Build fails at hex_editor.cpp on CachyOS (Arch derivative)

[emersonrp]

Building from a fresh pull of master on CachyOS, I get as far as hex_editor.cpp, and it fails thusly:

[733/1067] CXX src/gui2/+ui-plugin/dep/r/imhex/plugins/ui/source/ui/hex_editor.cpp
FAILED: [code=1] .obj/unix/src/gui2/+ui-plugin/dep/r/imhex/plugins/ui/source/ui/hex_editor.cpp/hex_editor.o 
sh .obj/unix/src/gui2/+ui-plugin/dep/r/imhex/plugins/ui/source/ui/hex_editor.cpp/rule-efda040e78a66a82ae0341d15859ee7c7d20ff5b.sh
In file included from .obj/unix/src/gui2/+libimhex_core_hdr/hex/api/content_registry/hex_editor.hpp:7,
                 from .obj/unix/src/gui2/+ui-plugin_hdr/ui/hex_editor.hpp:4,
                 from dep/r/imhex/plugins/ui/source/ui/hex_editor.cpp:1:
.obj/unix/src/gui2/+libimhex_core_hdr/hex/api/localization_manager.hpp: In instantiation of ‘auto fmt::format(const hex::Lang&, Args&& ...) [with Args = {long unsigned int&}]’:
dep/r/imhex/plugins/ui/source/ui/hex_editor.cpp:1222:202:   required from here
 1222 |                                 ImGui::SliderScalar("##separator_stride", ImGuiDataType_U64, &m_separatorStride, &min, &max, m_separatorStride == 0 ? "hex.ui.hex_editor.no_separator"_lang : fmt::format("hex.ui.hex_editor.separator_stride"_lang, m_separatorStride).c_str());
      |                                                                                                                                                                                               ~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
.obj/unix/src/gui2/+libimhex_core_hdr/hex/api/localization_manager.hpp:169:27: error: no matching function for call to ‘format(fmt::v12::runtime_format_string<>, long unsigned int&)’
  169 |         return fmt::format(fmt::runtime(entry.get()), std::forward<Args>(args)...);
      |                ~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  • there is 1 candidate
    • candidate 1: ‘template<class ... Args> auto fmt::format(const hex::Lang&, Args&& ...)’
      .obj/unix/src/gui2/+libimhex_core_hdr/hex/api/localization_manager.hpp:168:10:
        168 |     auto format(const hex::Lang &entry, Args &&... args) {
            |          ^~~~~~
      • template argument deduction/substitution failed:
        •   cannot convert ‘fmt::v12::runtime(fmt::v12::basic_string_view<char>((& entry)->hex::Lang::get()))’ (type ‘fmt::v12::runtime_format_string<>’) to type ‘const hex::Lang&’
          .obj/unix/src/gui2/+libimhex_core_hdr/hex/api/localization_manager.hpp:169:40:
            169 |         return fmt::format(fmt::runtime(entry.get()), std::forward<Args>(args)...);
                |                            ~~~~~~~~~~~~^~~~~~~~~~~~~
[742/1067] CXX src/gui2/+ui-plugin/dep/r/imhex/plugins/ui/source/ui/pattern_drawer.cpp
ninja: build stopped: subcommand failed.
make: *** [build/ab.mk:126: +all] Error 1

My C++ fu is feeble so I don't have much clue where to start applying pressure.

[VorpalBlade]

I'm seeing the same on Arch Linux (unsurprising, Cachy is a downstream of Arch after all).

My guess is that this is due to a newer version of some dependency, probably fmt. Arch Linux packages fmt 12.2.0-1.

As I'm trying to update the AUR package which I maintain I will spend some time looking into this.

[VorpalBlade]

Not seeing what is wrong here (and I really don't like doing complex C++ as a hobby any more, I have mostly moved to Rust).

[davidgiven]

C++ is getting painfully complicated and I'm seriously considering rewriting everything in something better. Like, anything. What's most likely going on here is a version incompatibility between whatever C++ version it is and libfmt. Try upgrading both as far as you can and see if that helps.

[VorpalBlade]

My personal suggestion would be Rust (if you care about having a similar performance profile), but most important for a hobby project is to pick whatever is fun to code in for you.

[VorpalBlade]

I tried upgrading to C++26 in the Makefile (HOSTCXX) without success. All of C++20, C++23 and C++26 exhibit the same behaviour. Arch appears to have the latest libfmt as of writing this.

I hope that is the right place to change things, I cannot claim that I understand your custom build system (nor do I really feel to digging into how a makefile calls python code to generate Ninja build files).

What I can say is that changing just the C++ version in the makefile did not cause ninja to rebuild existing files, I had to make clean in between. This suggest some missing dependency tracking, or perhaps that I edited the wrong place.

[emersonrp]

I'm not really adding anything here, just confirming that ensuring I have the latest everything from the Cachy repos (which track the Arch repos closely) doesn't affect anything as far as this compile failure. fmt is version 12.2.0-1.1

[VorpalBlade]

So, someone posted (to a comment for the AUR PKGBUILD of fluxengine) a patch, that they said was Claude generated. I'm not sure about your AI policy, but the fix looks trivial enough, changing the include of fmt/core.h to fmt/format.h in localization_manager.hpp inside the cloned imhex repo. There is a much longer (presumably AI generated) comment explaining why this is needed in the PKGBUILD.

Based on the sort of weird errors I have seen using fmt at my dayjob, I'm not surprised that changing the include fixes an obscure error.