From 2ef83da01b9a8849fb94c32608819b51bbbb91a7 Mon Sep 17 00:00:00 2001 From: Damien L-G Date: Wed, 22 May 2024 17:43:45 -0400 Subject: [PATCH] libstdc++: the specialization atomic_ref should use the primary template Per [[atomics.ref.int]](https://eel.is/c++draft/atomics.ref.generic#atomics.ref.int) `bool` is excluded from the list of integral types for wich there is a specialization of the `atomic_ref` class template and [*Note 1*](https://eel.is/c++draft/atomics.ref.generic#atomics.ref.int-note-1) clearly states that `atomic_ref` "uses the primary template" instead. ```c++ void test() { bool x = true; std::atomic_ref const a{x}; a &= false; // error should indicate that the compiler // could not find a matching 'operator&=' // but currently fails in the instantiation // of implementation details. } ``` I also expect that one should be able to detect that all the `fetch_` member functions and other arithmetic operators are not valid expressions. Maybe that could be the basis for testing the correct behavior is implemented. See https://godbolt.org/z/1eor9TbYj Signed-off-by: Damien L-G --- libstdc++-v3/include/bits/atomic_base.h | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/libstdc++-v3/include/bits/atomic_base.h b/libstdc++-v3/include/bits/atomic_base.h index 062f1549740..7c342d3bcd0 100644 --- a/libstdc++-v3/include/bits/atomic_base.h +++ b/libstdc++-v3/include/bits/atomic_base.h @@ -1478,7 +1478,8 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION #undef _GLIBCXX20_INIT template, bool = is_floating_point_v<_Tp>> + bool = is_integral_v<_Tp> && !is_same_v<_Tp, bool>, + bool = is_floating_point_v<_Tp>> struct __atomic_ref; // base class for non-integral, non-floating-point, non-pointer types -- 2.39.3 (Apple Git-146)