On blog location: This directory contains source and generated content which appears on https://blog.r-project.org/ It includes RSS feed, accessible via https://blog.r-project.org/index.xml, which needs to validate according to https://validator.w3.org/feed/check.cgi Absolute URLs need to be included in the feed, and the server part has to match the URL of the feed, so it is not possible to serve the (static) content from a different URL. A docker image: There is now a docker image with pre-installed software to generate the (static) site from the blog posts. It is to be used from a working copy of subversion https://svn.r-project.org/R-dev-web/trunk/Blog2 (so from the directory where this README file is) To download and use a pre-built version of the docker image, run: docker run -it -p 4321:4321 -v `pwd`:/Blog kalibera/rblog To build and use a local version, run: docker build -t rblog . docker run -it -p 4321:4321 -v `pwd`:/Blog rblog Running the image gives an R prompt and the container has read-write access to the current directory, with all software necessary to (re)-generate blog posts. To display the blog site locally (e.g. when working on a blog post), run blogdown::serve_site(host="0.0.0.0") and open a web browser on the host system to point to http://localhost:4321 The content should update automatically when any of the files is changed. To stop the server that is diplaying/generating the site locally: blogdown::stop_server() To regenerate all blog posts: blogdown::build_site(build_rmd=TRUE) If you prefer not to regenerate all posts, but only the new one(s), before committing updates: blogdown::check_content() blogdown::build_site() While the following text has instructions to install the needed software without docker, it includes important information also for bloggers using docker. To add a blog post and rebuild the site (all content is static): 1. checkout svn from https://svn.r-project.org/R-dev-web/trunk (it helps to do this in a clean installation that does not have any more packages than needed for the blogs to build, because some blogs include R code which may be affected by extra packages, such as new conflicts between names exported from different packages) 2. the blog is under "Blog2" 3. get a recent version of R-devel to build the blog (some posts demonstrate new R-devel features, so one needs an R version with the features to build the .Rmd files) 4. install blogdown package install.packages("blogdown") # or from github, see below blogdown uses "hugo" and provides functionality to install hugo (gives some R command to run the installation), but one can also install hugo manually on the local machine - on Ubuntu or Debian, it is "apt-get install hugo". hugo must not be too old (Ubuntu 18.04 has hugo 0.40 and it works fine, hugo 0.16 from Ubuntu 16.04 fails with "Error: unknown flag: --themesDir") On Ubuntu 20.04 (November 2, 2020, hugo version 0.68.3), the CRAN version of blogdown (0.21) did not recognize a running hugo server, so it killed it in serve_site(). A fixed version 0.21.33 could be found at https://github.com/rstudio/blogdown and worked. At the time of writing, the CRAN version of blogdown didn't have this problem. 5. install these packages (needed to build existing blog posts): dplyr, MASS, colorspace, tidyr, bench, ggplot2, tidyr, ggbeeswarm, kableExtra [ please update this list for any new dependencies, as well as the docker file ] 6. run svn update it may help to delete unneeded local files, so that you only have what is in the svn and in the current version (because many new files will then be generated and one needs to add them, without accidentally adding files not intended for commit) 7. go to directory Blog2, run "blogdown::build_site()" and check whether it finished without errors, rebuilding all blog posts; if not, go back to previous steps and try to fix the problem; note that from some recent version, blogdown stopped automatically rebuilding posts from Rmd files it has built already - use "blogdown::build_site(build_rmd=TRUE)" then to force the rebuilding 8. when build_site() is happy, run "blogdown::serve_site()", this will open a web browser and display the current version of the blog from the SVN working copy, so even local uncommitted changes; it used to get automatically updated when you edit your blog post (Rmd file), but later it broke and wasn't automatically happening, later it get fixed again; in previous versions, the Rmd files were rebuilt to be shown (hard-coded URLs), currently that is no longer happening 9. in another terminal window, go to "content/post" (to create a new post) start from copy of an existing post, e.g. a copy of something like 2018-03-23-dll-limit.Rmd, edit the meta data at the top of the Rmd file and edit the content; when one saves the file, hugo automatically rebuilds the blog _locally_ and the changes become visible in the local web browser (nothing is committed automatically) 10. when happy with the changes/new post, quit the R session running "serve_site()", re-start R, run "blogdown::build_site()"; that will take several seconds and rebuild all the blog posts (remember the need for option build_rmd=TRUE mentioned above); the blog posts should hence be written to be robust against rebuilding at a later time (do not include computations based on current time, on unfixed version of a package, etc). 11. "svn add" newly created files and directories for the blog (in addition to the Rmd files, these will include a html version, rendered images, meta-data for categories and tags, etc) 12. svn commit an updated version of the blog will appear eventually (when the website gets auto-updated) at https://blog.r-project.org/index.html General advice / experience: - the generated output unfortunately depends a bit on the version of hugo and blogdown; sometimes the difference is unimportant, but it is still a bit of pain checking the diffs - the committed version is usually generated with the latest Ubuntu LTS, hugo installed from Ubuntu and blogdown from CRAN and using recent R-devel; the plan is to use the docker image for it now that it is available - the posts should contain only a small amount of code generation (not run benchmarks, not run tests, only use R for convenient rendering), so that the result is more portable and easier to maintain - the posts should not depend on the current time of generation, because they are re-generated often (they should not print the current date/time, current R versions, etc); R should be used, if needed, only for rendering the content, not for demonstrating what R does/did/etc - the posts should not depend on features/properties of R that are likely to change soon (e.g. to demonstrate that something didn't work in an old version of R - such things have to be copied, not executed, because the blogs will eventually be re-generated with newer versions of R) - the instructions here are based on experience with earlier versions and are intentionally defensive; it has been reported that one can now again get away now without rebuilding all of the blog posts, by running blogdown::stop_server() # if serve_site() was used before blogdown::check_content() blogdown::build_site() - in either case, it is necessary to check any changes before committing, and all the blog posts should be "re-buildable", because time to time they will have to be rebuilt as the software for the site evolves