Fully delegate package installation to rstantools

[andrjohns]

With the latest release of rstantools (2.3.0), we now support exposing stan functions during the installation process (so you no longer have to manually create the c++ sources).

This PR updates your package structure to fully delegate the installation/configuration to rstantools, allowing your package to remain compatible with future versions of rstan/StanHeaders.

I've also added a GHA R-CMD-CHECK workflow from the r-lib/actions repository so you can verify that all still works & passes

[jtimonen]

Thanks!

[jtimonen]

So the installation process now seems to generate an src/ directory, R/RcppExports.R and R/stanmodels.R inside the source location. So should I just .gitignore those?

[andrjohns]

Yep that would be best

[andrjohns]

When you're able, would you mind submitting a new release to CRAN with this PR?

[jtimonen]

I tried it, but got a bunch of warnings and notes that I don't get locally with R CMD check --as-cran. Some of them are easy to fix but some I might need help with.

[jtimonen]

I don't know how to fix this:

* checking whether package 'lgpr' can be installed ... WARNING
Found the following significant warnings:
  WARNING: The tools required to build C++ code for R were not found.

It is now the only warning that I get on winbuilder with r-devel

[andrjohns]

That's not a warning that you need to fix. It's referring to the system being used to run the checks, rather than the configuration of your package itself

[jtimonen]

But I also get this on my own mac, even though my c++ toolchain should be okay and the installation works, so I don't really understand what is going on.

[andrjohns]

The check/warning has no relation to your package or its configuration, just the system itself. If you receive it on your system, but the tools are installed, then it's just a false positive.

[jtimonen]

Okay thanks, I have sent a message to CRAN explaining this.

[andrjohns]

@jtimonen It looks like this warning might have rstantools-related after all (apologies for the misdirection!), since it's also being seen in another package using the same framework.

Did CRAN have a response to your message? I'm trying to replicate locally (no luck yet), but was hoping that they might have responded with more context/information

[jtimonen]

No they haven't responded

[andrjohns]

Sorry again for the wrong advice! The warning occurs when calling rstan::expose_functions() with rstan 2.21 and only when rstudio isn't being used, which was why this was missed initially.

I've submitted a PR to rstantools to fix the warning, and I'll let you know when it's on CRAN

[andrjohns]

rstantools 2.3.1 has been released and now the warning should no longer pop up

[andrjohns]

@jtimonen sorry to double-ping, but would you mind trying the submission again soon?

[jtimonen]

Done

[andrjohns]

Done

Thanks, very much appreciated!

[jtimonen]

https://cran.r-project.org/web/packages/lgpr/index.html <- now on CRAN (binaries not yet updated)