Questions regarding fixing broken nixpkgs

[mihem]

Hey @b-rodrigues, after installing nixos, I also wanted to see whether I could help fixing any broken rPackage.

Your intro is great: https://docs.ropensci.org/rix/articles/contributing-to-nixpkgs.html

However, I am little confused.

1. The link that you use there https://hydra.nixos.org/jobset/nixpkgs/r-updates These are buiilds from 2024. Where are latest broken builds?

2. And there is this list that you put together: https://raw.githack.com/b-rodrigues/nixpkgs-r-updates-fails/targets-runs/output/r-updates-fails.html . This is also based on Hydra but seems to be uptodate, 2026-03-23 .. so probably when you fixed packaged when the last R version was released. And these 78 CRAN packages are then broken? Because I thought the R nix team would fix all CRAN/BioC packages when there is a new R-release. No judgment, just asking, it would be nice if there is stuff that need to be fixed ;).

3. https://github.com/rstats-on-nix/build_tree/blob/main/.github/workflows/build_tree.yml is the workflow for you weekly checks. Where is the workflow for your daily CRAN snapshots? The last build was not successful (though some connection error and no package build failure).

Thanks!

[b-rodrigues]

Hi, sorry for the delayed response! So what happened is that hydra maintainers had to turn off builds of CRAN packages, because those for taking up quite a lot of resources. So hydra has not been building them since 2024.
However, the lead maintainer, jbedo, has been using his own hydra instance to build packages for x86-linux. This is what the my custom list shows.

So indeed, those 78 packages are broken. On each bump, we try to fix as many as possible, so we're in a pretty good state I'd say :)

We have not started the R 4.6.0 and Bioc 3.23 bump, but I have these already on the rstats-on-nix fork.

However, there are certain things to keep in mind:

• platforms other than x86-linux are not so well-tested, and there's likely more broken packages on aarch64-{linux, darwin}.
• some packages technically ok, but fail at runtime. For example, the package might need to further download stuff and expects a standard linux filesystem. That might be ok on non-NixOS linuxes, or on macOS, but not necessarily on NixOS.

You can find the action for the daily builds here: https://github.com/rstats-on-nix/nixpkgs/blob/r-daily-source/.github/workflows/r-daily.yml

[mihem]

Thanks.

So it would make sense if I have free time to try to fix some of those https://raw.githack.com/b-rodrigues/nixpkgs-r-updates-fails/targets-runs/output/r-updates-fails.html ?

But hydra is only for each new R release? There is no failed packages list for the weekly tests you run right? You "only" tests the packages listed here and then just fix them yourself every week https://github.com/rstats-on-nix/build_tree/blob/main/default.nix ?

What do you think where help is most needed?

[b-rodrigues]

So in a perfect world, hydra would build the whole tree for each new release, or if some input changes (downstream packages would then get rebuilt as well). Unfortunately, that is currently not the case.

But yes, you could go through the package in my list. Perhaps, since you're familiar with biocondcutor, you could try fixing bioc packages?

Also, because there's been a new bioc release, and a new R release, and they're not yet in upstream nixpkgs, I'd say use the daily branch of rstats-on-nix/nixpkgs fork to test if the listed packages are still broken. Then, once we bump R and Bioc upstream, you could open PRs to fix these upstream as well.

[mihem]

Thanks.

But jbedo runs all 23 000 CRAN + 2000 Bioconductor packages on his own hydra instance?
Doesn't this take many weeks? Or which packages are built for each release?

Yes, agree in a perfect world this would be built continuously or similar to your setup weekly or so.
Do you have any idea how much resource this costs?

So this list https://raw.githack.com/b-rodrigues/nixpkgs-r-updates-fails/targets-runs/output/r-updates-fails.html get's updated soon, right?

[mihem]

@b-rodrigues sorry, do you find time to answer my last post?

Also you might be interested: over the WE i added nix-based actions for my shiny app (inspired by your book and rix workflows and nix workflows https://github.com/mihem/cerebroAppLite/tree/master/.github/workflows)
I started with rix "latest_upstream" but changed to "bleeding_edge", so at least once a week i recreate this. There are quite some packages needed (1.1GB in my cachix), so maybe this is a good additional test if any of those packages are broken. Right now they all seem to build fine.
https://github.com/mihem/cerebroAppLite/blob/master/default.nix
https://github.com/mihem/cerebroAppLite/actions/runs/25640102218/job/75258939271

[b-rodrigues]

Thanks.

But jbedo runs all 23 000 CRAN + 2000 Bioconductor packages on his own hydra instance? Doesn't this take many weeks? Or which packages are built for each release?

No because most packages build quickly or are cached (they didn't change between releases nor their inputs), so it takes like a day.

Yes, agree in a perfect world this would be built continuously or similar to your setup weekly or so. Do you have any idea how much resource this costs?

Honestly probably not that much. Github actions can be used to build for free and the cost would only be storage. Less than 10 bucks a month I reckon.

So this list https://raw.githack.com/b-rodrigues/nixpkgs-r-updates-fails/targets-runs/output/r-updates-fails.html get's updated soon, right?

Yes when we'll bump.

Also you might be interested: over the WE i added nix-based actions for my shiny app (inspired by your book and rix workflows and nix workflows https://github.com/mihem/cerebroAppLite/tree/master/.github/workflows) I started with rix "latest_upstream" but changed to "bleeding_edge", so at least once a week i recreate this. There are quite some packages needed (1.1GB in my cachix), so maybe this is a good additional test if any of those packages are broken. Right now they all seem to build fine. https://github.com/mihem/cerebroAppLite/blob/master/default.nix https://github.com/mihem/cerebroAppLite/actions/runs/25640102218/job/75258939271

sweet !

[mihem]

Thanks.

Okay then I'll close this.
Maybe consider adding some of the stuff that you explained in the vignettes because they are interesting for a broader audience I think.