On Mon, Jan 04, 2021 at 08:48:55PM +0100, Otto Moerbeek wrote:
> On Mon, Jan 04, 2021 at 05:50:53PM +0100, Otto Moerbeek wrote:
>
> > tOn Mon, Jan 04, 2021 at 01:42:48PM +0100, Theo Buehler wrote:
> >
> > > > > + return log_sockaddr(addr2sa(addr, 0, &len), len);
> > > >
> > > > Perhaps I haven't yet had enough coffee this year, but I'm unsure
> > > > whether it is actually guaranteed that addr2sa() is called before the
> > > > second len in this line is passed to log_sockaddr().
> > >
> > > Answering my own question: C99 and C11 6.5.2.2.12 require that all side
> > > effects must be completed before log_sockaddr() is called. As addr2sa()
> > > changes the second len as a side effect, this should be fine.
> > >
> > > ok tb
> > >
> >
> > I am not convinced. Consider
> >
> > #include <stdio.h>
> >
> > char x;
> >
> > char f(char *p)
> > {
> > *p = 'f';
> > return 'r';
> > }
> >
> > int main()
> > {
> > x = 'm';
> > printf("%c %c %c\n", x, f(&x), x);
> > }
> >
> > prints "m r f" here. The first x is not influenced by the call to
> > f(), while the second is. So the side effetc only affects one of the args.
> >
> > -Otto
> >
> >
>
> And if you compile with gcc on amd64 you'll get "f r m".
>
> In my reading, the C standard is clear; the order of evaluation of the
> arguments is undefined. The side effects should indeed take effect
> before calling the function, but that's something different, e.g.
> before the side effetcs are complete, some of the args may already be
> on the stack,
Yes. Thanks for the example and the clarifications. I agree that the
spec is clear that the order is unspecified. I fooled myself into
thinking the identity function should somehow be special.
Anyway, this will be fixed by claudio's follow-up on tech.
