rather than the legacy one.
  --legacy                     with --text and --html: use the legacy text
                               formatter, rather than the v3 one.

5   Syntax Changes in v3 XML Sources

First of all, please note that you can use v2 XML source files with xml2rfc, and still request v3 output formatters. In this case, xml2rfc will run the v2-to-v3 converter internally, to convert any v2 elements in your input source to the equivalent v3 constructs. Everything should work as expected, but as long as you are using v2 elements in your input file, there will be a conversion step, and you will not be in full control of the XML which is actually sent to the renderers.

If you wish to transition to v3 source files, you can always convert v2 sources explicitly to v3 by using the --v2v3 switch, and then continue working with the resulting XML file. When you do so, you need to know which constructs are no longer acceptable, and which v3-only constructs to use instead.

5.1   Replaced elements and constructs

Here are the v2 elements you should not use any more, and their replacements:

5.1.1   List elements

Don't use these to generate lists:

t           # to indicate list element, as a child of <list>

Instead, use one of the 3 new list types: <ul>1, <ol>2 and <dl>3. These map directly to the identically named HTML elements, and are used in the same way. For <ul> (unordered list) and <ol> (ordered list) use <li>4 to indicate individual list elements. This:

  <li>First item of an unordered list</li>
  <li>Second item</li>

should translate to:

  • First item of an unordered list
  • Second item

and similarly this:

  <li>First item of an ordered list</li>
  <li>Second item</li>

should translate to:

  1. First item of an ordered list
  2. Second item

and finally, for definition lists:

  <dt newline="true">what</dt>
    Definition lists associate a term with
    a definition.

should give:

Definition lists associate a term with a definition.

5.1.2   Tables

These, previously used to build tables, are deprecated:


Instead, use nested <table>5/ <tbody>6/ <tr>7 / <td>8 the same way you would in HTML:

      <td>Cell 1.1</td>       <td>Cell 1.2</td>
      <td>Cell 2.1</td>       <td>Cell 2.2</td>

in order to get:

Cell 1.1 Cell 1.2
Cell 2.1 Cell 2.2

To add table headers and footers, use <thead> and <tfoot>.

5.1.3   Text surrounding Tables and Figures

Vocabulary version 2 had special elements associated with text before or after figures and tables. These have been deprecated:


Instead, simply add regular <t> paragraphs before and/or after the figure or table.

5.1.4   Text Attributes



Instead, specific text attribute elements have been introduced:

  • <em> for emphasised text (typically rendered as slanted or italic text )
  • <strong> for boldly rendered text
  • <sub> for subscripttext
  • <sup> for superscripttext
  • <tt> for 'teletype' text (typically a generic mono-spaced text).

5.1.5   Miscellaneous

These are also deprecated:


<facsimile> has no replacement.

<vspace> has been replaced by an attribute newline="true" when used with definition lists, in order to make the definition start on a new line. For other use cases, there is no replacement. A suggestion to support <br/> generally in any inline context was vigorously opposed by some design team members.

5.2   New v3 Features and Elements

5.2.1   Artwork Alternatives

The v3 vocabulary introduces the possibility of providing multiple alternative artwork executions. Where the artwork type is different between the alternatives, this lets the renderer pick the best available alternative, as a function of the output format.

The current implementation of xml2rfc will prefer SVG type artwork over other alternatives when rendering HTML and PDF output, and will prefer ascii-art when rendering plain text output.

In order to specify a set of alternatives for a given artwork, you enclose all of the alternative executions within an <artset> element:

<artset anchor="flow-chart">
  <artwork type="svg" src="flowchart.svg"/>
  <artwork type="ascii-art" src="flowchart.txt"/>

The XML snippet above also illustrates a few other noteworthy features of the <artwork> and <artset> elements under v3:

  • The type attribute is necessary on <artwork> if xml2rfc is to select the best execution for a given output format.

  • The src attribute on <artwork> makes it easy to work with external artwork files, as produced by various drawing tools. When run through the preptool, all external content will be pulled into the final prepped file, in order to have a publication file without any external dependencies.

  • When referring to artwork from document text, any reference to a particular format out of several grouped within an <artset> is inappropriate, as there is no guarantee that one particular <artwork> entry will be used in the rendering.

    For this reason, it is best to place any anchor attribute on the <artset> element instead of on the <artwork> elements. If there are anchors on <artwork> elements within an <artset> element, and no anchor on <artset>, the preptool will promote the first <artwork> anchor to the <artset> element. Remaining anchors on enclosed <artwork> elements will be removed.

5.2.2   Section Naming

Section naming is no longer done by setting a title attribute on a <section> elements. Instead use the new <name> element, placed as the first child element of the <section>.

5.2.3   Reference Anchor Display

In v3 XML it is possible to specify an alternative display tag for a reference, using the new <displayreference> element.

This element gives a mapping between the anchor of a reference and a name that will be displayed instead.

For example, if the reference uses the anchor "RFC6949", the following would cause that anchor in the body of displayed documents to be "RFC-DEV":

<xref target="RFC6949" />


<displayreference target="RFC6949" to="RFC-DEV"/>


<reference anchor='RFC6949' target=''>
    <title>RFC Series Format Requirements and Future Development</title>

5.2.4   Reference Grouping

In order to be able to better render for instance STD references which consist of multiple individual RFCs, <referencegroup> provides a way to group references under one reference anchor:

<referencegroup anchor="STD78" target="" >
  <xi:include href=""/>
  <xi:include href=""/>
  <xi:include href=""/>
  <xi:include href=""/>

5.2.5   Unicode Content

The following describes the v3 Unicode handling as implemented, with the modifications and additions described in draft-levkowetz-xml2rfc-v3-implementation-notes. For the <u> element in particular, this goes beyond what is described in RFC 7991.

In v3, the elements <author>, <organization>, <street>, <city>, <region>, <code>, <country>, <postalLine>, <email>, and <seriesInfo> may contain non-ascii characters for the purpose of rendering author names, addresses, and reference titles correctly. They also have an additional "ascii" attribute for the purpose of proper rendering in ascii-only media.

In order to insert Unicode characters in any other context, v3 requires that the Unicode string be enclosed within an <u> element. The element will be expanded inline based on the value of its format attribute. This provides a generalised means of generating the 6 methods of Unicode renderings listed in [RFC7997], Section 3.4, and also several others found in for instance the RFC Format Tools example rendering of RFC 7700, at

The format attribute accepts either a simplified format specification, or a full format string with placeholders for the various possible Unicode expansions.

The simplified format consists of dash-separated keywords, where each keyword represents a possible expansion of the Unicode character or string; use for example <u "lit-num-name">foo</u> to expand the text to its literal value, code point values, and code point names.

A combination of up to 3 of the following keywords may be used, separated by dashes: num, lit, name, ascii, char. The keywords are expanded as follows and combined, with the second and third enclosed in parentheses (if present):

The numeric value(s) of the element text, in U+1234 notation
The Unicode name(s) of the element text
The literal element text, enclosed in quotes
The literal element text, without quotes
The provided ASCII value

In order to ensure that no specification mistakes can result for rendering methods that cannot render all Unicode code points, "num" must always be part of the specified format.

The default value of the format attribute is "lit-name-num".


   Temperature changes in the Temperature Control Protocol are
   indicated by the character U+0394 ("Δ").

   Temperature changes in the Temperature Control Protocol are
   indicated by the character U+0394 (GREEK CAPITAL LETTER DELTA).

   Temperature changes in the Temperature Control Protocol are
   indicated by the character U+0394 ("Δ").

   Temperature changes in the Temperature Control Protocol are
   indicated by the character U+0394 (GREEK CAPITAL LETTER DELTA,

   Temperature changes in the Temperature Control Protocol are
   indicated by the character GREEK CAPITAL LETTER DELTA ("Δ",

   Temperature changes in the Temperature Control Protocol are
   indicated by the character "Δ" (GREEK CAPITAL LETTER DELTA,

If the <u> element encloses a Unicode string, rather than a single code point, the rendering reflects this. The element

<u format="num-lit">ᏚᎢᎵᎬᎢᎬᏒ</u>

will be expanded to U+13DA U+13A2 U+13B5 U+13AC U+13A2 U+13AC U+13D2 ("ᏚᎢᎵᎬᎢᎬᏒ").

Unicode characters in document text which are not enclosed in <u> will be replaced with a question mark (?) and a warning will be issued.

In order to provide for cases where the simplified format above is insufficient, without relinquishing the requirement that the number of a code point always must be rendered, the format attribute can also accept a full format string. This format uses placeholders which consist of any of the key words above enclosed in curly braces; outside of this, any ascii text is permissible. For example,

The <u format="{lit} character ({num})">Δ</u>.

will be rendered as

The "Δ" character (U+0394).

[1]RFC 7991, Section 2.63: ul
[2]RFC 7991, Section 2.34: ol
[3]RFC 7991, Section 2.20: dl
[4]RFC 7991, Section 2.29: li
[5]RFC 7991, Section 2.54: table
[6]RFC 7991, Section 2.55: tbody
[7]RFC 7991, Section 2.61: tr
[8]RFC 7991, Section 2.56: td