Compile all arbor source files with -fvisibility=hidden

[noraabiakar]

The visibility of symbols in the python shared library is set to 'hidden'. However, checking the dynamic symbol table of the generated .so file reveals that all the symbols of the static arbor libraries are still visible. This causes some issues on the Mac M1. To resolve, all source files should be compiled with fvisibility=hidden.

Fixes #1557
Fixes #1559

[noraabiakar]

bors try

[bors]

try

Build failed:

• ci/gitlab/gitlab.com

[noraabiakar]

bors try

[bors]

try

Build failed:

• ci/gitlab/gitlab.com

[noraabiakar]

bors try

[bors]

try

Timed out.

[noraabiakar]

bors try

[bors]

try

Build succeeded:

• ci/gitlab/gitlab.com

[kanzl]

Works for me on M1 Macbook

[thorstenhater]

Looks good, also works on my non-M1 MacBook. A tiny standards nitpick and we are good to merge.

[thorstenhater]

Since C++17 we are allowed to use [[gnu::visibility("default")]] as part of the standard. The __attribute__ is technically an extension by GCC.

[noraabiakar]

But that would only work for GCC. According to this: https://gcc.gnu.org/wiki/Visibility, we should add this entire block:

#if defined _WIN32 || defined __CYGWIN__
  #ifdef BUILDING_DLL
    #ifdef __GNUC__
      #define DLL_PUBLIC __attribute__ ((dllexport))
    #else
      #define DLL_PUBLIC __declspec(dllexport) // Note: actually gcc seems to also supports this syntax.
    #endif
  #else
    #ifdef __GNUC__
      #define DLL_PUBLIC __attribute__ ((dllimport))
    #else
      #define DLL_PUBLIC __declspec(dllimport) // Note: actually gcc seems to also supports this syntax.
    #endif
  #endif
  #define DLL_LOCAL
#else
  #if __GNUC__ >= 4
    #define DLL_PUBLIC __attribute__ ((visibility ("default")))
    #define DLL_LOCAL  __attribute__ ((visibility ("hidden")))
  #else
    #define DLL_PUBLIC
    #define DLL_LOCAL
  #endif
#endif

I think we need that top block for WSL, but I'm not sure and I don't have a system to test it on.

[thorstenhater]

That page predates the C++17 attribute (last change 2016). We also can savely assume GCC > 4 (else C++14 would be tricky). And finally, we do not care about Windows right now. In endeffect it comes down to what you have in the PR.

Regarding your comment, this is working in clang and gcc. To quote my last comment __attribute__ is a GCC extension in itself. However, Clang shares most of GCC's extensions. Reference: https://gcc.godbolt.org/z/Edev9PGnK

[espenhgn]

Hi. As I raised #1557 I tested this PR out locally, and it appears to have sorted out the problem(s) I encountered there. Thanks!