> On 19 Mar 2025, at 00:16, Thiago Macieira <thiago_at_[hidden]> wrote:
>
> On Tuesday, 18 March 2025 14:16:51 Pacific Daylight Time Hans Åberg wrote:
>> Features in a new language version that are not yet implemented can be
>> marked, and the user will know, and at need implement, as in the example in
>> my first post of this thread.
>>
>> One may need preprocessor macros so that when implemented by the compiler,
>> the ones done by the user are inhibited.
>>> Under what conditions would some code compile with the attribute, not
>>> print
>>> the warning, and would not compile with =delete?
>>
>> Compiler options can regulate that. The main point is that it is marked, so
>> that the compiler can recognize the situation.
>
> You have not answered the question. You've given a vague "behaviour could be
> different" answer. I am asking what behaviour would be different:
>
> Under what conditions would the compiler behave differently when attribute is
> used compared to =delete("message")?

GCC does nothing, so one gets a linker error with a mangled name which may be difficult to interpret. Clang uses in some case “delete” which says, in violation of the standard, that semantically it is an error to implement it, prohibiting an override.

As for an example how an override can be written see my other post.