rmarkdown 2.9

• Fix a regression in version 2.8 when a url is used in css argument (thanks, @vnijs, #2163).

• All HTML dependencies are now correctly supported, included those with only an href component but not file component in their src attribute. Previously, rmarkdown would throw the error 'path for html_dependency not provided' when rendering documents containing HTML dependencies with href components (thanks, @crazycapivara, @matthewstrasiotto, #1805, #1948, #2151).

• Fix an error thrown with output format using a file_scope function (like in bookdown) (thanks, @rfaelens, #2149).

• Fix an issue with copy_ressource = TRUE in html_document_base where very long HTML documents were truncated during post processing (thanks, @oliviermeslin, #2145).

• When run()-ing a runtime: shiny document, an extra temp folder will be used in the output path. With the extra temp random folder in the path, predictable output file names may be used. (#2137)

• When run()-ing a runtime: shiny document with a {bslib} theme, the global theme value wasn't being restored properly. (#2160)

• Floating ToC in html_document can now hide headings with unnumbered and unlisted classes (thanks, @atusy, #1993).

• Fix prefix handling in R Markdown website's navbar for Fontawesome V5 and compatibility with V4. For icon only available in V5, the full prefix + name should be use, especially with new fab prefix (e.g. fab fa-r-project). If no prefix is used (e.g fa-home instead of fas fa-home), the fa prefix will be added for V4 compatibility as it has been deprecated in V5. We advice to use the full prefix + name for icons following Fontawesome documentation. (#1994)

• rmarkdown::site_generator() can hang session waiting for input when the site field is not found in the YAML frontmatter of index.Rmd (thanks, @kevinushey @mirh, #2043).

rmarkdown 2.8

• Fix a issue with Pandoc 2.5 and latex-div.lua - documents can now be rendered as expected without error (thanks, @DAVIDWALES, #2121).

• Fix an issue with styling and code folding button behavior when default is code-folding: show. The Button can now be correctly style according to state as aria-expanded attributes is correctly updated. Also, new classes has been added on the button to allow styling during transition: btn-collapsing and btn-expanding are respectively applied during transition Show to Hide and Hide to Show. (This follow Bootstrap behavior for the collapsible block) (thanks, @steveharoz, #2085).

• Fix an issue with citation_package having no effect when using .md file as input to render() with latex and PDF output formats (thanks, @andrewheiss, #2113).

• A new internal option rmarkdown.knit.ext has been added to control the extension of the intermediary knit output during a rendering. It defaults to md to produce *.knit.md. Only useful for very advanced usage (#2098).

• render() won't produce any *.utf8.md intermediary file anymore. This was a leftover from previous versions of rmarkdown. Since knitr 1.24 and rmarkdown 2.0, only UTF-8 input files are allowed. (#2098).

• Fix an Invalid cross-device link error when tempdir() is used for intermediates_dir in render() (thanks, @gorgitko, #2096).

• Fix a regression in HTML default template with floating toc incorrectly placed on small size window (thanks, @grimbough, #2071)

• Provided a runtime: shiny fix for output formats that pass a modified bslib::bs_theme() object to html_document_base()'s theme (thanks, @cpsievert, #2049).

• Rendering using runtime: shiny_prerendered or runtime: shinyrmd will now produce valid HTML by not inserting anymore the full document as body in the resulting shiny apps (thanks, @dakep, #1912). Header content usually containing html dependencies will be inserted in the HTML document at the end of the head before </head>, unless the rendered HTML contains <!-- HEAD_CONTENT --> special comment (see htmltools::renderDocument()). A new Pandoc variable is set in for shiny prerendered document to allow conditional insertion of such content in the the HTML template using $if(shiny-prerendered)$. This has been done in all HTML template in this package. Users of custom template should make this change to provide support for this runtime. See rmarkdown default template (default.html) for an example (#2064).

• Added tectonic as a supported LaTeX engine for generating PDF output (thanks, @dpryan79, #2078). You can specify to use this by adding engine: "tectonic" to your output format in YAML, such as pdf_document.

• When no output_format is provided in any way but an output_file is provided in render(), the default format will be determined based on the extension: "pdf_document" for .pdf, or "word_document" for .docx. Otherwise, it will be "html_document" as previous version (thanks, @pearsonca, #1569).

• Added a new global option rmarkdown.render.message. When set FALSE, render() will not output the message starting by Output created: allowing RStudio IDE to open a preview of the document. This is useful for package developers that would need to emit there own output message for there custom format. See ?render_site for more info on this special message (#2092).

• Internal changes regarding Lua filters. They have now an explicit Pandoc version minimal requirement: A filter will be skipped with a warning printed by the Lua filter if this requirement is not met. For now, all filters work for Pandoc 2.1 and above (thanks, @atusy, #2088). There is also now a new mechanism to have a share Lua filter script loadable by other Lua files: render() will set the RMARKDOWN_LUA_SHARED env var to the path of Lua filter shared.lua so that other filters can access functions defined in it using dofile(os.getenv 'RMARKDOWN_LUA_SHARED'). This is for internal usage only to avoid duplication (thanks, @tarleb, #2103).

• html_document_base gains a css argument, to which html_document's css argument is now passed. This also fix an issue when .sass or .scss files are used with this css argument when self_contained: FALSE. Moreover, sass caching mechanism can now be used when passing .sass or .scss files to the css argument (thanks, @cpsievert, #2095).

• The fig_crop option of PDF document formats (such as pdf_document and beamer_presentation) supports the value "auto" now, which means fig_crop = TRUE when figure cropping tools pdfcrop and ghostscript are available.

• The default value of the fig_crop option of PDF output formats has been changed from TRUE to "auto" (#2077).

• rmarkdown::tufte_handout has been deprecated and will be removed in the future from this package. It has been moved to the tufte package since rmarkdown 0.9.5 (released on 2016-02-22). Please use tufte::tufte_handout instead.

rmarkdown 2.7

• html_document (and html_document_base)'s theme parameter now understands bslib::bs_theme() objects/arguments, meaning that one may opt-into Bootstrap 4 and more easily create custom themes. For examples, see #1706, and for context, see https://rstudio.github.io/bslib/ (thanks, @cpsievert, #1706).

• Files with .scss/.sass extension (i.e., Sass files) provided to html_document's css parameter are now compiled to CSS using the {sass} package. Also, if theme is a {bslib} object, these Sass files may utilize Sass code inside the theme (thanks, @cpsievert, #1706).

• Fix an issue with line numbering in code chunks when .numberlines with Pandoc's highlighting (thanks, @aosavi, #1876).

• Fix an issue with shiny runtime and global.R (thanks, @liaojiahui-r, rstudio/flexdashboard#298).

• Accept latex="{options}", latex=1, or latex=true for Latex Divs.

• Add output_format_filter function to default_site_generator(). Enables custom site generators to customize or even entirely replace the output format right before rendering of each page.

• Automatically exclude renv directory for render_site() (thanks, @jmbuhr, #1996)

• Do not force options(htmltools.preserve.raw = TRUE) when this option has been set, otherwise it is impossible for other packages to turn this option off, e.g., yihui/xaringan#293.

• knitr_options_pdf() will now throw a warning when fig_crop = TRUE but is disabled because required tools pdfcrop and/or ghostscript are missing (thanks, @netique, #2016).

• Eliminated the unnecessary padding in code blocks in the html_document output with Bootstrap 4 themes (thanks, @atusy, #2019).

• github_document() will produce a working TOC even if some headers start with number (#2039).

• Fix an issue with knit_print.data.frame. The ... arguments are no more passed to print() to avoid passing knit_print() arguments options and encoding to custom print() methods (#2047).

rmarkdown 2.6

• Encoding is correctly handled now in html_vignette when checking for identical title and vignette index entry (thanks, @py-b, #1978).

• clean_site() now default to preview = TRUE and will no more remove files without notice. This change will affect the "Clean All" button in the "Build" pane for website project. clean_site(preview = FALSE) must be run to effectively remove files (#1973).

• The intermediate .tex file is now correctly deleted if keep_tex = FALSE when the R Markdown document is not rendered from the working directory (thanks, @vqv, #1308).

• Fix a bug causing certain resources files to be deleted as intermediate files when intermediates_dir is the same as the input (thanks, @bellma-lilly, #1248).

• Fix issues with anchor_sections = TRUE and learnr (thanks, @gadenbuie, #1938).

• Enable use of server.R and global.R alongside runtime: shinyrmd documents.

• pkg_file_lua() now works with devtools::load_all() and testthat when used in other packages.

• Fix pandoc_convert(citeproc = TRUE) not supressing the --natbib or --biblatex options (thanks, @atusy, #1932).

• pandoc-citeproc is now activated if a bibliography field is defined in another YAML block instead of the first YAML block (thanks, @bwiernik, #1364).

• Specify that htmltools::htmlPreserve() should use the pandoc raw attribute
  rather than preservation tokens when pandoc >= v2.0. Note that this option will
  have the intended effect only for versions of htmltools >= 0.5.0.9003.

• anchor_sections in html_documents() now defaults to FALSE. It was introduced in previous version with a default to TRUE, but it is reverted now after hearing feedbacks from the community (thank you!). The # is still used as the character for the anchor but you can easily change that using CSS rules. Examples have been added to the help page ?html_document.

• Using Pandoc's default for --email-obfuscation now. Previously, it was set to none explicitly, which is the default for Pandoc 1.17.2+ anyway. Only users with a Pandoc version before 1.17.2 may see a change in the content of the html source file produced if the document contains email addresses. This change allows to pass the Pandoc's command line flag if you want to set it to another value (thanks,@seankross, #1969).

  output:
    html_document: 
      pandoc_args: ["--email-obfuscation", "javascript"]

  See Pandoc's manual for the meaning of this option.

rmarkdown 2.5

• Tables without header rows (wich can be possible in Pandoc's simple table) are now formatted correctly when using html_document() format (thanks, @fkohrt, #1893).

• html_document gains the anchor_sections argument, which is TRUE by default, so that readers can get links to section headers easily---when you mouse over a section hader, you will see a hash symbol # at the end of the header, which contains the anchor link to this header. You can click on this link and get the URL in the addres bar of your web browser, or right-click on it and copy the URL from the context menu. The hash symbol is defined by the CSS rule a.anchor-section::before {content: '#';}. You can customize it by overriding this rule (e.g., via the css argument of html_document) and use any other symbols or icons, e.g., content: "\02AD8;" (thanks, @atusy, #1884).

• pkg_file_lua() should have thrown an error if the expected Lua file does not exist.

• Provide files_dir and intermediates_dir as attributes on return from render() when run_pandoc = FALSE.

• Supports new Pandoc 2.11 --citeproc flags usage instead of pandoc-citeproc external filter. pandoc_convert() and pandoc_citeproc_convert() will now use the correct flags according to the Pandoc version used. The logic is exported in pandoc_citeproc_args(). See Pandoc release note for more information about the new citeproc processing (#1916).

• Fixed the code highlighting when code block is hidden. Previous version introduced a regression where non default code highlighting was still shown when code_folding is activated and code block is hidden (thanks, @matthewcarlucci, #1921).

• The minimal required version for the xfun package (v0.15) has been specified for R Markdown Notebooks to work properly (thanks, @jmcphers, #1923).

• Fixed a bug in convert_ipynb() when the language is not specified in the .ipynb file (thanks, @acca3003, #1925).

• Introduce runtime: shinyrmd as a more user friendly alias for runtime: shiny_prerendered.

rmarkdown 2.4

• Lua filters handling has been improved internally with some user-facing changes (#1899):

  • New exported function pandoc_lua_filter_args() to return the Pandoc command-line argument to add a Lua filter.
  • New argument lua_filters in pandoc_options() to pass the Lua filter paths to use with a format. This allow output format authors to add filters for a custom format using the pandoc argument of output_format() and to get filters from a format using fmt$pandoc$lua_filters.
  • The Lua filters of an output format are now passed to Pandoc in render(). By default, they are passed to Pandoc before any other format-defined or user-defined Pandoc arguments (usually via the pandoc_args option of an output format). This ensures that filters of an output format are executed first. To change the default, you need to deal with it in a custom format (i.e., modify the elements in fmt$pandoc$lua_filters, such as reordering them).
  • New exported function pkg_file_lua() to get the full system path of a Lua filter included in a package source within the inst/rmarkdown/lua folder (thanks, @atusy, #1903).

• Fixed the path separators for the css parameter in YAML frontmatter for HTML output files under Windows. Previously, forward slashes in css paths were converted to backslashes (thanks, @jonathan-g, #1862).

• Since rmarkdown 1.16, Pandoc's fenced Div's are converted to LaTeX environments when the output format is LaTeX, e.g., ::: {.center data-latex=""} is converted to \begin{center}. The attribute data-latex of the Div was mandatory, even if it is empty. In rmarkdown 2.2, we silently drop this requirement, which means ::: {.center} is converted to \begin{center}. This turns out to be a bad idea, because users have no control over which Div's to be converted to LaTeX environments. Previously, they could opt-in by the data-latex attribute, but with rmarkdown 2.3, all Div's are converted to LaTeX environments unconditionally. What's more, this change led to bugs like https://stackoverflow.com/q/62340425/559676 and rstudio/bookdown#883. Therefore the data-latex attribute became mandatory again in this version. If the LaTeX environment does not need arguments, you may use data-latex="".

• The two Lua fitlers pagebreak.lua and latex-div.lua (introduced in rmarkdown 1.16) are also applied to the output format beamer_presentation now (thanks, @XiangyunHuang, #1815).

• When customizing formats with the output_format function, pre_knit, opts_hooks, and knit_hooks can now refer to rmarkdown::metadata. Previously, rmarkdown::metadata returned list() in these functions (thanks, @atusy, #1855).

• rmarkdown::find_external_resources() now discovers external template files. This in turn fixes the rendering issue of html_document with the shiny runtime and intermediate_dir set (thanks, @atusy, @cderv, #1865).

• Added the number_sections argument to following formats: github_document, ioslides_presentation, md_document, odt_document, powerpoint_presentation, rtf_document, slidy_presentation, word_document. These are powered by a Lua filter and requires Pandoc > 2.0. It will silently have no effect has before with previous pandoc version (thanks @atusy 1893). Pandoc >= 2.10.1 adds --number-sections for docx format, and thus word_document prefers the native feature to the Lua filter (thanks, @jooyoungseo, #1869).

• For the output format pdf_document, the option fig_crop will not be enabled unless both the programs pdfcrop and ghostscript are found (thanks, @dalupus, yihui/knitr#954).

• Fixed a bug that a chunk with a class fold-hide hides the rest of the chunks even the output format setting html_document(code_folding = "show") (thanks, @atusy, #1906).

• Updated documentation for render() to make it clearer how options are set for the output_format parameter (thanks, @jonathan-g, #1907 and rstudio/bookdown#930).

• Ported some CSS styles (e.g., underlines, small caps, and multi-column layouts) from the latest Pandoc's HTML template into rmarkdowns HTML templates (thanks, @atusy, #1878, #1908).

rmarkdown 2.3

• Addressed an accessibility issue in highlighted code blocks of HTML output for screen reader users: screen readers no longer read out an unnecessary code line id values (thanks, @jooyoungseo and @atusy, #1833).

• Added file_scope option to output format definition. This enables handling of duplicate numeric footnote identifiers (e.g. across bookdown chapters) via the pandoc --file-scope option (#1837).

• Added the customizable lang attribute to ioslides_presentation output (thanks, @jooyoungseo, #1841).

• Added publish_site() function for "one-button" publishing of R Markdown websites.

• When the df_print option is kable and the output format is not HTML, <div class="kable-table"> is no longer added to the kable() output, because recent versions of Pandoc will convert the div to a LaTeX environment when the output format is LaTeX (thanks, Laurens, https://stackoverflow.com/q/62340425/559676).

• html_vignette() only warns against differences in the vignette title and the vignette index entry for R >= 3.6.0 (thanks, @krlmlr, #1832).

• html_document() can apply code_folding on any chunk engines if the foldable class is added to a source code block via the chunk options (class.source or attr.source). You may apply this feature to all the source code blocks, for example, by setting knitr::opts_chunk$set(class.source = "foldable") at the beginning of your document (thanks, @atusy, #1835).

rmarkdown 2.2

• Exported the internal function find_pandoc(), and also added two arguments, dir and version, so that users can provide a custom directory under which this function may find Pandoc, as well as an expected version of Pandoc to be found (thanks, @connorp, #1785).

• pandoc_metadata_arg() is a new Pandoc helper function to generate --metadata argument for Pandoc command line (thanks, @cderv, #1789).

• The output format html_vignette() now warns against differences in the vignette title specified in the title field in the YAML metadata and the one specified inside \VignetteIndexEntry{}. Normally they are expected to be identical (#1789).

• Fixed a bug with encoding when rendering html_notebook containing HTML widgets (thanks, @cderv, #1799).

• TOC title can now be specified for html_document via the top-level option toc-title in the YAML frontmatter (thanks, @atusy, #1771).

• Floating TOC can now distinguish upper/lower-cases (thanks, @atusy, #1783).

• When code_folding='show' for the output format html_document, code blocks can be individually hidden initially by specifying the chunk option class.source='fold-hide' (thanks, @atusy, #1798).

• For LaTeX/PDF output formats pdf_document, beamer_presentation, and context_document, the argument citation_package = 'none' was deprecated, and citation_package = 'default' should be used instead if citations are to be processed by pandoc-citeproc (thanks, @njbart, rstudio/bookdown#754).

• output_format() can now inherit keep_md and clean_supporting from base_format when NULL is passed to these arguments. Previously, you must explicitly specify keep_md and/or clean_supporting as TRUE or FALSE in output_format() since they could not inherit the corresponding options of base_format. This behavior was not consisent with other arguments of output_format() (thanks, @atusy, #1823).

• The smart argument of most output formats has been removed, because Pandoc's smart extension is enabled by default, and setting smart: false for an output format did not really have any effect (which could be considered a bug, but we want to get rid of this option since it existed only for a historical reason for Pandoc 1.x, and Pandoc 2.x has been released for more than two years). If you want to disable the smart extension, you can use the option md_extensions: -smart of the output format (thanks, @atusy, #1774).

• pdf_document() should not specify the geometry variable when the documentclass variable is passed to Pandoc (thanks, @jpcirrus, #1782).

• render() now respects the YAML metadata in the R script when rendering the script with Pandoc 2.8 or later (thanks, @nsoranzo #1740, @cderv #1741).

• For pandoc_convert(), when the argument to = 'pdf', it will be changed to 'latex' internally (thanks, @JohannesFriedrich, #1802).

• render(run_pandoc = FALSE) no longer cleans up the Markdown file (typically knitted from Rmd) (thanks, @BrianDiggs, #1812).

rmarkdown 2.1

• Added the returned output from shiny::runApp() within rmarkdown::run() (thanks, @schloerke, #1760).

• YAML header is now correctly parsed in html_notebook's intermediate .knit.md file so that features like adding bibliography works again (thanks, @everdark, @cderv, #1747).

• ioslides_presentation template no longer generates an empty <h2> tag when subtitle is not specified in YAML (thanks, @jooyoungseo #1735, @cgrudz #1663).

• No longer center the #header element in the html_vignette() output (thanks, @EmilHvitfeldt, #1742).

• Ensure the tempdir() exists (via tempdir(TRUE)) when writing HTML dependencies to a temporary file, because this directory might be erased by accident (thanks, Kurt Hornik, and also @karawoo #1743).

• Added the slide_level argument to slidy_presentation() (https://stackoverflow.com/q/59157211/559676).

• Removed the jQuery dependency in html_document_base() (#1723). To avoid bugs like #1723, Pandoc 2.8 users have to upgrade to Pandoc 2.9+.

• For pdf_document, horizontal rules generated by Pandoc (before v2.8) stopped working in recent versions of TeX Live, and the same fix as the one to jgm/pandoc#5801 (i.e., hard-code \linethickness to 0.5pt) was applied in rmarkdown (thanks, @cderv, https://stackoverflow.com/a/58646915/559676).

rmarkdown 2.0

• For the output format pdf_document(), we no longer adjust the vertical spacing of the title area. This means there will be a larger spacing above the document title in PDF. If you prefers the old (smaller) spacing, please download https://github.com/rstudio/rmarkdown/blob/f6961af/inst/rmd/latex/compact-title.tex and include it to the preamble via the includes option of pdf_document. However, please note that this means you won't be able to have multiple authors in the author field of the YAML frontmatter, unless you use a custom LaTeX template. With the default LaTeX template, you will run into the error in #1716. Besides, the compact-title option in YAML is no longer supported.

• R code in the header-includes field in the YAML frontmatter stopped working in the previous version of rmarkdown. The code should be evaluated before passing to Pandoc (thanks, @mcol #1709, @cderv #1710).

• The encoding argument is no longer passed to the intermediates_generator of R Markdown output formats. The intermediates_generator function can only accept arguments input_file and intermediates_dir now (see ?rmarkdown::output_format). This is a breaking change to developers. If you are an output format developer, you have to remove the encoding argument in your intermediates_generator if your output format uses this function.

• The encoding argument is no longer passed to the render element of the site generator (see ?rmarkdown::render_site).

• The encoding argument has been removed from many of the internal functions in the rmarkdown package. Now all input files are assumed to be encoded in UTF-8. If you see an error message like "Error in FUN(arg = ...): unused argument (arg = ...)", please file an issue (with a reproducible example) to https://github.com/rstudio/rmarkdown.

• Added a new output format context_document() to support ConTeXT output (thanks, @RLesur #1725, @jooyoungseo #1713).

• render_site() can render R scripts in addition to Rmd files if you set autospin: true in _site.yml (thanks, @zeehio, #1564).

• Added ext argument to md_document(). Its default value is ".md". This argument is intended to be used together with variant argument (e.g., variant = "context" and ext = ".pdf") (thanks, @atusy, #1715).

• ioslides_presentation() stylesheet is updated for printing. Browsers are notified that the presentation should preferentially be printed in landscape orientation and without margin (thanks, @RLesur, #1718).

• Reverted the fix #1703 and applied an alternative fix to #1700, because the original fix brought a new bug #1714 (thanks, @pablobernabeu @cderv @everron @aronatkins).

• Tabsets don't work with Pandoc 2.8 (thanks, @mnazarov, #1723).

• The pdf_document format failed to work if the header-includes field in YAML is an empty list (thanks, @cderv, #1728).