Update extconf.rb so Mac OS X always uses clang (if correct version is available) and stops looking for GCC
xMinh129 opened this issue · 15 comments
I'm having this issue while trying to install a gem. Anybody could help me? Appreciate it a lot!
Gem::Ext::BuildError: ERROR: Failed to build gem native extension.
current directory:
/Users/minhvu/.rvm/gems/ruby-2.3.0/gems/bindex-0.5.0/ext/bindex
/Users/minhvu/.rvm/rubies/ruby-2.3.0/bin/ruby -r
./siteconf20170507-3967-bvkwf8.rb extconf.rb
creating Makefile
current directory:
/Users/minhvu/.rvm/gems/ruby-2.3.0/gems/bindex-0.5.0/ext/bindex
make "DESTDIR=" clean
current directory:
/Users/minhvu/.rvm/gems/ruby-2.3.0/gems/bindex-0.5.0/ext/bindex
make "DESTDIR="
compiling cruby.c
gcc: error: unrecognized command line option ‘-Wshorten-64-to-32’
gcc: error: unrecognized command line option ‘-Wdivision-by-zero’; did you mean
‘-Wdiv-by-zero’?
gcc: error: unrecognized command line option ‘-Wextra-tokens’; did you mean
‘--extra-warnings’?
make: *** [cruby.o] Error 1
make failed, exit code 2
Gem files will remain installed in
/Users/minhvu/.rvm/gems/ruby-2.3.0/gems/bindex-0.5.0 for inspection.
Results logged to
/Users/minhvu/.rvm/gems/ruby-2.3.0/extensions/x86_64-darwin-16/2.3.0/bindex-0.5.0/gem_make.out
An error occurred while installing bindex (0.5.0), and Bundler cannot
continue.
Make sure that `gem install bindex -v '0.5.0'` succeeds before bundling.
This is my ruby version : ruby 2.3.0p0 (2015-12-25 revision 53290) [x86_64-darwin16]
@v0dro We should probably fix this. The weird stuff in extconf is a workaround for when clang wasn't able to compile NMatrix and we had to use GCC. Now the reverse is true, and GCC won't work, but clang does.
Yes you have a point. I don't have access to a Mac though so I can't work on this as of now.
I think it should be a small change in extconf which will first check the environment and set the variables in CONFIG appropriately.
See this: http://stackoverflow.com/questions/4554837/how-to-specify-compiler-for-gem-install
Or maybe use use rake-compiler.
We don't need to rewrite extconf totally. It's already overriding the system default to use GCC on Mac. We just need to remove the override.
If you're on Mac you should clang not gcc. #584
how exactly? don't assume i have any idea what i'm doing.
if someone's at least willing to help me understand the problem, i'd be happy to try and work out a solution.
worked around by temporarily symlinking to clang.
mv /usr/local/bin/gcc-4.9 /usr/local/bin/gcc-4.9-orig
mv /usr/local/bin/g++-4.9 /usr/local/bin/g++-4.9-orig
ln -s $(which clang) /usr/local/bin/gcc-4.9
ln -s $(which clang++) /usr/local/bin/g++-4.9
gem install nmatrix
rm /usr/local/bin/gcc-4.9 /usr/local/bin/g++-4.9
mv /usr/local/bin/gcc-4.9-orig /usr/local/bin/gcc-4.9
mv /usr/local/bin/g++-4.9-orig /usr/local/bin/g++-4.9
I am having this issue, too. Or one that is closely related. The error is:
mkmf.rb:50:in 'gplusplus_version': unable to determine g++ version (match to get version was nil) (RuntimeError)
extconf failed, exit code 1
This occurred both before and after I followed the instructions for OS X here: https://github.com/SciRuby/nmatrix/wiki/Installation
I also tried setting up a symbolic ink for g++-4.8, to match the one for gcc. When I did that, I got a different error, as shown in the mkmf.log file
gcc: error: unrecognized command line option '-Wshorten-64-to-32'
gcc: error: unrecognized command line option '-Wdivision-by-zero'
gcc: error: unrecognized command line option '-Wextra-tokens'
checked program was:
/* begin */
1: #include "ruby.h"
2:
3: int main(int argc, char **argv)
4: {
5: return 0;
6: }
/* end */
This is a big problem. mikecmpbll's fix didn't work for me, either, as I am now on OS X 10.10.5.
I should also note: this is a local problem. I did manage to get nmatrix installed on our Linux server. But once I did, every time we run a script we get the following message, repeated 8 times, whenever a script starts up (even scripts that do not use nmatrix):
DEPRECATION WARNING: alias_method_chain is deprecated. Please, use Module#prepend instead. From module, you can access the original method using super. (called from require at /home/deploy/.gem/ruby/2.3.0/gems/bundler-1.16.0/lib/bundler/runtime.rb:81)
That is on Ruby 2.3.3p222, Rails 5.0.0.1.
@MohawkJohn Only if I can successfully upgrade to High Sierra. So far, it has been no go.
@lonny You should still be able to compile with clang/clang++. What happens if you open irb and do the thing I suggested above?
2.3.0 :002 > puts CONFIG['CC']
gcc
=> nil
I am going to continue my efforts to upgrade to High Sierra. But I have to do that on my "off time", as it were.
What about if you do CC=clang CXX=clang++ bundle exec rake compile?
<path>/mkmf.rb:50:in 'gplusplus_version': unable to determine g++ version (match to get version was nil) (RuntimeError)
from /Users/lonny/.rvm/gems/ruby-2.3.0@ey/gems/nmatrix-0.2.3/lib/nmatrix/mkmf.rb:71:in '<top (required)>'
from extconf.rb:28:in 'require'
from extconf.rb:28:in '<main>'
Just as a followup:
I checked the config for my ruby as in the description here:
https://stackoverflow.com/questions/4554837/how-to-specify-compiler-for-gem-install
and it is already set to use clang and clang++. When I type "which gcc" I get usr/bin/gcc/. When I type
gcc --version
it definitely says clang. Both in Yosemite (10.10.5) and El Capitan (10.11).