I am trying to compile R under a new setup, and frankly, I have had a lot
of problems, but I think the stuff below points to a possible bug in R's
(custom) configure checks for C++11/14/17, but not for C++98.
This is a report about R from the R-3-6 branch, with a svn checkout from
today, revision r77135.
In my case the compiler name is x86_64-conda_cos6-linux-gnu-g++, not g++. I
denote this in my configure call, using the CC variable. A snippet of the
full configure is
../${SRCDIR}/configure SHELL='/bin/bash' \
--prefix="${CONDA_PREFIX}/R/${R_VERSION}" \
CC="x86_64-conda_cos6-linux-gnu-gcc" \
CXX="x86_64-conda_cos6-linux-gnu-g++" \
F77="x86_64-conda_cos6-linux-gnu-gfortran" \
FC="$F77" \
CFLAGS="-Wall -mtune=amdfam10 -g -O2 -I${CONDA_PREFIX}/include"\
CXXFLAGS="-Wall -mtune=amdfam10 -g -O2 -I${CONDA_PREFIX}/include" \
F77FLAGS="-Wall -g -O2 -mtune=amdfam10 -I${CONDA_PREFIX}/include" \
CXX11="g++" \
CXX11STD="-std=c++11" \
CXX11FLAGS="-Wall -mtune=amdfam10 -g -O2 -I${CONDA_PREFIX}/include" \
CXX11PICFLAGS="-fPIC" \
Where $CONDA_PREFIX is given in my script.
The output in config.log is given below. Note that in the test for c++98,
it uses the "right" CC, but in the test for c++11 it uses g++. This looks
wrong to me:
configure:28111: checking whether x86_64-conda_cos6-linux-gnu-g++ supports
C++98 features with -std=gnu++98
configure:28130: x86_64-conda_cos6-linux-gnu-g++ -std=gnu++98 -c -Wall
-mtune=amdfam10 -g -O2
-I/jhpce/shared/jhpce/core/conda/miniconda3-4.6.14/envs/svnR-3.6/include
-fpic -DNDEBUG -D_FORTIFY_SOURCE=2 -O2 conftest.cp
p >&5
configure:28130: $? = 0
configure:28139: result: yes
configure:28315: checking whether g++ -std=c++11 supports C++11 features
configure:28607: g++ -std=c++11 -c -Wall -mtune=amdfam10 -g -O2
-I/jhpce/shared/jhpce/core/conda/miniconda3-4.6.14/envs/svnR-3.6/include
-fPIC -DNDEBUG -D_FORTIFY_SOURCE=2 -O2 conftest.cpp >&5
../R-3.6-src/configure: line 2355: g++: command not found
configure:28607: $? = 127
configure: failed program was:
I have similar issues (wrong CC using when compiling the test program) with
the test for c++14, whereas the test for c++17 has empty space where the CC
variable should be?
I can fix this issue by adding a soft link in my PATH from g++ to my
compiler of choice. In this case configure finishes and reports that I have
full C++17 capabilities. Weirdly, in the output, note that the C++ compiler
is "wrong" again, despite my configure call:
Source directory: ../R-3.6-src
Installation directory:
/jhpce/shared/jhpce/core/conda/miniconda3-4.6.14/envs/svnR-3.6/R/3.6
C compiler: x86_64-conda_cos6-linux-gnu-gcc -Wall
-mtune=amdfam10 -g -O2
-I/jhpce/shared/jhpce/core/conda/miniconda3-4.6.14/envs/svnR-3.6/include
Fortran fixed-form compiler:
/jhpce/shared/jhpce/core/conda/miniconda3-4.6.14/envs/svnR-3.6/bin/x86_64-conda_cos6-linux-gnu-gfortran
-fno-optimize-sibling-calls -fopenmp -march=nocona -mtune=haswell
-ftree-vectorize -fPIC -fstack-protector-strong -fno-plt -O2
-ffunction-sections -pipe
Default C++ compiler: g++ -std=c++11 -Wall -mtune=amdfam10 -g
-O2 -I/jhpce/shared/jhpce/core/conda/miniconda3-4.6.14/envs/svnR-3.6/include
C++98 compiler: x86_64-conda_cos6-linux-gnu-g++ -std=gnu++98
-Wall -mtune=amdfam10 -g -O2
-I/jhpce/shared/jhpce/core/conda/miniconda3-4.6.14/envs/svnR-3.6/include
C++11 compiler: g++ -std=c++11 -Wall -mtune=amdfam10 -g
-O2 -I/jhpce/shared/jhpce/core/conda/miniconda3-4.6.14/envs/svnR-3.6/include
C++14 compiler: g++ -std=gnu++14 -Wall -mtune=amdfam10 -g
-O2 -I/jhpce/shared/jhpce/core/conda/miniconda3-4.6.14/envs/svnR-3.6/include
C++17 compiler: g++ -std=gnu++17 -Wall -mtune=amdfam10 -g
-O2 -I/jhpce/shared/jhpce/core/conda/miniconda3-4.6.14/envs/svnR-3.6/include
Fortran free-form compiler:
/jhpce/shared/jhpce/core/conda/miniconda3-4.6.14/envs/svnR-3.6/bin/x86_64-conda_cos6-linux-gnu-gfortran
-fno-optimize-sibling-calls
Obj-C compiler:
--
Best,
Kasper
[[alternative HTML version deleted]]
______________________________________________
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
