is there a way we can have this flag be autodetected instead of manually being added for anyone who wants to mruby-docopt?

I just tried without it, and it failed.

http://ruby-doc.org/core-mruby/doc/guides/compile_md.html#label-C-2B-2B+ABI

mruby can use C++ exception to raise exception internally. It is called C++ ABI mode. By using C++ exception it can release C++ stack object correctly. Whenever you mix C++ code C++ ABI mode would be enabled automatically. If you need to enable C++ ABI mode explicitly add the following: ruby conf.enable_cxx_abi

So it should be detected automatically, moreover, I've enabled it into mruby-docopt. So it seems that it's too late when it detects and enables it.

Should MRuby::Build and MRuby::CrossBuild configuration of mrbgems infect the project depending on them?

We should be able to add defines in the mrbgem.rake of mruby-docopt

Could you explain @zzak ? How does it work?

I dug into mruby code and find the following:

• When setting up the gems, it detects if any cpp files are involved and enable c++ abi
• If C++ abi is enabled, the mruby core is compiled with that feature, It consists in listing the files to compile, in particular, if cxx abi is enabled, to select vm.cxx and error.cxx rather than their C version.
• Unfortunately, the detection in the gems is done too late, as it occurs after the preparation of mruby core.
• When fixing it (see patch below), I bumped into the following issue

CC    build/host/mrbgems/mruby-compiler/core/y.tab.c -> build/host/mrbgems/mruby-compiler/core/y.tab.o

This issue is the same than previously. As the C++ code has been enabled, some files should be appropriately selected as for vm.cxx and error.cxx

• y.tab.c belongs to mruby-compiler. Its C++ version should be selected, it's not yet enabled when it comes to that gem, explaining why it fails.

The patch I've mentioned above:

diff --git a/Rakefile b/Rakefile
index 3021bcc..6b4a9ec 100644
--- a/Rakefile
+++ b/Rakefile
@@ -21,9 +21,28 @@ end

 # load custom rules
 load "#{MRUBY_ROOT}/src/mruby_core.rake"
+cxx_abi_enabled_before_gems = MRuby.each_target.inject({}) { |hsh, target| hsh[target] = target.cxx_abi_enabled?; hsh }
 load "#{MRUBY_ROOT}/mrblib/mrblib.rake"

 load "#{MRUBY_ROOT}/tasks/mrbgems.rake"
+MRuby.each_target do
+  if !cxx_abi_enabled_before_gems[self] && cxx_abi_enabled?
+    path_mruby_core_rake = "#{MRUBY_ROOT}/src/mruby_core.rake"
+    current_dir = File.dirname(path_mruby_core_rake).relative_path_from(Dir.pwd)
+    relative_from_root = File.dirname(path_mruby_core_rake).relative_path_from(MRUBY_ROOT)
+    current_build_dir = "#{build_dir}/#{relative_from_root}"
+    libmruby.each do |objs|
+      Array(objs).each do |obj|
+        if found = obj.match(/(error|vm)#{exts.object}$/)
+          filename = File.basename(found.to_s).sub(/#{exts.object}$/, "")
+          obj.replace(compile_as_cxx "#{current_dir}/#{filename}.c", "#{current_build_dir}/#{filename}.cxx")
+        end
+      end
+    end
+  end
+end
+

I don't like it at all, but it helps to understand the problem and what to do.

All of that makes me ask the following question:

• does the automatic detection works correctly for other C++ cases?
• shouldn't the C++ detection should be done through all the mruby and mrbgems before preparing the files to compile?
• which mruby contributor may I contact about that?