CMake: Find dependencies in QUIET mode from pagmo-config

[jschueller]

only when pagmo is used as a client library, not when configuring pagmo itself

[bluescarni]

@jschueller thanks for the PR, LGTM.

Regarding the clang>=17 issues on OSX, this seems hairy and I will try to explain to the best of my understanding what is going on.

The issue is a std::bad_cast exception being thrown during (de)serialisation of problems/algorithms/etc. In pagmo, we employ the boost::serialization library, which we use to (de)serialize derived classes via pointers to their bases. For this to work, RTTI is used, and boost::serialization needs to register a caster between base and derived classes in some internal global state, and to perform std::dynamic_casts at runtime.

This has been working great for years in pagmo, but somehow, for reasons I cannot understand, it has stopped working since the clang 17 upgrade. I have run in the same issue in another project of mine, where I also have started pinning clang on osx:

https://github.com/conda-forge/heyoka.py-feedstock/blob/main/recipe/conda_build_config.yaml

According to my testing in this other project, the issue arises only on x86_64, apparently osx_arm64 is fine.

It looks like this is a quite difficult issue to debug and pin down. Perhaps someone with deep toolchain knowledge (pinging @isuruf and @h-vetinari ) could chime in...

[bluescarni]

Ok maybe not so difficult to diagnose after all. See:

https://releases.llvm.org/17.0.1/tools/clang/docs/ReleaseNotes.html#c-language-changes

Improved code generation for dynamic_cast to a final type. Instead of dispatching to the runtime library to compare the RTTI data, Clang now generates a direct comparison of the vtable pointer in cases where the ABI requires the vtable for a class to be unique. This optimization can be disabled with -fno-assume-unique-vtables. This optimization is not yet implemented for the MS C++ ABI.

And:

llvm/llvm-project#71196

I will try the suggested workaround...

[bluescarni]

@jschueller would you like to rebase/merge master back into this PR so I can merge this?

[jschueller]

done

[bluescarni]

Cheers!