pthread_cond_clockwait glibc forwarder (was Re: [RFCv4] Add pthread_cond_timedwaitonclock_np)

[Mike Crowe <mac at mcrowe dot com>]

[pthread_cond_clockwait was previously known as
pthread_cond_timedwaitonclock_np.]

On Monday 02 October 2017 at 19:26:35 -0300, Adhemerval Zanella wrote:
> >>> diff --git a/nptl/Versions b/nptl/Versions
> >>> index 0ae5def464..1cf8c2fbad 100644
> >>> --- a/nptl/Versions
> >>> +++ b/nptl/Versions
> >>> @@ -28,6 +28,9 @@ libc {
> >>>      pthread_cond_wait; pthread_cond_signal;
> >>>      pthread_cond_broadcast; pthread_cond_timedwait;
> >>>    }
> >>> +  GLIBC_2.26 {
> >>> +    pthread_cond_timedwaitonclock_np;
> >>> +  }
> >>
> >> The libc addition seems wrong, whhy this is required?
> > 
> > pthread_cond_timedwait is in libc, so I followed suit and added
> > pthread_cond_timedwaitonclock_np too. I must admit that I can't quite see
> > how the dummy empty implementations of the pthread functions end up in
> > libc, but they apparently do according to:
> > 
> > https://stackoverflow.com/questions/11161462/why-glibc-and-pthread-library-both-defined-same-apis
> > 
> > Maybe this means that I've missed something.
> 
> This is an GLIB implementation detail, where some libraries which
> are expected to be usable without explicit linking to libpthread use
> some pthread functions (for instance aio from librt). 
> 
> In the single thread case libc.so uses wrappers that may or not call the
> libpthread.so implementation (nptl/forward.c) if libpthread.so.0 is
> loaded (by dlopen for instance).
> 
> So since there is no current use of pthread_cond_timedwaitonclock_np,
> there is no requirement to add it on libc.so.

I was digging through some old review comments, and came across this. It
looks like I didn't address it. :( afe4de7d283ebd88157126c5494ce1796194c16e
added forwarders for __pthread_cond_clockwait and pthread_clock_clockwait
to glibc:

$ nm /lib64/libc.so.6 |grep _clockwait
000000000007f410 t __pthread_cond_clockwait
000000000007f410 t pthread_cond_clockwait

Should we remove them now, before they are included in a release and it's
too late? (I tried removing them from forward.c, pthread-functions.h and
nptl-init.c and all tests passed on x86_64.)

Alternatively, if they should stay, do they need adding to the libc part of
nptl/Versions?

Thanks, and sorry for not discovering this earlier.

Mike.