R devel - Jun 2022 - R CMD Check should allow packages with C and C++ and OpenMP without generating a NOTE

Dear Developers,

I have the following issue with my package collapse:
https://github.com/SebKrantz/collapse,
whose code base is 34.5% C code and 28.3% C++, and to which I now want to
add OpenMP multithreading and send it to CRAN. For this I have set up a
Makevars file with

PKG_CFLAGS = $(SHLIB_OPENMP_CFLAGS)
PKG_CXXFLAGS = $(SHLIB_OPENMP_CXXFLAGS)
PKG_LIBS = $(SHLIB_OPENMP_CXXFLAGS)

as linking is through C++ via Rcpp. The problem is that this generates a
NOTE on R CMD Check:

src/Makevars: SHLIB_OPENMP_CFLAGS is included in PKG_CFLAGS but not in
PKG_LIBS

If I alternatively put

PKG_LIBS = $(SHLIB_OPENMP_CXXFLAGS) $(SHLIB_OPENMP_CFLAGS)

I get another NOTE stating that this is not portable. Finally putting

PKG_LIBS = $(SHLIB_OPENMP_CFLAGS)

gives a note (or warning) that linking is through C++. CRAN maintainers
have not engaged constructively with me on the issue, but I do believe this
is an issue requiring constructive engagement i.e. I believe R CMD Check
should be adjusted to allow for such a configuration, and possibly some
instructions should be added to the Writing R Extensions Manual. I have
also asked Dirk Eddelbuettel (
https://github.com/SebKrantz/collapse/issues/258) who agrees that this is
an upstream issue. I am happy for any proactive comments or suggestions.

Best regards,

Sebastian Krantz

	[[alternative HTML version deleted]]

For the record: it seems a simple text-substitution approach in the
Makevars file solves the issue, e.g. to have a package combining C, C++ and
OpenMP, your Makevars file could look like this:

PKG_CFLAGS = $($(subst OPENMP,OPENMP_CFLAGS,SHLIB_OPENMP))
PKG_CXXFLAGS = $(SHLIB_OPENMP_CXXFLAGS)
PKG_LIBS = $(SHLIB_OPENMP_CXXFLAGS)

where $($(subst OPENMP,OPENMP_CFLAGS,SHLIB_OPENMP)) substitutes
OPENMP_CFLAGS for OPENMP, thus reconstructing SHLIB_OPENMP_CFLAGS and
avoiding errors from static code analysis on R CMD Check.

Needless to say this is not very elegant and it would be great if R CMD
Check could be adjusted.

Best regards,

Sebastian

On Tue, 7 Jun 2022 at 20:44, Sebastian Martin Krantz <
sebastian.krantz at graduateinstitute.ch> wrote:
> Dear Developers,
>
> I have the following issue with my package collapse:
> https://github.com/SebKrantz/collapse,
> whose code base is 34.5% C code and 28.3% C++, and to which I now want to
> add OpenMP multithreading and send it to CRAN. For this I have set up a
> Makevars file with
>
> PKG_CFLAGS = $(SHLIB_OPENMP_CFLAGS)
> PKG_CXXFLAGS = $(SHLIB_OPENMP_CXXFLAGS)
> PKG_LIBS = $(SHLIB_OPENMP_CXXFLAGS)
>
> as linking is through C++ via Rcpp. The problem is that this generates a
> NOTE on R CMD Check:
>
> src/Makevars: SHLIB_OPENMP_CFLAGS is included in PKG_CFLAGS but not in
> PKG_LIBS
>
> If I alternatively put
>
> PKG_LIBS = $(SHLIB_OPENMP_CXXFLAGS) $(SHLIB_OPENMP_CFLAGS)
>
> I get another NOTE stating that this is not portable. Finally putting
>
> PKG_LIBS = $(SHLIB_OPENMP_CFLAGS)
>
> gives a note (or warning) that linking is through C++. CRAN maintainers
> have not engaged constructively with me on the issue, but I do believe this
> is an issue requiring constructive engagement i.e. I believe R CMD Check
> should be adjusted to allow for such a configuration, and possibly some
> instructions should be added to the Writing R Extensions Manual. I have
> also asked Dirk Eddelbuettel (
> https://github.com/SebKrantz/collapse/issues/258) who agrees that this is
> an upstream issue. I am happy for any proactive comments or suggestions.
>
> Best regards,
>
> Sebastian Krantz
>
>
>
>
>
	[[alternative HTML version deleted]]