Re: [Qemu-devel] [PATCH] qemu-thread-posix: Fix build against older glibc version

[Michael S. Tsirkin]

On Tue, Mar 11, 2014 at 05:51:16PM +0000, Peter Maydell wrote:
> On 11 March 2014 16:36, Dr. David Alan Gilbert <address@hidden> wrote:
> > * Peter Maydell (address@hidden) wrote:
> >> On 11 March 2014 16:13, Dr. David Alan Gilbert <address@hidden> wrote:
> >> > * Peter Maydell (address@hidden) wrote:
> >> >> Can we have a configure test for the function instead, please?
> >> >> We don't generally do tests for specific glibc versions
> >> >> (among other things, glibc is not the only C library we
> >> >> might be compiled against -- consider MacOSX, the BSDs,
> >> >> and occasionally somebody tries to compile against one
> >> >> of the embedded libcs).
> >> >
> >> > Except pthread_setname_np is not portable and was previously
> >> > ifdef'd _GNU_SOURCE anyway
> >>
> >> I still think it's a generally worse way to do checks.
> >> We currently have only one check against __GLIBC_MINOR__
> >> in the tree (and that's in extremely Linux-specific
> >> code).
> >>
> >> > the parameters on other OSs
> >> > maybe different (freebsd has got a 3rd parameter for no
> >> > apparent reason).
> >>
> >> This rather suggests we should abstract the "set thread
> >> name" functionality out into its own function so we
> >> can easily provide other implementations for those
> >> other OSes later.
> >
> > OK, well there is already a
> >
> > void os_set_proc_name(const char *s)
> >
> > in os-posix.c but it has ~2 problems:
> >   1) If the OS doesn't support doing it that way it exit(1)'s
> >   2) it uses prctl to set it on the current thread, where
> >      pthread_setname_np has the advantage that it sets it on
> >      something you have a pthread* for.
> >   3) There isn't a matching entry in os-win32.c
> >
> > is the best thing to add another function to os-posix and os-win
> > then?
> 
> Given this function's only for use in qemu-thread-posix.c
> I think it would be reasonable to just have it live there.
> It would be nicer to have an
> 
> static void qemu_thread_set_name(QemuThread *thread, const char *name)
> {
> #ifdef CONFIG_THREAD_SETNAME_NP
>     pthread_setname_np(thread->thread, name);
> #endif
> }
> 
> rather than #ifdefs in the middle of functions,
> that's all.
> 
> thanks
> -- PMM

Shouldn't we prevent the unsupported option from
appearing in qemu output though?
Otherwise how does user know whether it works?