Re: Issue: #include <stdio.h> shall not cause intmax_t to be defined

[Pavel M <pavel.morozkin@gmail.com>]

> Is there actually a "MUST NOT" defined anywhere in the standards or was
this change unnecessary?
C11 specifies when header X includes header Y.
Examples:
7.25 Type-generic math <tgmath.h>:
> The header <tgmath.h> includes the headers <math.h> and <complex.h>
7.26 Threads <threads.h>:
> The header <threads.h> includes the header <time.h>
7.8 Format conversion of integer types <inttypes.h>:
> The header <inttypes.h> includes the header <stdint.h>
There are no more occurrences of "includes the header".
--
Pavel

On Wed, 4 May 2022 at 16:13, Corinna Vinschen <vinschen@redhat.com> wrote:

> On May  4 11:02, Sebastian Huber wrote:
> > On 04/05/2022 10:54, Corinna Vinschen wrote:
> > > On May  4 10:41, Sebastian Huber wrote:
> > > > On 04/05/2022 10:37, Corinna Vinschen wrote:
> > > > > On May  4 09:59, Sebastian Huber wrote:
> > > > > > On 03/05/2022 19:00, Corinna Vinschen wrote:
> > > > > > > On Apr 27 00:41, Pavel M wrote:
> > > > > > > > Hi all,
> > > > > > > >
> > > > > > > > Issue: #include <stdio.h> shall not cause intmax_t to be
> defined. However,
> > > > > > > > now it causes. This is because now <stdio.h> includes
> <sys/types.h>, which
> > > > > > > > includes <sys/_stdint.h>.
> > > > > > > > Note: per C11 the types intmax_t and uintmax_t defined in
> the header
> > > > > > > > <stdint.h>, and <stdint.h> is not included in <stdio.h>.
> > > > > > > > Consider fixing.
> > > > > > > I pushed a patch to fix this.
> > > > > > In FreeBSD, <sys/types.h> provides the stdint.h types. Could we
> bring back
> > > > > > this with
> > > > > Is that with FreeeBSD only, or is that with BSDs in general?
> > > > It seems to be a general BSD feature:
> > > >
> > > > https://github.com/openbsd/src/blob/master/sys/sys/types.h#L75
> > > >
> > > >
> http://cvsweb.netbsd.org/bsdweb.cgi/src/sys/sys/types.h?rev=1.105&content-type=text/x-cvsweb-markup
> > > If the BSDs expose stdint.h types via stdio.h anyway, what's the sense
> > > of not exposing it in the non-_BSD_VISIBLE scenario?  _BSD_VISIBLE is
> > > default anyway, so the non-exposure of the stdint types is restricted
> to
> > > files which define _XOPEN_SOURCE or some such.
> > >
> > > Is there actually a "MUST NOT" defined anywhere in the standards or
> > > was this change unnecessary?
> >
> > I think the reported problem with stdio.h is because the Newlib stdio.h
> > includes <sys/types.h>. In FreeBSD for example, <sys/_types.h> is
> included
> > with local type definitions, see:
> >
> > https://sourceware.org/pipermail/newlib/2022/019592.html
>
> I guess I'm an idiot.  I just dropped my patch and applied yours
> instead. It fixes *all* the problems requiring more patches to other
> files I saw while building with my change to sys/types.h.
>
> So, I think what I'll do is revert my patch and apply yours instead.
>
> The other patch of your patch series is obseleted by this.
>
>
> Thanks a lot!
>
>
> Corinna
>
>