On Tue, Apr 4, 2023 at 3:57 PM Marcin Jaczewski via Std-Proposals
<std-proposals_at_[hidden]> wrote:
>
> wt., 4 kwi 2023 o 18:12 Barry Revzin <barry.revzin_at_[hidden]> napisał(a):
> >
> >
> >
> > On Tue, Apr 4, 2023, 10:59 AM Marcin Jaczewski via Std-Proposals <std-proposals_at_[hidden]> wrote:
> >>
> >> I wonder why not make infinite loops simply explicit?
> >> Like:
> >>
> >> ```
> >> while (std::infiniti) { }
> >> ```
> >>
> >> Now it will be clear to the reader and compiler that we expect it to
> >> be infinite.
> >> Reading `std::infiniti` would be considered an observable behavior or something
> >> like that.
> >>
> >> Alternative add attribute to signal to everyone that this loop should not break.
> >> We could even reuse attribute:
> >>
> >> ```
> >> [[noreturn]] while(true) {}
> >> ```
> >>
> >> This would have benefits that compilers could more aggressively warn
> >> if they see a loop that can't break and therefore create UB.
> >
> >
> > Because the goal is to make all the existing infinite loops non-UB, rather than add some new construct that people would have to go and update all their [not actually broken] code to.
> >
> > Barry
>
> It's hard to say that is "not broken" as it breaks a contract with the
> compiler, of course we can argue who is right there.
> Another thing, how do we know if the given loop was not an error? If
> we try to guess what the intention was then we can guess wrongly,
> better would be to allow the user to mark his intention to the
> compiler and other programmers.

The proposal in question only makes it not-UB for very specific cases:
the case of a condition which is a constant expression that converts
to the boolean `true`. I suggest that it's *very* difficult to write
that by accident. Not impossible, but hardly something people do every
day.