[Core][Apps] Making visibility hidden by default

[roigcarlo]

📝 Description
This PR aims to unify the windows and Linux behavior making all symbols in Linux (gcc and llvm based compilers) hidden by default. This will probably help with linking times and linking restrictions (see llvm in win refusing to link targets with more than 65K symbols)

Since exporting template classes with definitions in sources works a bit different (finally i figured out how to make them visible!), some additional decorators will need to be added to several explicit instantiations.

Comments:

• Changes in find_global_nodal_neighbours_for_entities_process.cpp are not strictly necessary, but is better this way (imo).
• @KratosMultiphysics/dem DEM application was failing due to a forward declaration of the SphereElement. Please take a look (0e8e547)
• @KratosMultiphysics/mpi There was an error in the AMGCSolve function as the header was never being used. I made it so it uses the header declaration and the source definition instead of using the definition as a header in the source. Please confirm this was the intended behavior.
• @AlejandroCornejo CL is mostly yours check everything needed is exposed, I really don't know what is used from outside so I treid to make everything visible but maybe I am missing something used in other apps I don't have access to.

[loumalouomega]

You don't need to kodify KRATOS_API for non MSVC ?

[matekelemen]

Oh yes thank you! I got back into compile-heavy development again and was about to tear my nonexistent hair out during linking.

[loumalouomega]

Almost compiled, just Metis is failing apparently

[roigcarlo]

KRATOS_API

Should work as it is rn, I foresaw this when I did the Kratos api, but if some application gets grumpy I will touch it.

[roigcarlo]

It's alive

[loumalouomega]

@matekelemen I am trying to compile with Intel LLVM in Windows and your multigrid assemblers are giving issues

[matekelemen]

@matekelemen I am trying to compile with Intel LLVM in Windows and your multigrid assemblers are giving issues

They're fine on master. I don't know why this PR would trigger an ICE in intel's wonderful compiler but as a general rule, I don't think we should play around defective compilers.

[rfaasse]

To chip in here, we use Clang-LLVM (with llvm-cov) on windows for generating our coverage report in our pipelines. Since this PR seems to be close to merging, I tested it in our pipelines and for the clang-windows build (core + linear solvers + structural + geo apps) we still see some linker errors, for example (certainly not a complete list):

• lld-link: error: undefined symbol: public: static class Kratos::DataCommunicator & __cdecl Kratos::ParallelEnvironment::GetDefaultDataCommunicator(void)
• lld-link: error: undefined symbol: public: __cdecl Kratos::Exception::Exception(class std::basic_string<char, struct std::char_traits, class std::allocator> const &, class Kratos::CodeLocation const &)
• lld-link: error: undefined symbol: class Kratos::Variable Kratos::TEMPERATURE

For this particular build, we use Clang 19.1.0 with MSVC-like command-line. Is this expected @roigcarlo?

[loumalouomega]

@matekelemen I am trying to compile with Intel LLVM in Windows and your multigrid assemblers are giving issues

@matekelemen I am trying to compile with Intel LLVM in Windows and your multigrid assemblers are giving issues

They're fine on master. I don't know why this PR would trigger an ICE in intel's wonderful compiler but as a general rule, I don't think we should play around defective compilers.

Well, is not just Intel LLVM, Clang for Windows is also failing. And the reason this PR exists is making work any of those 2 compilers because MSVC is terrible kin providing properly optimized code. We have tests in internal projects that just compiling with any of those compilers instead of MSVC improves performance 20-25% out of the box. Any case we can fix compilation in independent PR.

[loumalouomega]

I need to check internally this to see if affects us.

[loumalouomega]

I cherry-picked some fixes I did in #14346, anyway I thought that maybe fix the compilation in LLVM compilers can be done in a later PR.

[loumalouomega]

Fantastic, I just broke Windows, some symbols were missing to compile locally in Ubuntu. Well at least is the last 2 commits, so easy to revert o to track