Re: AC_C_INLINE with -Werror llvm 3.4

[Eric Blake]

On 06/09/2014 09:59 AM, Kurt Schwehr wrote:
> Eric,
> 
> Thanks for the feedback.  I definitely had the feeling that it would be
> impossible to get rid of all warnings for all compilers.
> 
> Two questions back at you:
> 
> 1) Do you have a example of a source package that handles -Werror in a
> clean way so that I can use it a s model try to add the same to these other
> packages?

Sure.  GNU coreutils:

http://git.savannah.gnu.org/cgit/coreutils.git/tree/configure.ac#n108
http://git.savannah.gnu.org/cgit/coreutils.git/tree/src/local.mk

Note how it probes for -Werror separately from other flags, and sticks
the result in a different variable for substitution at make time.  At
maketime, it then avoids -Werror for gnulib (which has a looser set of
standards, and therefore fails some of the warning flags that coreutils
turns on for itself), and enables -Werror for itself (all according to
whether warnings-as-errors was requested based on configure options,
which in turn have a default based on whether it detects a build from
git sources or from a tarball).

> 
> 2) Does the suggested fix I gave make sense?  I think that means: a) It's
> not going to break any environments without -Werror or equivalent for
> compilers out there. b) It will generally reduce the number of
> compilers/flag combinations that would issue a warning for this test?  If
> this proposed fix is a bad idea, why is it bad?

Your fix is a one-liner, and easy to maintain, once you resubmit it
without the syntax error of a missing (); so I see no reason against
applying your patch.  I'm just pointing out that expecting -Werror to
work during configure is a lost cause, and that while we can apply
simple patches, we cannot solve the inherent technical problem of being
able to make ALL compilers warning-free.

-- 
Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org

signature.asc
Description: OpenPGP digital signature