Support new formatter rules introduced in fmt v10 and v11

[ArmstrongCN]

This PR updates our codebase to comply with the new formatting rules and API changes introduced in fmtlib v10+ while keeping the compatity of fmt v8.

Key changes include:

1. add default enum formatter that without stringfy impl.
2. fix some non-activated stringfy struct.

relate issue: #173

[ArmstrongCN]

I have formatted the commit, please review again.

the commit is tested with both fmt v8.0.1(the original) and fmt v11.1.4(the newest). Updating the fmt also requires a compatible version of spdlog.

Changes:

1. make custom formatter const
2. make default enum formatter for non-stringfy type
3. fmt::join is moved to <fmt/ranges.h>
4. some source files are not properly include the enum formatter

[hulibruce]

Thank you for your contribution. We are excited to announce an open-source rewards event—every valid merge request qualifies for a reward! For more details, please join the Feishu and WeChat groups listed in our documentation or contact us directly. All prizes will be distributed collectively at the end of the event.
https://bytedance.larkoffice.com/docx/YJSVdaSaIo6pmgxf9WgcZcvwnge?from=from_copylink

[JackLau1222]

First of all, thanks very much for your great work.
However, this multi-version compatibility issue involves API changes. It may be more appropriate to get the version number in the cmake build and then implement it according to the version in the code.

[ArmstrongCN]

The patch is compatible with fmt versions ranging from v8 to the latest v11. Therefore, I believe there is no need to modify the CMake build script.

outdated

[ArmstrongCN]

UPDATE: support both char and wchar_t

[JackLau1222]

there're some conflicts with latest commit, may you rebase this patch?

[ArmstrongCN]

rebase to main with no conflicts

[hulibruce]

In the GPU environment, compilation fails with fmt versions 10.0.0 and 11.0.2. It's likely that some fmt functions used exclusively in the GPU context require version-specific adaptations.

[ArmstrongCN]

include the header <hmp/format.h> resolves the cudaError_enum issue.

[hulibruce]

still have compilation errors when using fmt 10.0.0 (tag a0b8a92e3d1532361c2f7feb63babc5c18d00ef2) in the GPU environment

[ArmstrongCN]

still have compilation errors when using fmt 10.0.0 (tag a0b8a92e3d1532361c2f7feb63babc5c18d00ef2) in the GPU environment

Add the line set(SPDLOG_FMT_EXTERNAL ON CACHE BOOL "SPDLOG_FMT_EXTERNAL" FORCE) to ignore the fmt bundled with spdlog which cause the conflict.

By the way, this may probably reduce the final package size. I GUESS.

[h-vetinari]

Hi! 👋

This patch would be very helpful for conda-forge, where we generally only build for one fmt version across the entire ecosystem, and we've been on fmt v11 for a long time already. The fact that bmf is not yet compatible with this makes it difficult to keep the package up-to-date and usable.

I've tried backporting this patch in conda-forge/bmf-feedstock#36 (on top of v0.1.0), and it still fails.