https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88576
--- Comment #18 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Florian Weimer from comment #17)
> See integral_argument in gcc/opts-common.cc:
>
> 181 HOST_WIDE_INT
> 182 integral_argument (const char *arg, int *err, bool byte_size_suffix)
> 183 {
> 184 if (!err)
> 185 err = &errno;
> 186
> 187 if (!ISDIGIT (*arg))
> 188 {
> 189 *err = EINVAL;
> 190 return -1;
> 191 }
> 192
> 193 *err = 0;
> 194 errno = 0;
ICK.
> I agree it's rare, but it's not entirely unheard of.
Makes me think of a possible -fdirect-errno-only? It might be possible to
diagnose address-taking of it. It would still require marking of errno
accesses, of course.
Marking of errno accesses could also try to DWIM, avoid optimizing
known-errno accesses but still optimize the maybe-errno cases.
I'm not entirely sure how the frontend can effectively check for the
possible pp token (which has to be 'errno') after macro-expansion.
