[Lldb-commits] [libcxx] [lldb] [libcxx] adds size-based `__split_buffer` representation to unstable ABI (PR #139632)

[Louis Dionne via lldb-commits]

================
@@ -288,92 +476,82 @@ template <class _Tp, class _Allocator>
 template <class _ForwardIterator>
 _LIBCPP_CONSTEXPR_SINCE_CXX20 void
 __split_buffer<_Tp, _Allocator>::__construct_at_end_with_size(_ForwardIterator __first, size_type __n) {
-  _ConstructTransaction __tx(std::addressof(this->__end_), __n);
+  _ConstructTransaction __tx(this, __data_.end(), __n);
   for (; __tx.__pos_ != __tx.__end_; ++__tx.__pos_, (void)++__first) {
-    __alloc_traits::construct(__alloc_, std::__to_address(__tx.__pos_), *__first);
+    __alloc_traits::construct(__data_.__alloc_, std::__to_address(__tx.__pos_), *__first);
   }
 }
 
 template <class _Tp, class _Allocator>
 _LIBCPP_CONSTEXPR_SINCE_CXX20 inline void
 __split_buffer<_Tp, _Allocator>::__destruct_at_begin(pointer __new_begin, false_type) {
-  while (__begin_ != __new_begin)
-    __alloc_traits::destroy(__alloc_, std::__to_address(__begin_++));
+  pointer __begin = __data_.__begin_;
+  while (__begin != __new_begin)
+    __alloc_traits::destroy(__data_.__alloc_, std::__to_address(__begin++));
+  __data_.__update_begin(__begin);
----------------
ldionne wrote:

Although we may not have the same issue I described above with respect to exception handling since this is destroying objects (and not constructing them), I think a comment like this would be welcome:

```
// Updating __data_'s begin at every iteration is unnecessary because destruction can't throw
```

That explains why we're OK with doing things this way and doesn't pessimize anything.

https://github.com/llvm/llvm-project/pull/139632