TU-local entity error question

[Thomas Khyn]

Indeed, it looks like it's exactly the same issue. Thanks a lot!

Thomas

On 23/02/2025 10:38, Jonathan Wakely wrote:
> On Tue, 18 Feb 2025 at 09:11, Thomas Khyn via Gcc-help
> <gcc-help@gcc.gnu.org> wrote:
>> Hi,
>>
>> I am trying to build {fmt} as a C++20 module with GCC-15 (15.0.1 20250214).
>>
>> I get an error that neither me nor the {fmt} maintainer can make sense
>> of. We are unsure if it's a code issue or a GCC issue (other compilers
>> work fine, however they do not implement P1815
>> <https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2020/p1815r2>).
>>
>> This is about the following code fragment (os.cc L72
>> <https://github.com/fmtlib/fmt/blob/master/src/os.cc#L72>, curated from
>> #ifdefs below), which ends up being included in fmt.cc:
>>
>> namespace {
>> // Return type of read and write functions.
>> using rwresult = ssize_t;
>>
>> inline std::size_t convert_rwcount(std::size_t count) { return count; }
>> }  // namespace
>>
>> The compiler reports an error for the line using rwresult = ssize_t; :
>>
>> /usr/bin/c++ -I/mnt/d/dev/libs/tools/cpp/fmt/include -O3 -DNDEBUG
>> -std=gnu++20 -fvisibility=hidden -fvisibility-inlines-hidden
>> -fdiagnostics-color=always -fmodules-ts -MD -MT
>> CMakeFiles/fmt.dir/src/fmt.cc.o -MF CMakeFiles/fmt.dir/src/fmt.cc.o.d
>> -fmodules-ts -fmodule-mapper=CMakeFiles/fmt.dir/src/fmt.cc.o.modmap -MD
>> -fdeps-format=p1689r5 -x c++ -o CMakeFiles/fmt.dir/src/fmt.cc.o -c
>> /mnt/d/dev/libs/tools/cpp/fmt/src/fmt.cc
>> In file included from /mnt/d/dev/libs/tools/cpp/fmt/src/fmt.cc:148:
>> /mnt/d/dev/libs/tools/cpp/fmt/src/os.cc:74:7: error: ‘using
>> {anonymous}::rwresult = ssize_t’ exposes TU-local entity ‘{anonymous}’
>>      74 | using rwresult = ssize_t;
>>         |       ^~~~~~~~
>> /mnt/d/dev/libs/tools/cpp/fmt/src/os.cc:62:1: note: ‘{anonymous}’
>> declared with internal linkage
>>      62 | namespace {
>>         | ^~~~~~~~~
>>
>> My shallow understanding of the standard
>> <https://timsong-cpp.github.io/cppwp/n4659/basic.link#8> and P1815
>> <https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2020/p1815r2> is
>> that it looks like gcc is correct to disallow this code (anonymous
>> namespace is exposed as TU-local because rwresult has linkage because
>> ssize_t is a fundamental type).
>>
>> However, when working around this error (either by moving the `using`
>> statement within the functions bodies or by creating a macro for
>> `rwresult`
>> <https://github.com/fmtlib/fmt/commit/9381bba8d967428dd4829b2fcc0c32199ef41ed3#diff-261f435b8f5004993798511d1dc333e917901cbb2e83e73592d5bd2f3082c7bc>),
>> gcc is happy with `convert_rwcount`, which should also have internal
>> linkage and therefore expose the anonymous namespace as TU-local and
>> therefore report the same error.
>>
>> Would anyone know what am I missing? If GCC is correct about
>> convert_rwcount then I would like to understand if there is anything we
>> can do to rwresult to make it happy.
> This is apparently https://gcc.gnu.org/PR118799
>
>
>> Thanks in advance,
>>
>> Thomas
>>
>>
>> PS: full steps to reproduce the error, with GCC 15.0.1 20250214
>> (installed in /opt/gcc-dev) and cmake 3.31.5:
>>
>> git clone -b gcc-15-fixes-test https://github.com/tkhyn/fmt
>> mkdir fmt/build && cd fmt/build
>> export CXX=/opt/gcc-dev/bin/g++
>> export CC=/opt/gcc-dev/bin/gcc
>> cmake .. -DFMT_MODULE=ON
>> cmake --build . --target fmt
>>
>>
>> To make the error disappear, make these changes
>> <https://github.com/fmtlib/fmt/commit/9381bba8d967428dd4829b2fcc0c32199ef41ed3#diff-261f435b8f5004993798511d1dc333e917901cbb2e83e73592d5bd2f3082c7bc>
>> to os.cc.