[Feature] Add C++20 modules support

[iTrooz]

Hello !

I'm trying to use CLI11 with C++20 modules, but I'm getting tons of errors when trying to pass a CLI::App through the module's interface. I think they are mostly about the dummy variable in TypeTools.hpp being TU-local due to being constexpr.

Full error log is available here: https://pastebin.com/PJYpZhAV, but it's mostly a spam of this:

/bin/g++   -std=gnu++23 -fmodules-ts -MD -MT CMakeFiles/math_mod.dir/src/math.ixx.o -MF CMakeFiles/math_mod.dir/src/math.ixx.o.d -fmodules-ts -fmodule-mapper=CMakeFiles/math_mod.dir/src/math.ixx.o.modmap -MD -fdeps-format=p1689r5 -x c++ -o CMakeFiles/math_mod.dir/src/math.ixx.o -c /home/itrooz/tmp3/CppModules/src/math.ixx
In file included from /usr/include/CLI/CLI.hpp:26,
                 from /home/itrooz/tmp3/CppModules/src/math.ixx:3:
/usr/include/CLI/TypeTools.hpp:456:6: error: ‘template<class T1, class T2, class T, typename std::enable_if<std::is_same< <template-parameter-1-1>, <template-parameter-1-2> >::value, CLI::detail::enabler>::type <anonymous> > decltype (CLI::detail::to_string(forward<T>(value))) CLI::detail::checked_to_string(T&&)’ exposes TU-local entity ‘CLI::detail::dummy’
  456 | auto checked_to_string(T &&value) -> decltype(to_string(std::forward<T>(value))) {
      |      ^~~~~~~~~~~~~~~~~
/usr/include/CLI/TypeTools.hpp:39:19: note: ‘CLI::detail::dummy’ declared with internal linkage
   39 | constexpr enabler dummy = {};
      |                   ^~~~~

Example program here: https://github.com/iTrooz/CppModules/tree/cli11, compile (with errors) with CC=/bin/gcc CXX=/bin/g++ cmake -G Ninja -B build && (cd build && ninja)

I am using gcc 15.2.1, cmake 4.1.2 and latest CLI11 commit (c992112), on ArchLinux

This problem does not happen on clang, probably because of llvm/llvm-project#112294

[phlptp]

can you try something? Change the typetools lines to

enum class enabler : std::uint8_t {null};

/// An instance to use in EnableIf
constexpr enabler dummy = {enabler::null};

I haven't worked with modules much, but it might be the declaration of an empty struct forcing internal linkage. If that doesn't work will probably have to play around with it some more, I have not worked much with modules yet.

[iTrooz]

Just tested, and nope, same error ):

You may want to check nlohmann/json#4764. In short, the author removed the static keywords in their code. I think that in this case, constexpr might be the culprit.

I'm not sure if the implications, but adding inline to dummy, as well as some other places in the code make the build work.

[phlptp]

I am going to try to get your example set up on a system with gcc 15 and see if I can replicate the errors and try out different ways to fix it. Might be a few days before I can get that tested.

[phlptp]

@iTrooz any issues if I use some of the code you referenced as a build test in CLI11?

[phlptp]

What I am running into is that there is no way to have the code compile with C++11 and be compatible with modules, so there will be a compile definition and CMAKE definition in place to switch out some of the keywords, which I think will make it work.

[iTrooz]

@iTrooz any issues if I use some of the code you referenced as a build test in CLI11?

No problem ! Code is mostly not from me but is under the MIT licence

that there is no way to have the code compile with C++11 and be compatible with modules,

I have no problem with the cmake compile definitions, but I'd just like to point out that nlohmann_json seems to be compatible with C++11 and have module support. I didn't look at the code though, no idea if such a thing is possible here