Hi Paul,

On 2026-02-14T12:12:02-0800, Paul Eggert wrote:
> On 2026-02-14 11:20, Alejandro Colomar wrote:
> > we can't allow that, because that would require path analysis
> 
> Then that's a fatal objection to the change proposed in alx-0087r5.

Then that'd be a fatal objection to the mere existence of the attribute.

It's more or less comparable to not being able to convert T**  to
const T**.  It's a corner case that might be safe in some cases, but
which we can't allow for safety reasons.  A programmer might be in a
better place to override the compiler (or maybe the compiler could
allow it as QoI, if it's smart enough).

> There's no such requirement for [[noreturn]] in C23, and for good reason.
> Why should there be such a requirement for [[reproducible]]?

Indeed, there's discussion on [[noreturn]] going on at the moment, both
in WG14 and WG21.  Don't consider any existing standard attributes to be
settled or good.


Cheers,
Alex

-- 
<https://www.alejandro-colomar.es>

Attachment: signature.asc
Description: PGP signature

Reply via email to