Why and how maintain a NEWS file for your R package?
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.
Your R package is doomed to evolve as you add new features and bug fixes. There is some value in cultivating a NEWS file as a response to package transformation. In this post, we shall go through why keeping track of changes, how to create and maintain a changelog, advocating for the NEWS.md format.
Why cultivate a changelog?
You will most probably change stuff in your package. Discussing package evolution is beyond the scope of this post, refer to e.g. this chapter of the rOpenSci dev guide. This post is about documenting changes. Even with a very clear version control history1, having a text version of changes can be useful:
to you and other developers, when looking back;
to contributors, if you acknowledge their user name;
to users, when updating the package or wondering when something changed;
to you and other communicators, when drafting content (blog post, tweet, email) about a release.
How to write such a changelog, in practice? We shall discuss format in the next section. As regards contents, beside referring to the changelogs you like, the NEWs chapter of the tidyverse style guide is a very useful read.
Why write the changelog as NEWS.md?
There are several possible formats for maintaining an R package changelog, according to the documentation of
util::news(): inst/NEWS.Rd formatted like other Rd files, a Markdown NEWS.md, plain-text NEWS or inst/NEWS.
In all cases, it means users can access the changelog
From the package CRAN page (in the case of CRAN packages, of course). See
rhubNEWS. The differences in URL path, display, page title are due to differences in the way the different changelog formats are parsed and rendered.
utils::news(), either reading it all at once
news(package = "rhub") Changes in version 1.1.1 Enhancements - cran_summary() now messages that we recommend to fix all NOTEs, WARNINGs and ERRORs before a CRAN submission when the check results aren't 0 NOTE, 0 WARNING, 0 ERROR. - cran_summary() now outputs informative messages when any of the builds of the group hasn't completed (yet, or at all). Bug fixes - cran_summary() now works for packages whose R CMD Check result include no NOTE/WARNING/ERROR, and gives an informative error message when not all builds are completed yet. - cran_summary() now prints lines to screen without unwanted indentation. Changes in version 1.1.0 New features - New local_check_linux() function to run an R-hub check on the local host's Docker. New local_check_linux_images() function to list R-hub Docker images. - New check_on_solaris() shortcut to check on Solaris X86, without building the PDF manual or the vignettes. - New get_check() function that works with check ids, or a check group id. - list_package_checks() and list_my_checks() now output a tibble, that is nicely formatted when printed to the screen. - The output of get_check(), check(), check_on_, check_for_cran(), etc. functions gained - an urls() method returning a data.frame with URLs to the html and text logs, as well as the artifacts, of the check(s); - a browse() method replacing the web() method for opening the URLs corresponding to a rhub_check object. - New cran_summary() method to print a summary for a group or set of checks. Bug fixes - In printing methods the submitted time is now always correct thanks to explicitly specifying units for as.numeric.difftime (@jimhester, #94 and @schloerke, #135). Changes in version 1.0.2 First public release.
or specifying a query
news( query = grepl("check_on_solaris", Text), package = "rhub" ) Changes in version 1.1.0 New features - New local_check_linux() function to run an R-hub check on the local host's Docker. New local_check_linux_images() function to list R-hub Docker images. - New check_on_solaris() shortcut to check on Solaris X86, without building the PDF manual or the vignettes. - New get_check() function that works with check ids, or a check group id. - list_package_checks() and list_my_checks() now output a tibble, that is nicely formatted when printed to the screen. - The output of get_check(), check(), check_on_, check_for_cran(), etc. functions gained - an urls() method returning a data.frame with URLs to the html and text logs, as well as the artifacts, of the check(s); - a browse() method replacing the web() method for opening the URLs corresponding to a rhub_check object. - New cran_summary() method to print a summary for a group or set of checks.
Some users might get creative and actually write their own function for browsing changelogs.
What NEWS format is the most popular? The actually serious CRAN package
ouch uses inst/NEWS.Rd, like 134 other packages at the time of writing.
In the case of
ouch, inst/NEWS.Rd is actually created from a plain-text inst/NEWS using
R CMD Rdconv.
As regards plain-text NEWS files like the one in commonmark source2, when preparing this post I found 413 inst/NEWS and 1,055 NEWS in CRAN packages.
Last but not least, at the time of writing there were 1,174 packages with a NEWS.md file
NEWS.md is a great format
Now, why do I think NEWS.md is the best format? Of course you are free to disagree and to present your worflow in the comments! ?
If you’re used to R Markdown, then it makes sense to use Markdown in NEWS.md i.e. it will come naturally to you.
If your website has a
pkgdownwebsite, NEWS.md will be rendered as a nice changelog page. See e.g.
rhubchangelog, notice how links to GitHub users and issues resolve with no effort made in the NEWS.md itself
There is an
usethisshortcut for creating a NEWS.md:
usethisto increase version will add the corresponding heading to NEWS.md; and
usethiswill help you create an informative GitHub release. Also note the interesting
newsmdpackage, like a
descfor NEWS, that might help you write your own helpers?
Limits of NEWS.md as a format
There are two limitations to NEWS.md.
One caveat is the
util::news()function. If the user uses that and doesn’t have
xml2installed, it will fail4 – which is documented but not especially pleasant.
Another caveat, more specific to GitHub, is that NEWS.md locally doesn’t render links to issues magically, unlike GitHub releases (and unlike
pkgdown). That is why Yihui Xie uses a placeholder inst/NEWS.Rd in e.g.
knitr. Maybe a good idea for a function would be one to convert from NEWS.md to an informative inst/NEWS.Rd as feature-rich as
news(package = "knitr") Changes in version 999.999 o This NEWS file is only a placeholder. The version 999.999 does not really exist. Please read the NEWS on Github: <URL: https://github.com/yihui/knitr/releases>
Limits to NEWS as a communication channel
Now, no matter as good your NEWS.md file is, you need to keep in mind that some (most?) users will never read it. ? How to be sure they are informed, in that case?
On the one hand, communicating widely about executed or planned releases in blog posts for users and for developers can be useful.
If your blog post RSS feed is not a part of R Weekly, you can submit such links via a PR to the R Weekly repo (“Updated packages” category) or via R Weekly webform.
Maybe you even have a mailing list for users?
And to come back to
pkgdown websites, you can tweak the changelog to have it display one page per major version and related minor versions, and to have the navbar feature release blog posts.
On the other hand, you can add information about changes inside the package manual!
For instance, in
tidyr docs there are lifecycle badges and explanatory text: compare the page of
tidyr::gather() (retired) and
How and when to update the changelog?
Hi #rstats ? developers, how to you auto generate your https://t.co/dcTogPqNa8 file when ready for a release? Do you actually update https://t.co/dcTogPqNa8 as you close an issue? 'Cause that's just crazyness ? @jimhester_— Stu Field (@stufield3) March 8, 2020
Now, in terms of workflow, you could
update the changelog for each contribution5;
only update the changelog before releases, by looking at version control history and the issue tracker, potentially using something like GitHub milestones;
fledgethat “has been designed to streamline the process of versioning R packages on Git, with the functionality to automatically update NEWS.md and DESCRIPTION with relevant information from recent commit messages”.
In all cases you’ll probably want to polish the changelog before releases, as e.g.
usethis would remind you.
In this post we made the case for maintaining a changelog for your package, and for doing it in a NEWS.md file. We also explained the limitations of NEWS.md as a way to efficiently inform users of changes, since users might lose track or never read changelogs: communicating around releases and adding lifecycle badges in the package manual itself can help mitigate those limitations. As a package user, how do you follow NEWS of packages?6 Do you ever read changelogs?7
A good way to achieve one is adopting git flow. You can make many stupid commits in a branch, and then squash and merge to master! It can also be wise to learn about rewriting history. ↩︎
commonmarkto parse Markdown NEWS files but
commonmark‘s own NEWS file is plain-text. ? ? ↩︎
commonmarkis awesome for parsing Markdown!
roxygen2uses it to parse tags with Markdown markup . ↩︎
Interestingly, for extracting URLs out of NEWS.md for URL checks, R uses another approach involving Pandoc. ↩︎
That’d pair well with git flow and some sort of PR automatic checks! ↩︎
For packages developed on GitHub, and if you don’t oppose using GitHub, there are different levels of repository watching including being notified of releases only. ↩︎
Reading changelogs can be useful but also pleasant, e.g. when discovering the package uses version nicknames. ✨ ↩︎
R-bloggers.com offers daily e-mail updates about R news and tutorials about learning R and many other topics. Click here if you're looking to post or find an R/data-science job.
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.