Add v3 r and rstudio images with R 4.5.3 and updated packages

[remlapmot]

Paging @bloodearnest

R 4.5.3, which according to the R-Core team will be the last patch release of R 4.5 https://stat.ethz.ch/pipermail/r-announce/2026/000718.html, was released last week on the 11th March. So I'd say it makes sense to have v3 r and rstudio images including that - which is what this does (and the v2 image is about 1 year old now - which is my personal limit for R related stuff).

The v3 additions are essentially a copy of what we do for v2 with the updated R and updated CRAN_DATE:

• The updated CRAN_DATE gets us the current versions of the packages notably including ggplot2 version 4.
• In a few places I've had to generalise the code to swap a hardcoded v2 for using the MAJOR_VERSION env var.
• Comparing the number of R packages - v2 has 544 - v3 has 545 (mainly due to different dependency requirements). I think the only notable difference is that on the new v3 CRAN_DATE the parglm R package is no longer available on CRAN https://cran.r-project.org/package=parglm, and the parglm GitHub repo has been untouched for 5 years https://github.com/boennecd/parglm, so in v3 I have replaced parglm with the fastglm package https://cran.r-project.org/package=fastglm.
• And I realised we never switched from the netlib BLAS/LAPACK to the faster OpenBLAS and its LAPACK - I've made that switch in v3 - running sessionInfo() this is reported in the following lines

  ❯ docker run -it --rm --platform linux/amd64 r:v3
  
  R version 4.5.3 (2026-03-11) -- "Reassured Reassurer"
  Copyright (C) 2026 The R Foundation for Statistical Computing
  Platform: x86_64-pc-linux-gnu
  
  R is free software and comes with ABSOLUTELY NO WARRANTY.
  You are welcome to redistribute it under certain conditions.
  Type 'license()' or 'licence()' for distribution details.
  
  R is a collaborative project with many contributors.
  Type 'contributors()' for more information and
  'citation()' on how to cite R or R packages in publications.
  
  Type 'demo()' for some demos, 'help()' for on-line help, or
  'help.start()' for an HTML browser interface to help.
  Type 'q()' to quit R.
  
  > sessionInfo()
  R version 4.5.3 (2026-03-11)
  Platform: x86_64-pc-linux-gnu
  Running under: Ubuntu 22.04.5 LTS
  
  Matrix products: default
  BLAS:   /usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3
  LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.20.so;  LAPACK version 3.10.0
  

• I also bumped the version of rstudio-server in v2.

As usual there was a nice summary of what's new in R 4.5 by Jumping Rivers https://www.jumpingrivers.com/blog/whats-new-r45/.

I don't believe I touched anything to do with v1 - so its epic 2hr build should still run. I have run the v2 and v3 r and rstudio builds and tests locally.

Maybe the opensafely CLI would need checking to see if it allows r:v3 and rstudio:v3?

Also - remember I have no permissions - so I can't formally 'request review' - and you'll have to trigger CI builds etc.

[rebkwok]

@remlapmot Many thanks for this. Everything looks good, and I've triggered the CI runs on the branch. I want to double-check the replacement of parglm with fastglm with the research team just so they're aware, and we'll have a wider discussion with the tech team about docs etc and the process for releasing a v3.

[remlapmot]

Thanks Becky

As far as I know fastglm is the best replacement for parglm in terms of features (i.e., they both fit Generalised Linear Models like the core stats library glm() function does - they just do so much faster). The only drawback is that unfortunately fastglm uses a slightly different syntax to parglm. I believe that parglm uses exactly the same syntax as the stats::glm() function, so if people had parglm::parglm(...) calls in their code they could simply amend those to stats::glm(...) and they would work - just with a longer runtime - or they could make the small syntax change within the call and use fastglm::fastglm(...slightly different syntax here...) to retain the faster runtime.

After releasing r:v3 and rstudio:v3 the only amend I would recommend you make in the opensafely CLI is changing the default version of rstudio to v3 in the opensafely launch command (such that

opensafely launch rstudio

will invoke opensafely launch rstudio:v3 instead of opensafely launch rstudio:v2 as it currently does). I think that would just be a one character amend on this line within the launch_rstudio() function.

https://github.com/opensafely-core/opensafely-cli/blob/e63d4c4b759c639810eee7cf81d9ff71b00f8e24/opensafely/launch.py#L165

I have checked that when the ghcr.io/opensafely-core/r:v3 and ghcr.io/opensafely-core/rstudio:v3 images exist the following two CLI calls already work as expected

opensafely exec r:v3

opensafely launch rstudio:v3

You can test this locally by building the two images from this branch then retagging e.g.,

just build v3
just build-rstudio v3
docker tag r:v3 ghcr.io/opensafely-core/r:v3
docker tag rstudio:v3 ghcr.io/opensafely-core/rstudio:v3

then running those commands.

[remlapmot]

Hi Becky

PR updated.

Good news. I have actually taken over maintainership of the parglm package myself and yesterday I got it reinstated on CRAN, so I have added included it the v3 image.

Also,

• I have kept the fastglm package in v3 because Will is interested in using it.
• I have also added SEQTaRget package to the v3 image as it our package for sequential target trial estimation and it is an alternative to the TrialEmulation package (which is already in v2 and v3).

[rebkwok]

Awesome, thanks @remlapmot