On Sat, Apr 13, 2024 at 7:32 PM Thiago Macieira wrote:
>
> That's great, but only if your implementation isn't hacky, non-portable and
> full of UB. It might show that the data is present so that they can implement
> it, but there are other ways of showing the data is present than to use it in
> hacky ways.


When I write a paper and provide a platform-specific "possible
implementation", I can write platform-specific code. Code that has
undefined behaviour according to the ANSI C++ Standard can have
well-defined behaviour according to the compiler and platform in
question.

But anyway more to the point: I don't expect the compiler vendor to
copy-paste my code into their compiler (which might be written in C
anyway). They can implement the new feature in assembler if they want.
The point of my "possible implementation" is to show where the data is
and how it can be gotten.