[libcxx-commits] [libcxx] [z/OS][libc++] Guard to be used only when `align_val_t` is available (PR #114396)

[Louis Dionne via libcxx-commits]

================
@@ -66,16 +66,24 @@ class __small_buffer {
     if constexpr (__fits_in_buffer<_Stored>) {
       return std::launder(reinterpret_cast<_Stored*>(__buffer_));
     } else {
+#  ifdef _LIBCPP_HAS_LIBRARY_ALIGNED_ALLOCATION
       byte* __allocation = static_cast<byte*>(::operator new[](sizeof(_Stored), align_val_t{alignof(_Stored)}));
       std::construct_at(reinterpret_cast<byte**>(__buffer_), __allocation);
       return std::launder(reinterpret_cast<_Stored*>(__allocation));
+#  else
+      return static_cast<_Stored*>(std::__libcpp_allocate(_BufferSize * sizeof(byte), _LIBCPP_ALIGNOF(byte)));
+#  endif // _LIBCPP_HAS_LIBRARY_ALIGNED_ALLOCATION
----------------
ldionne wrote:

I think what we want is this:

```c++
return static_cast<_Stored*>(std::__libcpp_allocate(_BufferSize * sizeof(byte), alignof(_Stored)));
```

We're always passing the right alignment, but we let `__libcpp_allocate` discard it based on `_LIBCPP_HAS_LIBRARY_ALIGNED_ALLOCATION`. So we don't need any `#if` here.

https://github.com/llvm/llvm-project/pull/114396