pandoc-crossref: Pandoc filter for cross-references

[ gpl, library, program, text ] [ Propose Tags ]

pandoc-crossref is a pandoc filter for numbering figures, equations, tables and cross-references to them.

[Skip to Readme]




Note: This package has metadata revisions in the cabal description newer than included in the tarball. To unpack the package including the revisions, use 'cabal get'.

Maintainer's Corner

Package maintainers

For package maintainers and hackage trustees


Versions [RSS],,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, (info)
Dependencies base (>=4.8 && <5), bytestring (>=0.9 && <0.11), containers (>=0.1 && <0.6), data-accessor (>= && <, data-accessor-template (>= && <, data-accessor-transformers (>= && <, data-default (>=0.4 && <0.8), mtl (>=1.1 && <2.3), pandoc (>=1.17.1 && <1.19), pandoc-crossref, pandoc-types (>=1.16 && <1.17.5), roman-numerals (>=0.5 && <0.6), syb (>=0.4 && <0.7), template-haskell (>= && <, utility-ht (>=0.0.11 && <0.1.0), yaml (>=0.8 && <0.9) [details]
License GPL-2.0-only
Author Nikolay Yakimov
Revised Revision 1 made by lierdakil at 2018-06-12T14:16:04Z
Category Text
Source repo head: git clone
this: git clone v0.2.4.1)
Uploaded by lierdakil at 2016-11-23T09:05:24Z
Distributions Arch:, NixOS:
Reverse Dependencies 1 direct, 0 indirect [details]
Executables pandoc-crossref
Downloads 48280 total (308 in the last 30 days)
Rating (no votes yet) [estimated by Bayesian average]
Your Rating
  • λ
  • λ
  • λ
Status Docs available [build log]
Last success reported on 2016-11-23 [all 1 reports]

Readme for pandoc-crossref-

[back to package description]

pandoc-crossref filter Build Status Build status

pandoc-crossref is a pandoc filter for numbering figures, equations, tables and cross-references to them.

Input file (like can be converted into html, latex, pdf, md or other formats.

Optionally, you can use cleveref for latex/pdf output, e.g. cleveref pdf, cleveref latex, and listings package, e.g. listings pdf, listings latex

You can also enable per-chapter numbering (as with --chapters for latex output). You need to specify -M chapters for non-latex/pdf output however. Examples: html, markdown, latex, pdf.

This work is inspired by pandoc-fignos and pandoc-eqnos by @tomduck.

This package tries to use latex labels and references if output type is LaTeX. It also tries to supplement rudimentary LaTeX configuration that should mimic metadata configuration by setting header-includes variable.


LaTeX output and --include-in-header

pandoc-crossref uses metadata variable header-includes to add LaTeX definitions to output. However, Pandoc's command line option --include-in-header/-H overrides this variable. If you need to use --include-in-header, add pandoc-crossref-specific definitions as well. See LaTeX customization for more information.

pandoc-citeproc and pandoc-crossref

Since pandoc-crossref uses the same citation syntax as pandoc-citeproc, you have to run former before latter. For example:

pandoc -F pandoc-crossref -F pandoc-citeproc -o file.html

Note on leading/trailing spaces in metadata options

Leading and trailing spaces in YAML metadata will most likely be stripped by either YAML parser or Pandoc itself. If you need leading and/or trailing spaces in pandoc-crossref metadata variables, use html entity for space instead, i.e. &#32;. For example, if you want reference ranges to be delimited by a dash with spaces (e.g. 2 - 5), include the following in YAML metadata:

rangeDelim: '&#32;-&#32;'

or pass -MrangeDelim='&#32;-&#32;' to pandoc on command line.

You can use other html entites of course, like &nbsp; etc.


Syntax is loosely based on discussion in

Image labels


To label an (implicit) figure, append {#fig:label} (with label being something unique to reference this figure by) immediately after image definition.

This only works on implicit figures, i.e. an image occurring by itself in a paragraph (which will be rendered as a figure with caption by pandoc)

Image block and label can not be separated by spaces.


It's possible to group figures as subfigures. Basic syntax is as follows:

<div id="fig:figureRef">
![subfigure 1 caption](image1.png){#fig:figureRefA}

![subfigure 2 caption](image2.png){#fig:figureRefB}

Caption of figure

To sum up, subfigures are made with a div having a figure id. Contents of said div consist of several paragraphs. All but last paragraphs contain one subfigure each, with captions, images and (optionally) reference attributes. Last paragraph contains figure caption.

If you put more than one figure in the paragraph, those will still be rendered, but Pandoc will omit subfigure caption in most outputs (but it will work as expected with LaTeX). You can use output-specific hacks to work around that, or use subfigGrid (see below).

Output is customizable, with metadata fields. See Customization for more information.

Default settings will produce the following equivalent Markdown from example above:

<div id="fig:figureRef" class="subfigures">



Figure 1: Caption of figure. a — subfigure 1 caption, b — subfigure 2


References to subfigures will be rendered as figureNumber (subfigureNumber), e.g., in this particular example, [@fig:figureRefA] will produce fig. 1 (a).

You can add nocaption class to an image to suppress subfigure caption altogether. Note that it will still be counted.

Subfigure grid

If you need to align subfigures in a grid, and using output format styles is not an option, you can use subfigGrid option. That will typeset subfigures inside a table.

Rows are formed by different paragraphs, with each image in a separate column.

Column widths will be taken from width attributes of corresponding images, e.g.

<div id="fig:coolFig">
![caption a](coolfiga.png){#fig:cfa width=30%}
![caption b](coolfigb.png){#fig:cfb width=60%}
![caption c](coolfigb.png){#fig:cfc width=10%}

![caption d](coolfigd.png){#fig:cfd}
![caption e](coolfige.png){#fig:cfe}
![caption f](coolfigf.png){#fig:cff}

Cool figure!

will produce a table with columns of 30%, 60% and 10% respectively.

Only first row of images is considered for table width computation, other rows are completely ignored.

Anything except images is silently ignored. So any text, spaces, soft line breaks etc will silently disappear from output. That doesn't apply to caption paragraph, obviously.

All images will have width attribute automatically set to 100% in order to fill whole column.

Specifying width in anything but % will throw an error.

If width for some images in first row is not specified, those will span equally in the remaining space.

If width isn't specified for any image in first row, those will span equally on 99% of page width (due to Pandoc otherwise omitting width attribute for table).

This option is ignored with LaTeX output, but paragraph breaks should produce similar effect, so images should be typeset correctly. TL;DR you don't need subfigGrid enabled for it to work with LaTeX, but you can still enable it.

Equation labels

$$ math $$ {#eq:label}

To label a display equation, append {#eq:label} (with label being something unique to reference this equation by) immediately after math block.

Math block and label can be separated by one or more spaces.

You can also number all display equations with autoEqnLabels metadata setting (see below). Note, however, that you won't be able to reference equations without explicit labels.

Equations numbers will be typeset inside math with \qquad before them. If you want to use tables instead, use tableEqns option. Depending on output format, tables might work better or worse than \qquad.

Table labels

a   b   c
--- --- ---
1   2   3
4   5   6

: Caption {#tbl:label}

To label a table, append {#tbl:label} at the end of table caption (with label being something unique to reference this table by). Caption and label must be separated by at least one space.

Section labels

You can also reference sections of any level. Section labels use native pandoc syntax, but must start with "sec:", e.g.

# Section {#sec:section}

You can also use autoSectionLabels variable to automatically prepend all section labels (automatically generated with pandoc included) with "sec:". Bear in mind that references can't contain periods, commas etc, so some auto-generated labels will still be unusable.

WARNING: With LaTeX output, you have to invoke pandoc with --number-sections, otherwise section labels won't work. It's also advised with other output formats, since with no numbers in section titles, it would be hard to navigate anyway.

Section numbering

Pandoc doesn't properly support numbering sections in some output formats, and section reference labels (see below).

You can let pandoc-crossref handle section numbering instedad. This is done via numberSections and sectionsDepth metadata options.

numberSections controls if pandoc-crossref handles numbering sections, while sectionsDepth controls what sections are numbered.

Set sectionsDepth to 0 to make section numbering consistent with chaptersDepth.

If sectionsDepth value is lesser than 0, all sections will be numbered.

Otherwise, only header levels up to and including sectionsDepth will be numbered.

Section reference labels

Not currently supported with LaTeX output

If you want to reference some section by a pre-defined label instead of by number, you can specify section attribute label, like this:

# Section {label="Custom Label"}

This label will be used instead of section number in chapters output and when referencing section directly (with @sec:section).

Note that with chapters output with depth>1, only given section will be referenced by custom label, e.g. with

# Chapter 1.

## Section with custom label {#sec:scl label="SCL"}


@sec:scl will translate into sec. 1.SCL, and @fig:figure into fig. 1.SCL.1

Code Block labels

There are a couple options to add code block labels. Those work only if code block id starts with lst:, e.g. {#lst:label}

caption attribute

caption attribute will be treated as code block caption. If code block has both id and caption attributes, it will be treated as numbered code block.

```{#lst:code .haskell caption="Listing caption"}
main :: IO ()
main = putStrLn "Hello World!"

Table-style captions

Enabled with codeBlockCaptions metadata option. If code block is immediately adjacent to paragraph, starting with Listing: or : , said paragraph will be treated as code block caption.

Listing: Listing caption

```{#lst:code .haskell}
main :: IO ()
main = putStrLn "Hello World!"


```{#lst:code .haskell}
main :: IO ()
main = putStrLn "Hello World!"

: Listing caption

Wrapping div

Wrapping code block without label in a div with id lst:... and class, starting with listing, and adding paragraph before code block, but inside div, will treat said paragraph as code block caption.

<div id="lst:code" class="listing">
Listing caption
main :: IO ()
main = putStrLn "Hello World!"


[@fig:label1;@fig:label2;...] or [@eq:label1;@eq:label2;...] or [@tbl:label1;@tbl:label2;...] or @fig:label or @eq:label or @tbl:label

Reference syntax heavily relies on citation syntax. Basic reference is created by writing @, then basically desired label with prefix. It is also possible to reference a group of objects, by putting them into brackets with ; as separator. Similar objects will be grouped in order of them appearing in citation brackets, and sequential reference numbers will be shortened, e.g. 1,2,3 will be shortened to 1-3.

You can capitalize first reference character to get capitalized prefix, e.g. [@Fig:label1] will produce Fig. ... by default. Capitalized prefixes are derived automatically by capitalizing first letter of every word in non-capitalized prefix, unless overriden with metadata settings. See Customization for more information.

Linking references

To make references into hyperlinks to referenced element, enable linkReferences metadata option. This has no effect on LaTeX output, since in this case, hyperlinking references is handled with hyperref LaTeX package.

Custom prefix per-reference

It's possible to provide your own prefix per-reference, f.ex. [Prefix @reference] will replace default prefix (fig./sec./etc) with prefix verbatim, e.g. [Prefix @fig:1] will be rendered as Prefix 1 instead of fig. 1.

In citation group, citations with the same prefix will be grouped. So, for example [A @fig:1; A @fig:2; B @fig:3] will turn into A 1, 2, B 3. It can be used to an advantage, although it's a bit more cumbersome than it should be, e.g. [Appendices @sec:A1; Appendices @sec:A2; Appendices @sec:A3] will turn into Appendices @A1-@A3 (with @A1 and @A3 being relevant section numbers). Note that non-contiguous sequences of identical prefixes will not be grouped.

Not supported with cleferef LaTeX output.

Prefix suppression

Prepending - before @, like so [-@citation], will suppress default prefix, e.g. [-@fig:1] will produce just 1 (or whatever number it happens to be) without fig. prefix.

In citation group, citations with and without prefixes will be in different groups. So [-@fig:1; @fig:2; -@fig:3] will be rendered as 1, fig. 2, 3, so be careful with this feature. Again, non-contiguous sequences are not grouped together.


It's possible to use raw latex commands \listoffigures, \listoftables and listoflistings, which will produce ordered list of figure/table/listings titles, in order of appearance in document.

\listoflistings depends on other options, and is defined in preamble, so it will work reliably only with standalone/pdf output.


Assuming you already installed Haskell platform, you can install pandoc-crossref with cabal:

cabal update
cabal install pandoc-crossref

However, I highly recommend you use a sandbox for installation, e.g.

cabal update
mkdir pandoc-crossref
cd pandoc-crossref
cabal sandbox init
cabal install pandoc-crossref

This will get pandoc-crossref installed into .cabal-sandbox/bin. Pandoc will also be built, if it's not installed as a Haskell library system-wide. You might also want to install pandoc-citeproc in the same sandbox, if that's the case (cabal install pandoc-citeproc).

For Windows users, there is a pre-built executable available at releases page. Bear in mind that it is a product of an automated build script, and as such, provided as-is, with zero guarantees.


Run pandoc with --filter option, passing path to pandoc-crossref executable, or simply pandoc-crossref, if it's in PATH:

pandoc --filter pandoc-crossref

If you installed with cabal, it's most likely located in $HOME/.cabal/bin on *NIX systems, $HOME/Library/Haskell/bin on Macs, or in %AppData%\cabal\bin on Windows.


There are several parameters that can be set via YAML metadata (either by passing -M to pandoc, or by setting it in source markdown)

A list of variables follows.

General options

  • cref: if True, latex export will use \cref from cleveref package. Only relevant for LaTeX output. \usepackage{cleveref} will be automatically added to header-includes.
  • chapters: if True, number elements as chapter.item, and restart item on each first-level heading (as --chapters for latex/pdf output)
  • chaptersDepth, default 1: header level to treat as "chapter". If chaptersDepth>1, then items will be prefixed with several numbers, corresponding to header numbers, e.g. fig. 1.4.3.
  • numberSections, default false: if True, pandoc-crossref will prepend section number to section titles (as counted by pandoc-crossref itself).
  • sectionsDepth, default 0:
    • sectionsDepth < 0 -- number all sections
    • sectionsDepth == 0 -- be consistent with chaptersDepths
    • sectionsDepth > 0 -- number section levels up to and including sectionsDepth
  • listings: if True, generate code blocks for listings package. Only relevant for LaTeX output. \usepackage{listings} will be automatically added to header-includes. You need to specify --listings option as well.
  • codeBlockCaptions: if True, parse table-style code block captions.
  • autoSectionLabels, default false: Automatically prefix all section labels with sec:. Note that this messes with pandoc's automatic header references.
  • autoEqnLabels, default false: Automatically number all display equations (i.e. ones defined using $$...$$/\[...\]). Note that you won't be able to reference equations without explicit labels.
  • tableEqns, default false: Typeset equations and equation numbers in tables instead of embedding numbers into equations themselves. Depending on output format, this might work better or worse.

Item title format

  • figureTitle, default Figure: Word(s) to prepend to figure titles, e.g. Figure 1: Description
  • tableTitle, default Table: Word(s) to prepend to table titles, e.g. Table 1: Description
  • listingTitle, default Listing: Word(s) to prepend to listing titles, e.g. Listing 1: Description
  • titleDelim, default :: What to put between object number and caption text.

See Subfigures

  • ccsDelim, default ,&nbsp;: delimiter for collected subfigure captions. See Subfigures and Templates
  • ccsLabelSep, default &nbsp;—&nbsp;: delimiter used between subfigure label and subfigure caption in collected captions. See Subfigures and Templates
  • subfigGrid, default false. If true, typeset subfigures inside a table. Ignored with LaTeX output. See Subfigures

List titles

  • lofTitle, default # List of Figures: Title for list of figures (lof)
  • lotTitle, default # List of Tables: Title for list of tables (lot)
  • lolTitle, default # List of Listings: Title for list of listings (lol)

Reference format

  • figPrefix, default fig., figs.: Prefix for references to figures, e.g. figs. 1-3
  • eqnPrefix, default eq., eqns.: Prefix for references to equations, e.g. eqns. 3,4
  • tblPrefix, default tbl., tbls.: Prefix for references to tables, e.g. tbl. 2
  • lstPrefix, default lst., lsts.: Prefix for references to lists, e.g. lsts. 2,5
  • secPrefix, default sec., secs.: Prefix for references to sections, e.g. secs. 2,5
  • chapDelim, default .: Delimiter between chapter number and item number.
  • rangeDelim, default -: Delimiter between reference ranges, e.g. eq. 2-5
  • linkReferences, default false: Make references hyperlinks to the referenced element

figPrefix, eqnPrefix, tblPrefix, lstPrefix can be YAML arrays. That way, value at index corresponds to total number of references in group, f.ex.

  - "fig."
  - "figs."

Will result in all single-value references prefixed with "fig.", and all reference groups of two and more will be prefixed with "figs.":

[@fig:one] -> fig. 1
[@fig:one; @fig:two] -> figs. 1, 2
[@fig:one; @fig:two; @fig:three] -> figs. 1-3

They can be YAML strings as well. In that case, prefix would be the same regardless of number of references.

They can also be used with first character capitalized, i.e. FigPrefix, etc. In this case, these settings will override default reference capitailzation settings.

Custom numbering

See Custom Numbering Schemes

  • figLabels, default unset: custom numbering scheme for figures.
  • subfigLabels, default alpha a: custom numbering scheme for subfigures.
  • eqnLabels, default unset: custom numbering scheme for equations.
  • tblLabels, default unset: custom numbering scheme for tables.
  • lstLabels, default unset: custom numbering scheme for listings.
  • secLabels, default unset: custom numbering scheme for sections.

Item title templates

See Templates

  • figureTemplate, default \\[figureTitle\\] \\[i\\]\\[titleDelim\\] \\[t\\]: template for figure captions
  • tableTemplate, default \\[tableTitle\\] \\[i\\]\\[titleDelim\\] \\[t\\]: template for table captions
  • listingTemplate, default \\[listingTitle\\] \\[i\\]\\[titleDelim\\] \\[t\\]: template for listing captions
Subfigure templates

See Subfigures

  • subfigureTemplate, default \\[figureTitle\\] \\[i\\]\\[titleDelim\\] \\[t\\]. \\[ccs\\]: template for subfigure divs captions.
  • subfigureChildTemplate, default \\[i\\]: template for actual subfigure captions.
  • ccsTemplate, default \\[i\\]\\[ccsLabelSep\\]\\[t\\]: template for collected subfigure captions.

Reference templates

See Templates

  • figPrefixTemplate, defualt \\[p\\]&nbsp;\\[i\\] -- figure reference template
  • eqnPrefixTemplate, defualt \\[p\\]&nbsp;\\[i\\] -- equation reference template
  • tblPrefixTemplate, defualt \\[p\\]&nbsp;\\[i\\] -- table reference template
  • lstPrefixTemplate, defualt \\[p\\]&nbsp;\\[i\\] -- listing reference template
  • secPrefixTemplate, defualt \\[p\\]&nbsp;\\[i\\] -- section reference template
  • refIndexTemplate, default \\[i\\]\\[suf\\] -- individual reference index template
  • subfigureRefIndexTemplate, default \\[i\\]\\[suf\\] (\\[s\\]) -- subfigure reference index template

LaTeX customization

Support for above variables with LaTeX/PDF output is limited. In particular, the following variables are honored:

  • figureTitle
  • tableTitle
  • listingTitle
  • lofTitle -- ignores formatting
  • lotTitle -- ignores formatting
  • lolTitle -- ignores formatting
  • *Prefix, upper-/lowercase and single/plural form. Note that with cleveref output, if *Prefix is array, only first two items are used, and the rest is ignored.

Templates are not supported.

You can add arbitrary LaTeX commands to document header, however, using header-includes metadata field. Please bear in mind, that pandoc-crossref up to and including requires header-includes to be YAML array, e.g.

    - "\\newcommand{\\pcdoc}{Pandoc-crossref documentation}"

This will be added before any customization applied by pandoc-crossref. For a complete list of what is added to template, consult ModifyMeta.hs.


pandoc-crossref supports advanced caption customization via caption templates. Templates are specified as YAML metadata variables (see Customization), and are parsed as default Pandoc Markdown. Variables are specified with display math syntax, i.e. $$var$$ in a template will be replaced with value of variable var. Variables can be specified in YAML metadata block, or from command line (with -M switch). There are two special variables, that are set internally:

  • i -- object number, possibly with chapter number (if chapter=True)
  • t -- object caption, as given in source Markdown

Also there is a number of specific variables that are meaningful only in certain contexts:

  • ccs -- collected subfigure captions. Only applicable to subfigureTemplate. Collected captions will be separated by ccsDelim and individual captions will be printed with ccsTemplate. See Subfigures
  • suf -- reference suffix, applicable to refIndexTemplate, subfigureRefIndexTemplate
  • s -- subfigure index, applicable to subfigureRefIndexTemplate

xPrefixTemplate, where x is fig, eqn, etc, are a special case. Those don't have t variable, since there is no caption in source markdown, but instead have p variable, that binds to relevant xPrefix. This is done this way, since actual prefix vaule can depend on i.

Please note that at the moment, templates are not supported with LaTeX/PDF output.

Custom Numbering Schemes

It's possible to use other numbering schemes apart from arabic. This is controlled by several metadata options, consult Customization for a list. Possible values are:

  • arabic -- arabic numbers (1, 2, 3 ...)

  • roman -- roman numbers (I, II, III, IV, ...)

  • alpha x, where x is first letter to start from. This will work for any letter, but will use UTF-8 codepage to determine what's next, so using something strange is not advised. For example, you can safely use alpha a to get lowercase latin letters for 26 figures. After that, it will get weird (since basic latin alphabet has 26 letters). Specifically, it will go into characters space ({, |, }, etc). You can consult for general idea on letter progression.

  • list of strings. You can define a YAML array for numbers. Mapping is 1:1. For example,

    figLabels: [α, β, γ, 1, 2, 3, I, II, III]

    will give first figure label 'α', second -- 'β', etc.

    Note that it will repeat last item in list indefinitely if there are more images than items in list.

Settings file

It is also possible to set variables used by pandoc-crossref with a separate YAML file. If a given variable is not set in metadata, then pandoc-crossref will attempt to read it from file specified by crossrefYaml metadata variable, or, if not set, from pandoc-crossref.yaml from current working directory. This allows for reusable configurations. One possible application is ad-hoc internationalization.

For example, consider $HOME/misc/pandoc-crossref-es.yaml:

figureTitle: "Figura"
tableTitle: "Tabla"
figPrefix: "fig."
eqnPrefix: "ec."
tblPrefix: "tbl."
loftitle: "# Lista de figuras"
lotTitle: "# Lista de tablas"

pandoc-crossref will send this data to pandoc wrapped in lines of ---. The YAML file's first line should specify a variable; it will not pass the variables if it is --- or a blank line.

One could use this with pandoc-crossref as follows:

pandoc -F pandoc-crossref.hs -M "crossrefYaml=$HOME/misc/pandoc-crossref-es.yaml"


This software is licensed under GNU GPL 2. See for details.

© 2016 Nikolay Yakimov et al

Contributors (per GPL, holders of copyright on their respective contributions):

  • Nikolay Yakimov
  • Wlad
  • scoavoux
  • Matthew Salganik
  • Hadrien Mary
  • Felix Yan
  • Chris Black
  • Bart Mesuere