[libcxx-commits] [PATCH] D81174: [libcxx] Return "real" pointer from array<T, 0>::data.

[Zoe Carver via Phabricator via libcxx-commits]

zoecarver marked an inline comment as done.
zoecarver added inline comments.


================
Comment at: libcxx/include/array:250
+    ~__array_storage_wrapper() = default;
+    _LIBCPP_CONSTEXPR __array_storage_wrapper(__array_storage_wrapper const& __other)
+        : __t(__other.__t) { }
----------------
ldionne wrote:
> zoecarver wrote:
> > ldionne wrote:
> > > I think this is UB: we're only ever constructing the `__b` member, yet the copy-constructor copies the `__t` member, which hasn't been initialized. I think this triggers UB when we copy an empty array, which ends up calling this copy constructor on a union member that hasn't been initialized yet.
> > > 
> > > WDYT?
> > Yes, I think you're right. `memcpy` is probably the way to go here. I'll use `__builtin_memcpy` so it's still a constexpr on clang. 
> I mean, even if you use `memcpy`, it's UB. You can't summon a non trivially copyable object into existence merely by `memcpy`ing it, it's still UB to access the object afterwards IIUC.
> 
> You can try, but I would expect that Clang is going to error out if you try to take a pointer to `__t` if it has been "initialized" via `memcpy` -- which in that case means it hasn't been initialized properly -- in a constexpr context.
Is `memcpy`ing an object any different than copying a char array of its data? I'
ll give it a try in a constexpr context. 


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D81174/new/

https://reviews.llvm.org/D81174