A new namespace was introduced in 11.1.0 `fmt_detail`

[odygrd]

I believe code should be consistent and use the existing fmt::detail namespace. There is a new namespace that was introduced in latest version fmt_detail. Would it be possible to use the existing fmt::detail or a fmt::detail::chrono namespace ?

fmt/include/fmt/chrono.h

Line 25 in 41539c2

namespace fmt_detail {

For example when bundling libfmt under a custom namespace with a script, that is now an additional namespace that needs to be renamed

[vitaut]

We can't use fmt or fmt::detail namespace there but a subnamespace of fmt::detail could potentially work.

[odygrd]

I made a PR with the simplest change I could think:

#4328

[odygrd]

It looks like when current_zone is placed in a fmt::fmt_detail nested namespace—while also having using namespace fmt_detail;—unqualified lookup finds the template version first within fmt_detail, preventing it from considering std::chrono's function.

On the other hand, if the user-defined templated current_zone function is in a completely separate namespace, unqualified lookup doesn't find it before std::chrono's non-template version.

https://godbolt.org/z/49Tr89Pej

I also tried the following approach, completely removing the fmt_detail namespace:

template <class, class = void> struct has_tzset : std::false_type {};

template <class T>
struct has_tzset<T, void_t<decltype(_tzset())>> : std::true_type {};

template <typename, typename = void>
struct has_current_zone : std::false_type {};

template <typename T>
struct has_current_zone<T, void_t<decltype(std::chrono::current_zone())>>
    : std::true_type {};

Attempting to call std::chrono::current_zone() via SFINAE in a pre-C++20 environment fails due to the decltype check. Adding preprocessor checks like #if defined(__cpp_lib_chrono) && __cpp_lib_chrono >= 201907L and #if defined(_GLIBCXX_USE_CXX11_ABI) would likely resolve this.

I think that if someone tries to use two different versions of libfmt simultaneously in the future—like fmt v12 and fmt v11—they might run into the same collision issue with the current fmt_detail namespace, though I’m not sure how common that scenario would be.

Feel free to close this issue, I think it's not worth the effort to change. Just wanted to share my findings in case they’re useful.

Thanks again for the amazing work on this library! 🚀

[vitaut]

Replaced fmt_detail with a nested namespace in b3d45e1.