The goal of downlit is to provide syntax highlighting and automatic linking of R code in a way that is easily used from RMarkdown packages like pkgdown, bookdown, and hugodown.
Install downlit from CRAN with:
install.packages("downlit")
downlit has two slightly different highlighting/linking engines:
highlight()
works with multiline code blocks and does syntax highlighting,
function linking, and comment styling.autolink()
works with inline code and only does linking.Multiline code blocks have:
The following forms of inline code are recognized and automatically linked:
fun()
, pkg::fun()
.?fun
, pkg::fun
, type?topic
.help("fun")
, help("fun", package = "package")
, help(package = "package")
.vignette("name")
, vignette("name", package = "package")
.library(package)
, require(package)
, requireNamespace("package")
.{package}
gets linked (if possible) and formatted as plain text. If downlit can find a pkgdown site for the remote package, it will link to it; otherwise it will link to https://rdrr.io/ for documentation, and CRAN for vignettes. In order for a pkgdown site to be findable, it needs to be listed in two places:
In the URL
field in the DESCRIPTION
, as in
dplyr:
URL: https://dplyr.tidyverse.org, https://github.com/tidyverse/dplyr
In the url
field in _pkgdown.yml
, as in
dplyr
url: https://dplyr.tidyverse.org
When this field is defined, pkgdown generates a public facing
pkgdown.yml
file that
provides metadata about the site:
pandoc: '2.2'
pkgdown: 1.3.0
pkgdown_sha: ~
articles:
compatibility: compatibility.html
dplyr: dplyr.html
urls:
reference: https://dplyr.tidyverse.org/reference
article: https://dplyr.tidyverse.org/articles
So when you build a pkgdown site that links to the dplyr documentation (e.g., dplyr::mutate()
), pkgdown looks first in dplyr's DESCRIPTION
to find its website, then it looks for pkgdown.yml
, and uses the metadata to generate the correct links.
downlit is designed to be used by other packages, and I expect most uses of downlit will use it via another package (e.g. hugodown). If you want to use it in your own package, you'll typically want to apply it as part of some bigger transformation process. You can get some sense of how this might work by reading the source code of downlit_html()
and downlit_md()
, which transform HTML and markdown documents respectively.