buffer::set_final_data *should* accept weak_ptr<T[]>

[tomdeakin]

KhronosGroup/SYCL-CTS#1011

The WG discussed that this is an oversight in the spec

[fknorr]

What is the set of types you propose?

• Any output iterator
• std::weak_ptr<T>
• std::weak_ptr<T[]>

at least I suppose. The CTS and DPC++ currently also requires / implements

• std::shared_ptr<T[]>

and if this is intentional, maybe

• std::shared_ptr<T>

should be added as well, with identical semantics to passing it as a buffer ctor argument?

[gmlueck]

We did not discuss this in depth in the WG today, so these are my own thoughts.

It seems like we should define separate overloads of set_final_data that take weak_ptr<T> and weak_ptr<T[]>. By defining them as separate overloads (vs. constraints on the template parameter), the user can pass anything that is implicitly convertible to these types. For example, an application could pass shared_ptr<T> because there is an implicit conversion from that to weak_ptr<T>.

We probably also want an overload which is a function template, where the template parameter is constrained to an output iterator to T.

I looked quickly at the DPC++ implementation. We do not have a special overload for shared_ptr<T>. Rather the template parameter just has no constraint. As a result, you can pass anything, include shared_ptr<T>.