[systemd-devel] compile with clang broken

[Lennart Poettering]

On Fri, 15.08.14 12:40, Thomas H.P. Andersen (phomes at gmail.com) wrote:

> 
> On Fri, Aug 15, 2014 at 12:35 PM, David Herrmann <dh.herrmann at gmail.com> wrote:
> > Hi
> >
> > On Fri, Aug 15, 2014 at 12:29 PM, Thomas H.P. Andersen <phomes at gmail.com> wrote:
> >> On Fri, Aug 15, 2014 at 11:49 AM, David Herrmann <dh.herrmann at gmail.com> wrote:
> >>> Thanks for trying!
> >>>
> >>> Result is as I expected. Evaluation takes place _after_ validating
> >>> compile-time constants, and thus __builtin_constant_p in combination
> >>> with ?: will not work if not both cases are constant. Maybe it works
> >>> with __builtin_choose_expr()?
> >>>
> >>> Can you try the attached patch?
> >>
> >> This patch works. It also needs the change to do the calculation to a
> >> seperate line. Also only if size is const, like so:
> >> const size_t size = MAX(sizeof(struct in_pktinfo), sizeof(struct in6_pktinfo));
> >
> > Again, thanks for trying it out!
> 
> no problem. I have inserted the relevant error messages for the two
> non-working cases.
> 
> > I don't understand your comment, though. You're saying this works:
> >
> >         const size_t size = MAX(...);
> >         uint8_t buffer[CMSG_SPACE(size) +...];
> >
> > ...but this doesn't work:
> >
> >         uint8_t buffer[CMSG_SPACE(MAX(...)) +...];
> 
> src/resolve/resolved-dns-stream.c:67:43: error: non-const static data
> member must be initialized out of line
>                 uint8_t buffer[CMSG_SPACE(MAX(sizeof(struct
> in_pktinfo), sizeof(struct in6_pktinfo)))
>                                           ^
> 
> > ...and this doesn't work either (mind the dropped 'const'):
> >
> >         size_t size = MAX(...);
> >         uint8_t buffer[CMSG_SPACE(size) +...];
> 
> src/resolve/resolved-dns-stream.c:68:25: error: fields must have a
> constant size: 'variable length array in structure' extension will
> never be supported
>                 uint8_t buffer[CMSG_SPACE(size)
>                         ^
> 
> > Hm. This is weird. Maybe CMSG_SPACE does something weird. I'll see.
> > David

In this case I really think we should just detect LLVM and define MAX()
and MIN() to something weaker that doesn't use ({ ... }) expressions...

Lennart

-- 
Lennart Poettering, Red Hat