Add a compatibility type for C++20's std::type_identity.

[bangerth]

We have long had a identity type. In C++20, there is finally std::type_identity we can built on instead. Import that into namespace std_cxx20 and use it to define identity.

In a follow-up patch, I will switch all uses of identity<T> to std_cxx20::type_identity<T>. (Or, rather, to std_cxx20::type_identity_t<T>.

[masterleinad]

I think we should again use a feature test macro in the C++20 check we run while configuring via CMake.

[bangerth]

@masterleinad Maybe like the last commit here?

I think a good question is what we will do when we actually require C++20. Will we continue to check for the existence of features with these feature test macros? What if they are not satisfied? This is a question separate from the current patch, but one I'd be curious to hear how you think about for the long run.

[masterleinad]

@masterleinad Maybe like the last commit here?

We discussed some time ago (#10340), that it's not a good idea, in general, to use the feature macros directly in our source code but rather while configuring using CMake.

I think a good question is what we will do when we actually require C++20. Will we continue to check for the existence of features with these feature test macros? What if they are not satisfied? This is a question separate from the current patch, but one I'd be curious to hear how you think about for the long run.

We still have checks for C++14 support and I would think that we will continue to keep those for the oldest version supported. Of course, this also heavily depends on our decision for the oldest compilers we promise to support. If all of those support a particular feature anyway, we don't have to test for it (assuming that newer compilers will not drop support), though.

[bangerth]

Ah, thanks for the reminder. How about this commit?

As for the general question: I see these checks as things that are hard to get rid of. If we still have checks for specific C++14 features, then that's likely not because we need them, but because nobody has removed them. But that is not a discussion we need to have here.

[masterleinad]

As for the general question: I see these checks as things that are hard to get rid of. If we still have checks for specific C++14 features, then that's likely not because we need them, but because nobody has removed them. But that is not a discussion we need to have here.

I would agree that it's easy to forget to remove them but it's still a backard-comaptible change if we do. 🙂
At least w.r.t the oldest C++ standard supported, these checks are mainly there to make the life of users easier and help them figure out why they can't use a specific compiler.

[masterleinad]

We should deprecate this properly when we change all uses in the library.

[bangerth]

Will do in the follow-up!

[bangerth]

OK this way?