niedz., 7 kwi 2019 o 14:10 Peter Dimov via Boost napisał(a):

Bjorn Reese wrote:

...

The destructor should be marked as noexcept.

Destructors are noexcept by default.

Actually, that is not the case. The noexcept property of a destructor without explicit noexcept annotation is determined by inspecting noexcept property of destructors of non-static data members of the class and its base classes. Since these subobjects are private implementation of your class, and we are not aware of how their destructors are implemented, we cannot determine whether variant's destructor is noexcept or not. Here's an example: ``` class Detail { ~Detail() noexcept(false); }; class Variant { Detail _detail; }; static_assert(std::is_nothrow_destructible_v<Variant> == false); ``` Regards, &rzej;

pon., 8 kwi 2019 o 16:24 Peter Dimov via Boost napisał(a):

Andrzej Krzemienski wrote:

...

niedz., 7 kwi 2019 o 14:10 Peter Dimov via Boost napisał(a):

...

Bjorn Reese wrote:

...

The destructor should be marked as noexcept.

Destructors are noexcept by default.

Actually, that is not the case. The noexcept property of a destructor without explicit noexcept annotation is determined by inspecting noexcept property of destructors of non-static data members of the class and its base classes. Since these subobjects are private implementation of your class, and we are not aware of how their destructors are implemented, we cannot determine whether variant's destructor is noexcept or not.

This is an interesting point. However, the standard doesn't mark the destructors as noexcept, and I'm not sure I see why this same logic doesn't apply to it.

http://eel.is/c++draft/variant.dtor http://eel.is/c++draft/optional.dtor http://eel.is/c++draft/any.cons#23 http://eel.is/c++draft/unique.ptr.single.dtor

There is a global requirement for this: http://eel.is/c++draft/requirements#res.on.exception.handling-3 Regards, &rzej;

On 8/04/2019 01:07, Peter Dimov wrote:

Bjorn Reese wrote:

...

What about my questions about a tutorial and design rationale? Those are genuine questions. I am trying to understand if their absence is deliberate or due to lack of time.

The absence of a tutorial and a design rationale are not deliberate, in the sense that I don't believe that the documentation is better without them. Lack of time is more correct.

Absence of a tutorial is fine (I'm sure most people know how variants are supposed to work in general, especially as you're using mostly-compatible interface and behaviour as std::variant), but more motivation and/or design rationale would be nice. Perhaps in lieu of a more focused design rationale section (given time constraints), you could just include a link to http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2016/p0308r0.html (as recently posted by Michael)? I assume that your initial arguments there still apply to this implementation, and it seems like useful background information and/or motivation. (Although it then goes on to talk about unrelated things, and obviously at least some of your thinking has changed over time.) And probably most users (like me) would be unaware of the paper otherwise.