Using nameof_member on structures with non-constexpr destructor fails to compile on MSVC

[kamchatka-volcano]

Hi, I finally got around to updating nameof in my project to test the fix of #46 by @schaumb.
Unfortunately, it doesn't work when TestStruct has a member with a non-constexpr user-provided constructor. Is it possible to fix it?
I looked into the source, but I don't understand the union trick used in the member_name_v implementation for MSVC, so I can't tell if it can be fixed.

As previously, the following code doesn't compile on MSVC (2022) using the latest master of nameof, but works fine with clang and gcc

struct Foo{
    ~Foo(){};
};

struct TestStruct{
    std::string teststringfield = std::string{nameof::nameof_member<&TestStruct::teststringfield>()};
    Foo foo;
};

[schaumb]

Hi @kamchatka-volcano !
The current solution works only if the TestStruct destructor is constexpr, the default for trivially_destructible classes, and if all member destructor is constexpr.

So If Foo's destructor is constexpr, then it will work.

I Investigate any other solution for not constexpr destructor, but I think this is an MSVC compiler or C++std bug. It must not to call members user-defined destructor in the union, but the constexpr-ness property is inherited somewhy, based on the compiler error

TODO doc!

note:
the union trick is: We create a constexpr static variable, which is a union with a trivial active type (char) and the class (T) of the owner of the member. We cannot use the not active member (T), but we can query the member pointer of the constexpr union variable. This "constexpr relative" memory information is passed through a __FUNCSIGN__, and it will contain the name. So we won't create the TestStruct, and shouldn't be called the destructor either.
sidenote: this hack is needed because __FUNCSIGN__ is not contain the member pointer name default, only the offset

[kamchatka-volcano]

I see, that's what I was afraid of) In my project TestStruct is user-provided and the requirement of constexpr dtr is a pretty heavy restriction.
Anyway, thanks a lot for the detailed explanation, now it's much more clear what's going on there. I'll try to think of some workaround as well.

[kamchatka-volcano]

Hi, I think I've found a working solution by using union type as a static member of a class. https://godbolt.org/z/dzKcdfKzP

I opened a PR (#50), please check it out)