[PATCH] Cleanup std::inplace_vector implementation

François Dumont frs.dumont@gmail.com
Mon Sep 29 17:30:14 GMT 2025


Now that your part is in here is the new patch for the small cleanup left.

     libstdc++: std::inplace_vector implementation cleaup

     Remove duplicated std::swap implementations.

     libstdc++-v3/ChangeLog

             * include/std/inplace_vector:
             (std::swap(inplace_vector<>&, inplace_vector<>&)): Remove 
the duplicated
             implementation at std namespace level. Keep the friend 
inline one.
             (inplace_vector<Tp, 0>::assign(initializer_list<>): Add 
missing return
             statement.


Tested under Linux x86_64,

ok to commit ?

François


On 9/22/25 20:04, Jonathan Wakely wrote:
> On Mon, 22 Sept 2025 at 18:15, François Dumont <frs.dumont@gmail.com> wrote:
>> Some minor points spotted while working on the _GLIBCXX_DEBUG
>> implementation.
>>
>> I'm surprised that gcc did not complained about the duplicated std::swap
>> implementation. He seems to be smart enough to see that it is the same
>> despite the different noexcept qualification.
> No, they're not the same. The friend one is a non-template function
> which is defined for every std::inplace_vector specialization, and the
> one at namespace scope is a function template.
>
> It's still correct to remove the one at namespace scope, but GCC is
> right to not complain.
>
>>       libstdc++: std::inplace_vector implementation cleaup
>>
>>       Remove duplicated std::swap implementations.
>>
>>       Remove use of std::ref in std::erase_if.
>>
>>       libstdc++-v3/ChangeLog
>>
>>               * include/std/inplace_vector: Remove refwrap.h include.
>>               (std::swap(inplace_vector<>&, inplace_vector<>&)): Remove
>> the duplicated
>>               implementation at std namespace level. Keep the friend
>> inline one.
>>               (std::erase_if): Remove std::ref usage.
>>
>> 23_containers/inplace_vector tests run under Linux x86_64.
>>
>> Ok to commit ?
> No, I'm already changing erase_if and erase in
> https://forge.sourceware.org/gcc/gcc-TEST/pulls/78/commits/4d4a7dd885afca17a526caf504463310b3159587
> which I'm about to push.
>
-------------- next part --------------
diff --git a/libstdc++-v3/include/std/inplace_vector b/libstdc++-v3/include/std/inplace_vector
index 91ceace08f5..7aa6f9d4ab2 100644
--- a/libstdc++-v3/include/std/inplace_vector
+++ b/libstdc++-v3/include/std/inplace_vector
@@ -823,6 +823,7 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
 						      __detail::__synth3way);
       }
 
+      // [inplace.vector.special], specialized algorithms
       constexpr friend void
       swap(inplace_vector& __x, inplace_vector& __y)
       noexcept(is_nothrow_swappable_v<_Tp> && is_nothrow_move_constructible_v<_Tp>)
@@ -907,13 +908,6 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
 	}
     };
 
-  // [inplace.vector.special], specialized algorithms
-  template<typename _Tp, size_t _Nm>
-    constexpr void
-    swap(inplace_vector<_Tp, _Nm>& __x, inplace_vector<_Tp, _Nm>& __y)
-    noexcept(noexcept(__x.swap(__y)))
-    { __x.swap(__y); }
-
   // specialization for zero capacity, that is required to be trivally copyable
   // and empty regardless of _Tp.
   template<typename _Tp>
@@ -992,6 +986,7 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
       {
 	if (__il.size() != 0)
 	  __throw_bad_alloc();
+	return *this;
       }
 
       template<__any_input_iterator _InputIterator>


More information about the Libstdc++ mailing list