Build ROOT with C++23

[YanzhaoW]

Hello,

I would like to suggest to build ROOT program with C++23 standard. Currently it's built with C++17 standard. If our library is moving to C++23, ROOT cannot be used with a higher standard than the one it's built with.

[fuhlig1]

Hi,

if I am correct the required minimum standard is 17. You should be able to overwrite this passing -DCMAKE_CXX_STANDARD=23 when calling cmake.

[YanzhaoW]

I see. Thanks. Would it be also possible to do this in the cvmfs?

[YanzhaoW]

@fuhlig1 Ok, I tried to compile with C++23, but it doesn't work:

https://github.com/YanzhaoW/Dockerfiles/actions/runs/16077461308/job/45375832946

The cause is CLHEP, which is, as expected, another old software from CERN. Do you know why we need this library?

[YanzhaoW]

Update: The ROOT itself isn't supporting C++23 at the moment. I guess we should just wait a little bit longer until it's supported in ROOT.

[fuhlig1]

Thanks for your investigations. CLHEP is a package providing mathematical routines used by Geant4 so we need it.

Concerning ROOT, which ROOT version did you use?

[YanzhaoW]

Hi,

I'm using the latest version of ROOT. That's 6.36.0 at the moment.

[YanzhaoW]

Sorry, I just check the Github CI again. Yes, ROOT with 6.36.0 is working with C++23.

The one that doesn't work has the ROOT version 6.34.6.

[YanzhaoW]

Just a side note if everyone wants to use ROOT in C++23 (I guess it applies to C++ 20 as well):

Don't use range-v3 or any format_to in any headers that are used by Cling. This is because in ROOT Cling, it has this evil stuff:

using namespace std;

Thus any format_to from libfmt or ranges from range-v3 will directly conflict with std::format_to and std::ranges.

I don't know how to get around this with ROOT Cling.

[jose-luis-rs]

Hi, I also had problems with C++23. It seems to be an issue related to FairSoft, since C++23 is not fully supported when certain functions are generated during compilation. I had to add them manually...

[YanzhaoW]

Hi @jose-luis-rs

Thanks for the info.

I would like to point out that the problem could come from two kinds of compilation: the first is the compilation of the native C++ code (both headers and sources) into libraries or executables; the second is the compilation of ROOT Cling (it only compiles header files) to the ROOT dictionaries.

From my experience, the problem usually comes from ROOT Cling, where it has some nasty features, e.g. using namespace std. I have been looking around in the internet for the solution and it seems the best we could do are:

1. Use ifndef __CLING__, like

#ifndef __CLING__
template <>
class fmt::formatter<R3B::MilleDataPoint>
{
  public:
    static constexpr auto parse(format_parse_context& ctx) { return ctx.end(); }
    template <typename FmtContent>
    constexpr auto format(const R3B::MilleDataPoint& point, FmtContent& ctn) const
    {
        return fmt::format_to(ctn.out(),
                              "measurement: {}, sigma: {}\nlocals: {}\nglobals: {}",
                              point.measurement,
                              point.sigma,
                              point.locals,
                              point.globals);
    }
};
#endif

This will make ROOT Cling ignore the code.

2. Don't push all headers to the ROOT Cling. So in R3BRoot, we have the CMake variable ${HEADERS} for all header files. For normal C++ project without generating ROOT dictionaries, it is not necessary to let CMake know any header files. So if some header files are not needed by ROOT Cling, we'd better not to include them in ${HEADERS}.

3. Put the Cling-hostile code in the source file as ROOT Cling only compiles header files.

[fuhlig1]

Hi,

with the release candidate for the next FairSoft release the problem should be fixed. I was able to compile FairSoft setting the CMAKE_CXX_STANDARD to 17, 20 and 23. I did not encounter any problems beside CLHEP not compiling with C++23 which was a config issue which is fixed in FairSoft using a patch. A merge request was already created upstream but up to now there was no action on that side.

The code is currently in a branch of my repository

https://github.com/fuhlig1/FairSoft/tree/RC_aug25

This RC also updates to the latest versions of ROOT (6.36.04) and Geant4 (11.3.2) and also fixes the CMake and other issues. If have tested the compilation for Fedora 38, 40 and 42 as well as Debian 11 and 12 (vae24 and vae25) and haven't encountered any problems. Also test for macosx did not show any problems. For the time being the installations were not tested using FairRoot or some experiment code.

Maybe you can test it and provide some feedback. Please be aware that the commit history will change in near future to properly arrange the changes done during development

Ciao

Florian

[YanzhaoW]

Hi, @fuhlig1

Yeah, I could also compile FairSoft with C++23 without CLHEP. But it seems that there is a runtime conflict between ROOT and libfmt, which is used by fairlogger. Please check this issue: root-project/cling#553