Refactor some R_ext includes

[Enchufa2]

Closes #1417

• Drop <R_ext/Callbacks.h>, not used anymore and dropped from the API.
• Move <R_ext/Visibility.h> to Rcpp/r/headers.h.

Checklist

• Code compiles correctly
• R CMD check still passes all tests
• Preferably, new tests were added which fail without the change
• Document the changes by file in ChangeLog

[eddelbuettel]

Thanks for the prompt PR! The CI check from the 'dev' container probably may mean nothing as the update is weekly and may have missed the change. Let me update my local build real quick and check locally too.

[Enchufa2]

I missed the PrtUtil.h, thanks, fixed now.

[eddelbuettel]

I am turning the reverse-depends machinery on, though I don't expect too much. Packages using Rcpp as well as headers from R will have both include directories, and we are testing under R-release so this is bound to be washed over. But one never knows and we have gotten surprises before ...

[kevinushey]

LGTM. I'm a bit surprised these headers are being removed outright rather than just using a compiler pragma warning to note that they'll be removed. At least that would reduce the amount of CRAN breakage in the interim, while still giving package authors a chance to update...

[eddelbuettel]

I am equally shocked. It's plain cold breakage and I suspect the CRAN parts of R Core are not too happy either.

[Enchufa2]

I missed this one:

Rcpp/inst/include/Rcpp/Interrupt.h

Line 25 in 2b0a44b

#include <R_ext/GraphicsEngine.h>

This one is needed for Rf_onintr. But I think it's better if we move it to inst/include/Rcpp/r/headers.h too. Is this ok? This has no impact, so no need to rerun reverse dependencies.

[eddelbuettel]

I would prefer if we did that next cycle. It is entirely cosmetic and there are dozens of other cosmetic things we could be doing with the headers (the directory layout is a mess, the file naming is inconsistent etc). We (as well as "other parties") have also learned at times that moving header files around is never quite as clean and simple as you think. Now it is just a call from us to a header of theirs so as you say "should" be low-impact. But then again I had to put "should" in quotes because you never know.

[eddelbuettel]

Reverse-depends concluded, there is not thing new (package MIC failed, but it seems to be fighting with xgboost not us). An above-average set of fails, but these are same as in the last run (scrapnel from rncl needing to change, plus a handful or packages that already have deadlines at CRAN). Committed results as RcppCore/rcpp-logs@22496ce

So squash-merging ahead.