On 2023-08-28 01:42, Alejandro Colomar wrote:
> Hi,
>
> You seem to want a portable way of specifying unportable semantics to
> the language.
>
> That does exist. Or kind of. It's in #pragma and [[vendor::attribute]].
> That as far as the standard will take you by the hand. You need to cross
> the door alone. At the other side of the door, your compiler vendor will
> hopefully take your hand, if they are kind.
>
> You should be asking your compiler vendor to provide a vendor specific
> attribute or pragma that says "hey, the standard be ashes. I want some
> other semantics" In GCC, I guess you have what you want with
>
> $ g++ -fno-strict-overflow -fno-strict-overflow -fno-trapv -fno-strict-aliasing

One of those was meant to be -fwrapv.

>
> That's an entire new dialect of C++ that some programmers believe to be
> fine, and blame compiler authors for taking advantage of UB just because
> they can. That's fine, if you need it, just use it. If you want to
> enable those flags within the source code, you'll probably have pragmas
> that enable it just for a few specific lines, although I don't do that
> myself, so I didn't look that up.
>
> The standard provides minimal tools to specify "my code invokes UB, I
> know, and I won't fix it, so you better compile my code in my own
> dialect of C++ without going crazy". I don't think it should extend
> more than that, especially, when it's up to the compiler vendors to
> define such a new dialect of C++, in an agreement between them and you.
>
> These things are more often used in C, and not in C++, but hey, it's free.
>
> Cheers,
> Alex