I wrote code yesterday to manipulate VTables:

    #ifdef _MSC_VER
        struct VTable {
            RTTICompleteObjectLocator *locator;
            void *funcptrs[1u];
        };
    #else
        struct VTable {
            std::uintptr_t top_offset;
            std::type_info *p_type_info;
            void *funcptrs[1u];
        };
    #endif

    static_assert( std::is_standard_layout_v<VTable> );
    constexpr std::size_t vtable_prologue_size = offsetof(VTable, funcptrs);

See how I use the 'static_assert' to be sure that the 'offsetof' won't
have undefined behaviour? Well I think the Standard should be changed
so that if you use 'offsetof' on an inappropriate type, then you get a
compile-time static_assert failure rather than undefined behaviour at
runtime.

There might be a handful of other things in the Standard that could be
changed this way too. Better to have static_assert's failing at
compile time.