Fix test suite C++ test with GCC 5.1 dual ABI

[CyberShadow]

GCC 5.1 introduced new implementations of std::string and std::list:

https://gcc.gnu.org/onlinedocs/libstdc++/manual/using_dual_abi.html

This causes e.g. std::string to be actually defined as std::__cxx11::string.

On machines with GCC 5.1, this manifests as a linker error when running the cppa.d / cppb.cpp test:

cppa.o: In function `_D4cppa6test14FZv':
cppa.d:(.text._D4cppa6test14FZv+0x11): undefined reference to `foo14a(std::string*)'
cppa.d:(.text._D4cppa6test14FZv+0x18): undefined reference to `foo14b(std::basic_string<int, std::char_traits<int>, std::allocator<int> >*)'
cppa.d:(.text._D4cppa6test14FZv+0x3a): undefined reference to `foo14f(std::char_traits<char>*, std::string*, std::string*)'
cppa.o: In function `_D4cppa7testeh3FZv':
cppa.d:(.text._D4cppa7testeh3FZv+0x19): undefined reference to `throwle()'
collect2: error: ld returned 1 exit status
--- errorlevel 1

When the .cpp file is compiled with g++ 5.3.0, the actual function signatures in the cppb.o object file are:

foo14a(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >*)
foo14b(std::__cxx11::basic_string<int, std::char_traits<int>, std::allocator<int> >*)
foo14f(std::char_traits<char>*, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >*, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >*)

Fortunately, it is easily possible to disable the new feature by defining _GLIBCXX_USE_CXX11_ABI as 0 before including any standard headers.

[jpf91]

IIRC @ibuclaw also had a pull request for this issue?

[CyberShadow]

I don't see anything relevant on https://github.com/dlang/dmd/pulls/ibuclaw

[jpf91]

I was thinking of #5262 but it seems that has already been merged and probably wasn't meant to fix this problem.

[Geod24]

Maybe #5261 ?

[WalterBright]

You have a great explanation for this in the opening comment. But the code just contains the #define. Please add the explanation as a comment!

[CyberShadow]

OK, I copied the commit message as a comment before the #define.

[WalterBright]

Auto-merge toggled on

[ibuclaw]

Nope, I hadn't addressed the C++-11 problem. Maybe we can discuss a compromise between supporting the current compiler and supporting legacy at dconf. Because this is also hurting overall gdb experience too.

I honestly have half a mind in support of using something similar to -std=xxx to allow switching between ABIs.

[MartinNowak]

If at all it would need to be -c++-std ;).
How about we try to detect which C++ compiler gets used? We could insist on declaring CC=myc++ when linking against C++ code. That would also obsolete the problematic usage of -L-lstdc++ with a C compiler and provides a solution for the next C++ ABI exceptions (e.g. Intel compiler).

[MartinNowak]

BTW, it took me 15m. to figure out that stable tests fail on some auto-testers b/c of this, and another 30m. to find this PR.
Remember, safe fixes should always go into stable, all the time.