Mailing list review 2021-08-16 to 2021-09-13

Some interest to provide the fix proposed by the paper.
But several people also suggested to apply a similar constraint to the pointer constructor/assignment operators/reset, or at the very least to see that explored in the paper.

Do we need a "destroying delete type trait"? The breaking changes proposed in the paper are somewhat serious and impact sensible code.

The following suggestion was provided to constrain the pointer-taking methods. One downside is that it would have false negative for incomplete types.

template<typename U>
        requires (__is_default_delete<D>::value)
        && (!same_as<const volatile U, const volatile T>)
        && is_convertible_v<U*, T*>
        && (!has_virtual_destructor_v<T>)
        unique_ptr(U*) = delete;

However we observed that the proposed design has similar issues as regard to incomplete types and we encourage the author to mention these breaking change in the next revision.

Suggestion

• Explore constraining the pointer construction functions as suggested, in addition of the default_deleter
• Explore a has_destroying_delete type trait
• Come back for a second round of mailing list review with these changes, or motivation against making them

Closing due to more than 1 year of inactivity. Please re-open if further work is expected.