On Sun, Mar 19, 2017 at 02:01:42PM +0000, Antti Kantee wrote:
> On 19/03/17 13:20, Francesco Lattanzio wrote:
> > librumpuser's configure script checks the availability of the 2 and 3
> > arguments versions of the pthread_setname_np() function. The checks rely
> > on the compiler ability to emit a warning when an undefined function in
> > invoked.
> > This works on Clang and recent versions of GCC, but older versions of
> > GCC (e.g., 4.9.x) don't emit -- by default -- the warning, making the
> > checks to succeed unconditionally.
> >
> > The fix is as simple as adding -Wimplicit-function-declaration along
> > side the -Werror option to the CFLAGS variable.
> 
> So what's the failure mode?  Both checks succeed?  Because if you're 
> building in a env without pthread_setname_np() at all, the link phase of 
> the test should fail, making the compiler warnings irrelevant.  So what 
> the test actually relies on is throwing an error for a prototype mismatch.
> 
> Or are you building for an env with pthread_setname_np() in libpthread, 
> but no prototype for it in pthread.h?
> 
> I'm reluctant to change configure.ac in src-netbsd unless I'm absolutely 
> sure -Wimplicit-function-declaration is the right thing for your env, 
> and has no side effects for other compilers (e.g. it's not a supported 
> flag for some).
> 
> Can you post the relevant output from config.log under obj/tooldir/autoconf?

Please, forget the AC_TRY_LINK thing -- it was a stupid idea.

-- 
Francesco Lattanzio

Reply via email to