Hi Jakub, I investigated the source of the macro further.
I am not defining `__BFLT16_DIG__` in the project command line, in a user header, or in the source before including libstdc++ headers. The diagnostic is produced by CLion 2023.3.6's bundled JetBrains clangd 18 when parsing a WSL project configured with `/usr/bin/c++` (GCC 14.2.0). `std::format` is first marked as erroneous in the user source; following that diagnostic into `<format>` leads to: ``` else if constexpr (is_same_v<_Td, decltype(0.0bf16)>) ``` where clangd reports: ``` invalid suffix 'bf16' on floating constant ``` Consequently, uses that depend on `<format>`, including `make_format_args`, are also marked as erroneous. I enabled CLion's clangd protocol logging and found that CLion supplies GCC's predefined macro set to clangd through an `-imacros` file. The actual clangd command contains: ``` -imacros /mnt/x/clion_clangd_inmemory_directory/macros/... ``` and the contents of that virtual macro file include: ``` #define __GNUC__ 14 #define __GNUC_MINOR__ 2 #define __BFLT16_MANT_DIG__ 8 #define __BFLT16_DIG__ 2 ``` Thus the `__BFLT16_DIG__`-guarded code is visible to clangd because CLion injects the predefined macros collected from the selected GCC toolchain. The same JetBrains clangd run standalone with an ordinary GCC-style compile command, without CLion's injected `-imacros` file, does not define `__BFLT16_DIG__` and does not reach this code. So the issue is a mismatch: clangd is given GCC's bfloat16 capability macros but does not recognize the corresponding `bf16` literal suffix. Separately, I understand your point that `__bf16` cannot be assumed to be the type of `0.0bf16` on every target. My proposed change to `c++config` was checked on x86_64. And maybe the previous commit is better? Sincerely, Jakit Jakub Jelinek <[email protected]> 于2026年9月7日周一 22:16写道： > > On Mon, Sep 07, 2026 at 09:59:04PM +0800, Jakit Liang wrote: > > Tested on x86_64-linux-gnu with g++ and clangd 18/19. > > > > Signed-off-by: Jakit Liang <[email protected]> > > --- > > libstdc++-v3/include/bits/c++config | 2 +- > > libstdc++-v3/include/std/format | 2 +- > > 2 files changed, 2 insertions(+), 2 deletions(-) > > > > diff --git a/libstdc++-v3/include/bits/c++config > > b/libstdc++-v3/include/bits/c++config > > index 60456a48681..072da451df8 100644 > > --- a/libstdc++-v3/include/bits/c++config > > +++ b/libstdc++-v3/include/bits/c++config > > @@ -892,7 +892,7 @@ namespace std > > #if defined __cplusplus && defined __BFLT16_DIG__ > > namespace __gnu_cxx > > { > > - typedef __decltype(0.0bf16) __bfloat16_t; > > + typedef __bf16 __bfloat16_t; > > This is incorrect. > > __bf16 is the type of 0.0bf16 literals only on some targets, not all. > You haven't explained why is __BFLT16_DIG__ defined when compiling with > clangd, it shouldn't be when it doesn't support bfloat16_t properly. > Are you defining it by hand on the command line or in some user header > or main source before including the libstdc++ headers? > > Jakub >
