chore: Replace Rcpp by cpp11

[Antonov548]

• busy_callback_helper() -> how to catch eval error and interrupt?
• result_release() -> make this work with cpp11::external_pointer
• find replacement for Rcpp::warningcall(R_NilValue, std::string("Ignoring remaining part of query: ") + tail);
• figure out why cpp11::warning() crashes
• adapt astyle()
• maybe remove some STRICT_R_HEADERS and R_NO_REMAP

[Antonov548]

@krlmlr
Seems cpp11 hasn't any functions to handle exceptions from R functions called in C++ via cpp11::function.
In Rcpp they adding try catch block around functions passed from R to C++ also by installing additional scripts.

We can implement this logic in RSQLite just using R native functions or just remove catching of errors from R, at least for now.

[krlmlr]

I remember seeing a safe modifier, e.g., in https://github.com/cran/duckdb/blob/f2550069da4019ebb5f81103dd92d471c40d4d3b/inst/include/cpp11/list.hpp#L92

Would that help?

[krlmlr]

I also wonder why you commit with a different committer name and e-mail. Is this from gitpod?

[Antonov548]

Yes, this is due to the gitpod, I also wondering why it sometime from your name.
Maybe it's due gitconfig changes in gitpod configuration in RMariaDB project - https://github.com/r-dbi/RMariaDB/blob/66f890fdf08489b76e1a3dea20db28c1a15c9aba/.gitpod.yml#L7

[Antonov548]

I remember seeing a safe modifier, e.g., in https://github.com/cran/duckdb/blob/f2550069da4019ebb5f81103dd92d471c40d4d3b/inst/include/cpp11/list.hpp#L92

Would that help?

Yes, I didn't think about safe object. I will take a look, thanks.

[krlmlr]

Thanks, weird. Can you please try this in a clean repository that doesn't have a .gitpod.yml ?

[Antonov548]

Thanks, weird. Can you please try this in a clean repository that doesn't have a .gitpod.yml ?

Yes, I will try

[krlmlr]

Can we vendor the cpp11 package with cpp11::vendor_cpp11() and fix the tests that are now skipped?

[krlmlr]

We should un-skip the tests and fix the code instead. @Antonov548: can you please take a look?

[krlmlr]

@Antonov548: The "interrupt" test still fails legitimately. To reproduce, the package needs to be installed with R CMD INSTALL . from the same branch, we've omitted this today. Can you please take a look?

[krlmlr]

Thanks!