consider requiring C++11 for 'long long'

[kevinushey]

From @jjf234:

Hi Kevin,

I noticed over the past couple of days that most packages that rely on RcppParallel are now getting a 'long long' warning on CRAN for 'r-devel-windows-ix86+x86_64'. Here's a sample warning:

d:/RCompile/CRANpkg/lib/3.4/RcppParallel/include/tthread/tinythread.h:632:14: warning: ISO C++ 1998 does not support 'long long' [-Wlong-long]

For reference, these are the packages and 'long long' warning messages:

• dbmss: https://www.r-project.org/nosvn/R.check/r-devel-windows-ix86+x86_64/dbmss-00check.html
• FLSS: https://www.r-project.org/nosvn/R.check/r-devel-windows-ix86+x86_64/FLSSS-00check.html
• GERGM: https://www.r-project.org/nosvn/R.check/r-devel-windows-ix86+x86_64/GERGM-00check.html
• markovchain: https://www.r-project.org/nosvn/R.check/r-devel-windows-ix86+x86_64/markovchain-00check.html
• microclass: https://www.r-project.org/nosvn/R.check/r-devel-windows-ix86+x86_64/microclass-00check.html
• roll: https://www.r-project.org/nosvn/R.check/r-devel-windows-ix86+x86_64/roll-00check.html
• SpatMCA: https://cran.r-project.org/web/checks/check_results_SpatMCA.html
• SpatPCA: https://cran.r-project.org/web/checks/check_results_SpatPCA.html

I've followed the steps on the RcppParallel site for creating R packages (http://rcppcore.github.io/RcppParallel/#r_packages) but was wondering if we need to also include or modify something else? If not, do you think this is an RcppParallel issue or a CRAN issue (since it only appeared in the last couple of days)?

Thanks in advance,
Jason

We might want to require C++11 support on Windows to avoid these CRAN warnings.

[eddelbuettel]

@jjf234: Well. what makes you think the warning is new? AFAIK we had that for years, and I used the same trick (of turning on C++11, or just C++0x with g++-4.6.*) to get long long for a number of packages. (Though something may of course be different with RcppParallel, my experience is more with Rcpp in general).

[kevinushey]

My guess is that this is a new warning as otherwise the CRAN maintainers would've rejected these package submissions.

[kevinushey]

It's also possible that package authors just need to re-submit with CXX_STD = CXX11 in their Makevars or Makevars.win, to enforce adherence to the C++11 standard.

[eddelbuettel]

We had those warnings for years. I even once tried an entire dummy package around it: RcppLongLong. So I still doubt these are new -- g++ has always complained about 'no long long' in that whole stupid context of C++98 not having them wheres C99 (IIRC, or maybe a 2003 interim C++ or something) had them yet we were not supposed to have access to it.

Anyway, all over now, apart from those base-layer packages we need to be able to compile with g++-4.6 or older.

[kevinushey]

I should emphasize that these are compiler warnings that are now being treated as R CMD check warnings; hence why there is some more urgency in getting a fix.

For RcppParallel + client packages, I am guessing the simplest fix will be to just add users to add CXXSTD = CXX11 to their Makevars.

[kevinushey]

And I'm guessing that a recent R-devel change is now causing R to compile with the -pedantic flag set on Windows, which triggers this warning on gcc 4.9.3.

[eddelbuettel]

True dat on both counts. Cannot propagate a compilation standard down to a client when the "interface" is offering a directory with headers :-/

[kevinushey]

I've confirmed that no warnings are emitted when compiling RcppParallel client packages with CXX_STD = CXX11. I've contacted the maintainers of downstream packages and instructed them to add this to their src/Makevars[.win] files, and let the CRAN team know as well.

[kevinushey]

I'm going to update the RcppParallel documentation to suggest the use of CXX_STD = CXX11 for client packages, but otherwise I think we can consider this issue closed.