344 implement user traits

[Matthew-Whitlock]

Closes #344

[nmm0]

@Matthew-Whitlock will this require some changes to VT?

[cz4rs]

@Matthew-Whitlock PR tests with vt build is currently failing in the configuration stage and reports success incorrectly. #352 fixes both these issue, please rebase once it is merged.

[nmm0]

About half reviewed

[nmm0]

@Matthew-Whitlock Can you rebase and mark as ready for review?

[nmm0]

Whew, sorry it took so long I added a bunch comments

[nmm0]

Here and elsewhere in the PR make sure the casing matches the rest of magistrate, i.e. CamelCase for struct names

[Matthew-Whitlock]

Happy to make this change, but I think underscore casing makes sense for type manipulation and template helpers. The rest of the code is a bit inconsistent, though.

serializable_traits.h has:

template <typename U>
using specialized_reconstruct_t = decltype(
  checkpoint::CheckpointReconstructor<U>::reconstruct(
    std::declval<U*&>(),
    std::declval<void*>()
  )
);
using has_specialized_reconstruct =
  detection::is_detected<specialized_reconstruct_t, T>;

//But also this style
template <typename U, typename V>
using serializeThis_t =
  decltype(std::declval<U>().serializeThis(std::declval<V&>()));
using has_serializeThis = detection::is_detected<serializeThis_t, T, S>;

checkpoint_traits.h:

template <typename U>    
using contiguousBytes_t = decltype(std::declval<U>().contiguousBytes(    
  std::declval<void*>(), std::declval<SizeType>(), std::declval<SizeType>()    
));    
using has_contiguousBytes = detection::is_detected<contiguousBytes_t, T>;

reconstructor_traits.h:

template <typename T>
using isDefaultConsType =
  std::enable_if_t<std::is_default_constructible<T>::value>;

[nmm0]

Hah, I'll defer to @lifflander then for this one :P

[nmm0]

TraitPairA is a little confusing as they don't seem the be pairs?

[Matthew-Whitlock]

The naming is based on the check in UserObjectA that expects to have either both or neither of TraitPairA and TraitPairB

[nmm0]

should this test also check name to ensure that the UserObjectA was called correctly

[Matthew-Whitlock]

We can, but I don't think that's checking for anything that wouldn't already be caught by test_serialize_extra so I'd rather keep the checks within the test's scope

[Matthew-Whitlock]

With the extra check for correct trait detection, we should be fine leaving the name checking to the test_serialize_extra test

[nmm0]

same here