[PATCH] D51872: isPodLike: handle tuple and array

[JF Bastien via Phabricator via llvm-commits]

jfb planned changes to this revision.
jfb added inline comments.


================
Comment at: include/llvm/Support/type_traits.h:59
+template <> struct isPodLike<std::tuple<>> { static const bool value = true; };
+template <typename T> struct isPodLike<std::tuple<T>> {
+  static const bool value = isPodLike<T>::value;
----------------
ldionne wrote:
> ldionne wrote:
> > I'm not sure what the intent is, but `std::tuple` is not a POD even when its elements are -- it typically applies the EBO and I'm pretty sure it could actually use other tricks that would not make it a POD. Specifically, that would make it not Standard Layout AFAICT. 
> https://wandbox.org/permlink/uTOa0CFb15ZBQby0
Interesting! The idea is to work around GCC < 5 not having `std::is_trivially_copyable`... and `isPodLike` already says that `std::pair` matches `isPodLike` which is actually not true as you demonstrate.

My use case was `bit_cast` of `std::array` which *is* trivially copyable.

So now I have to change `isPodLike` to remove the `std::pair` overload, not add one for `std::tuple`, and do add one for `std::array`.

I'm therefore not just making it more capable, I'm also fixing a bug with `std::pair`.

To make everything worst, this is slightly inconsistent because of GCC 5! Methink we really want to upgrade minimum compiler past GCC 5. This BoF sounds like a great place to discuss it:
https://llvm.org/devmtg/2018-10/talk-abstracts.html#bof3


Repository:
  rL LLVM

https://reviews.llvm.org/D51872