take Additional_repositories into account

[ThierryO]

This PR should solve r-universe-org/help#109 directly. The suggested workaround is not longer needed.

[jeroen]

I appreciate the suggestion, but the problem is that this makes the repository inconsistent, and therefore the end-user who wants to install packages from your universe won't be able to install that package anymore.

We want to be sure that the user can always do:

install.packages("checklist", repos = "https://inbo.r-universe.dev")

And that means that the dependencies of your package will have to be either on CRAN or in your universe, because install.packages does not take Additional_repositories into account either.

[ThierryO]

Good point.

But what when the package depends on a package we can't add to our universe. E.g. the ladybird package has Additional_repositories: https://inla.r-inla-download.org/R/stable. Adding that package via Remotes: github::inbo/INLA fails on R-universe because it is too large...

Should I try to suggest a patch for utils::install.packages() to handle this?

[jeroen]

Should I try to suggest a patch for utils::install.packages() to handle this?

You could try that, but I doubt r-core will take it.

Maybe we can try to help fix the INLA package. There is no reason the package has to be that huge, it includes a ton of files that are not needed (e.g. binaries for linux and mac in a windows package, etc).

[jeroen]

It looks like the latest stable release of INLA is now slightly under 100MB. Maybe if you update https://github.com/inbo/INLA to this version, that will solve your issue?

https://inla.r-inla-download.org/R/stable/src/contrib/INLA_21.11.22.tar.gz

[jeroen]

I tried to build INLA from source but I can't get it to work :( The packages that they publish somehow have a mix of win/mac/linux binaries.

Looking at CRAN packages that depend on INLA, none of them have a "hard" dependency (imports) but instead they all seem to use Suggests: INLA such that the package can be built, even when INLA is not available:

• https://cran.r-project.org/web/packages/inlabru/index.html
• https://cran.r-project.org/web/packages/meta4diag/index.html

Perhaps you can also make INLA a Suggests in ladybird, such that the package can be built without it? But that requires also that you remove the @importFrom INLA and instead replace them with INLA::function() calls in ladybird. This would make ladybird suitable for CRAN as well.

[ThierryO]

I'm trying to move from Imports: INLA to Suggests: INLA. I'm running into problems when a package depends on another package that also uses Suggests: INLA. See https://stat.ethz.ch/pipermail/r-package-devel/2022q1/007715.html for a more elaborate description of the problem.

@jeroen, do you have suggestions to fix this?