Re: Using macro CMSG_NXTHDR generates warnings with CLANG

[Szabolcs Nagy <nsz@port70.net>]

* Rich Felker <dalias@libc.org> [2016-10-11 11:09:01 -0400]:
> On Mon, Oct 10, 2016 at 10:09:38PM +0000, Jan Vorlicek wrote:
> > The testing source is below:
> > 
> > #include <sys/socket.h>
> > cmsghdr* GET_CMSG_NXTHDR(msghdr* mhdr, cmsghdr* cmsg);
> > 

this is invalid c code (you cannot leave 'struct' off).

> > cmsghdr* GET_CMSG_NXTHDR(msghdr* mhdr, cmsghdr* cmsg)
> > {
> >     return CMSG_NXTHDR(mhdr, cmsg);
> > }
> > 
> > Would it be possible to fix it so that no warnings are generated? We
> > are building our application with -Weverything and currently we need
> > to disable these two warnings around the CMSG_NXTHDR macro
> > invocation.
> > Thank you in advance for considering that!
> 
> As these are system headers, the compiler should not be producing any
> warnings from them. If it does that's a compiler bug. Are you perhaps
> using an odd setup where musl's headers aren't in the default system
> include path but instead passed in via -I rather than -isystem? If you
> have a minimal test file I could see if the same warnings appear with
> clang on Alpine Linux.
> 

fwiw i see the warnings with clang -c -Wcast-align test.c

i assume they cannot easily tell if they should warn when a macro
defined in a system header expanded with user arguments.

i think this warning is problematic: it warns about casting char*,
but not about casting void* which has the exact same issue.

so if you have apis that uses char* instead of void* for generic
pointers (e.g. for abi compat reasons), then you get loads of
false positives (which is what happens here).

it can be worked around by adding a void* cast (but in my
opinion that makes things worse: users will add more casts
to get rid of the warning which has the opposite effect than
what you would want).