Rmodular cannot be built

[rekado]

I'm trying to build shogun with the Rmodular interface with GCC 4.9.3 and R 3.3.0 but I get this error:

error: conflicting declaration of C function ‘long long int std::abs(long long int)’
in gcc-4.9.3/include/c++/cstdlib

In file included from
/gnu/store/pbz6h59l4c4y77j5ik45krlvp4xq79qq-r-3.3.0/lib/R/include/R.h:40:0,
from
/tmp/guix-build-shogun-4.1.0.drv-0/shogun-4.1.0/src/interfaces/r_modular/sg_print_functions.cpp:2:

What then follows is a long stream of errors like "error: template with C linkage" in various files. I can provide the full build log in a few hours.

[rekado]

What follows are the cmake flags and the logs starting from the first error:
https://gist.github.com/rekado/01fcfd74e3defe4288ae3bab86078f3e

[karlnapf]

Can you please remove the long log here and paste a link to a gist instead? Otherwise chances that somebody reads it go to zero.
Next, could you post your system specs?

[rekado]

I moved the error log from the previous comment to a gist.
My system is an x86_64 machine with 8GB RAM but this doesn't seem to be related to the error which is about conflicting declarations in included files.

I'm building shogun in a chroot where only declared packages are available (that's a feature of GNU Guix). There were no problems with the modular R interface in version 4.0.0.

[karlnapf]

Sorry about the delay. This unfortunately is a bug. Will look into it and adding to 5.1 milestone

[karlnapf]

I locally have the same errors on ubuntu 16.04 with latest develop and standard packages, R version 3.3.1

https://gist.github.com/karlnapf/69946d524a9773b6cb4d9479ce93097e

[rekado]

It might be sufficient to #define NO_C_HEADERS before including R.h. The R NEWS file says that R.h should not be included from within extern C blocks, but that's what swig appears to do. With R 3.3.0 C headers are included when R.h is included. This can be avoided by defining NO_C_HEADERS.

I haven't tested this with Shogun yet, but it worked for SPAMS, which also includes bindings that were created with swig.

[karlnapf]

Thanks for the hint! This is since R 3.3.0

Just for the record, adding define NO_C_HEADERS in the sg_print_functions.cpp (manually, after it was generated by SWIG), results in different compilation errors

https://gist.github.com/karlnapf/9538ea56ead65bc1c470543ff6c3bd95

[karlnapf]

So the NEWS entry for 3.3.0 you refer to says:
https://cran.r-project.org/doc/manuals/r-release/NEWS.html

When R headers such as ‘R.h’ and ‘Rmath.h’ are called from C++ code in packages they include the C++ versions of system headers such as ‘<cmath>’ rather than the legacy headers such as ‘<math.h>’. (Headers ‘Rinternals.h’ and ‘Rinterface.h’ already did, and inclusion of system headers can still be circumvented by defining NO_C_HEADERS, including as from this version for those two headers.)

There is also this in the NEWS for 3.3.2

Use of the C/C++ macro NO_C_HEADERS is deprecated (no C headers are included by R headers from C++ as from R 3.3.0, so it should no longer be needed).

[karlnapf]

Same error with R 3.3.2

[karlnapf]

Fixed via #3538

[rekado]

Nice, thank you!

[karlnapf]

Let me know if it worked :)