Apparent C++20 compilation failure

[eddelbuettel]

Per email forwarded by BDR and originally to @stla the 'new at CRAN' package jack bonks:

This fails to install with LLVM clang 15.0.7 (current) with libc++ as its
(native) C++ library.

../Rcpp/include/Rcpp/sugar/functions/sapply.h:36:19: error: no template named
'result_of' in namespace 'std'; did you mean 'traits::result_of'?
        typedef typename ::std::result_of<Function(typename
SugarExpression::stored_type)>::type type;
                         ^~~~~~~~~~~~~~~~
                         traits::result_of
.../Rcpp/include/Rcpp/traits/result_of.h:30:8: note: 'traits::result_of'
declared here
struct result_of{
       ^
 ```
jack 5.0.0 specifies C++20, but std::result_of was removed in that standard
having been deprecated in C++17 according to
https://en.cppreference.com/w/cpp/types/result_of
and replaced by invoke_result since C++17.

[eddelbuettel]

FWIW I have not yet been able to tickle anything on Linux with clang++-15 when working with Rcpp itself and its tests.

[kevinushey]

It looks like std::result_of was removed in C++20, so perhaps that's related?

I think invoke_result is the modern alternative. https://en.cppreference.com/w/cpp/types/result_of

[Enchufa2]

A simple fix would be to add yet another branch to use invoke_result for C++17 or newer in Rcpp/sugar/functions/sapply.h.

It seems that this was added to support lambda functions in sapply. But the other users of ::Rcpp::traits::result_of (lapply, mapply and outer) still do not support lambdas. So it would be nice to move conditional compilation stuff to Rcpp/traits/result_of.h to generalize this support.

[eddelbuettel]

(Yes, BDR made that comment too. I added it to the web comment, if you replied to email you didn't see it.)

Fully agreed on the need to do this for all *apply functions.

[eddelbuettel]

The package experiencing the issue moved itself from C++20 to C++17.

[stla]

Yes. I used C++20 because it has the contains member function for maps. I replaced it with a classical find for C++17.

[eddelbuettel]

And that of course should not have failed on you. We'll look into it.

Errors apparently only come up with clang++(-15) and its own libc++.

[eddelbuettel]

I took care of the compilation issue, turning lambdas on in the *apply functions is more work ... Anyone?

[Enchufa2]

Ok to the quick patch for now. I currently do not have the time to dig deeper.

[eddelbuettel]

Yes we need to unblock this so once PRed and merged I will put a 1.0.10.2 onto the drat repo.

[eddelbuettel]

Closing this as #1248 is merged.