"Spring cleaning" of some ancient pre-processor directives

[eddelbuettel]

Include file inst/include/Rcpp/platform/compiler.h contains a number of ancient tests ("is gcc newer than 4.5 ?") that we should be able to simplify. One approach may be to open a branch and to chip away carefully, building on last year's variadic templates cleanup. I suspect we can get rid of most if not even all these 'else' branches as current Rcpp versions are very unlikely to encounter a compiler than does not have a minimal C++11 fallback.

For example, RCPP_USING_CXX0X_OR_LATER is defined there, but used only in one file (inst/include/Rcpp/sugar/functions/sapply.h) where it provides a fallback for typedef typename via local pre-C++11 implementation. That else case can likely be removed. By going carefully we may be able to get some simplifications in.

Thoughts?

[Enchufa2]

Definitely. +1

[eddelbuettel]

Should we be able to insist on 'C++11 or newer' ? It is a large enough watershed that it could lead to more cleanup. (Still noodling in platform/compiler.h after a few so-far-local-only tests.) Is there conceivably anything at CRAN insisting on C++ older than C++11? Non-CRAN builds can always be "cordially invited" to use an older Rcpp release...

[Enchufa2]

R does not support C++98 anymore (since v4.0.0). If you request it, R throws an error. So everything on CRAN compiles with at least C++11. I think it's safe to remove any older bits.

[eddelbuettel]

Yep that is what I thought. I started slowly (as I accidentally removed one line too many at the start) and carefully but now added an actual error if it is not at C++11. Stating that as a baseline will make future incremental commits easier.

First commit coming in a minute. Maybe keep an eye on them and eyeball as I go along.

[eddelbuettel]

"Because we can" our CI matrix go further back than R 4.0.0 so that version is actually now our baseline provided C++11 is opted into but of course as needed we can also remove some of the older checks. Will learn more about that as we go along.

[eddelbuettel]

🙄 Well that didn't work out. Bails well into the 4.* series.

PS I was too optimistic about __has_feature() but I can just #define given the changed baseline.

[kevinushey]

This sounds good to me. Maybe to make part of Rcpp 1.1.0, with C++11 now being the minimum supported standard? I'd love to tear out all of that generated code bloat.

[eddelbuettel]

That's the plan!

[eddelbuettel]

Rev.dep checks panned out fine so #1364 closes this -- and #1365 carries torch forward.