RE: [PATCH 2/3] Move ftime to a compatibility symbol

[Tamar Christina via Libc-alpha <libc-alpha@sourceware.org>]

> -----Original Message-----
> From: Florian Weimer <fweimer@redhat.com>
> Sent: Monday, October 19, 2020 10:51 AM
> To: Tamar Christina <Tamar.Christina@arm.com>
> Cc: libc-alpha@sourceware.org
> Subject: Re: [PATCH 2/3] Move ftime to a compatibility symbol
> 
> * Tamar Christina:
> 
> >> -----Original Message-----
> >> From: Libc-alpha <libc-alpha-bounces@sourceware.org> On Behalf Of
> >> Florian Weimer via Libc-alpha
> >> Sent: Monday, October 19, 2020 9:31 AM
> >> To: Szabolcs Nagy via Libc-alpha <libc-alpha@sourceware.org>
> >> Subject: Re: [PATCH 2/3] Move ftime to a compatibility symbol
> >>
> >> * Szabolcs Nagy via Libc-alpha:
> >>
> >> > The 10/15/2020 10:06, Adhemerval Zanella via Libc-alpha wrote:
> >> >> It was made deprecated on 2.31, so it moves to compat symbol after
> >> >> two releases.  It was also removed from exported symbol for
> >> >> riscv32 (since ABI will be supported on for 2.33).
> >> >>
> >> >> Checked on x86_64-linux-gnu and i686-linux-gnu.
> >> >> ---
> >> >>  include/sys/timeb.h                           |  1 -
> >> >>  .../unix/sysv/linux/riscv/rv32/libc.abilist   |  1 -
> >> >>  time/Makefile                                 |  5 +-
> >> >>  time/ftime.c                                  | 19 +++++-
> >> >>  time/sys/timeb.h                              | 44 --------------
> >> >>  time/tst-ftime.c                              | 59 ++++++++++---------
> >> >>  6 files changed, 50 insertions(+), 79 deletions(-)  delete mode
> >> >> 100644 include/sys/timeb.h  delete mode 100644 time/sys/timeb.h
> >> >
> >> > removing sys/timeb.h and ftime breaks several
> >> > spec2006 and spec2017 benchmarks.
> >> >
> >> > which means this header is used in practice, sometimes without
> >> > configure checks.
> >> >
> >> > is there a reason it cannot be supported?
> >>
> >> I think we should minimize the number of time-related symbols, in
> >> case we ever want to provide a dual ABI (with multiple time_t sizes).
> >>
> >> SPEC isn't going to change, so I don't think it makes sense to wait
> >> for it getting fixed.
> >
> > That's a pretty hard line to draw.. Yes released version of spec are
> > unlikely to change but doesn't mean future versions won't. Has anyone
> > ever pointed this out to the committee?
> 
> Has SPEC ever fixed any standards violations?  I don't have access to the SPEC
> web site, so I don't know if they have released patches to fix conformance
> bugs in benchmarks.  (I recall several discussions about aliasing violations in
> older benchmarks on the GCC lists.)

They do, and have in the past, but while I can reasonably get one for SPEC2017,
SPEC2006 is no longer maintained as far as I am aware.

> 
> > Regardless the issue now exists that we can't use glibc for benchmark
> > anymore and I don't really see a workaround.  Yes we can use older
> > glibc versions but then what's the point, we'll never see any of the
> > improvements we make anymore.
> 
> How does SPEC run on platforms that do not have ftime?  Even on UNIX-like
> systems, using ftime presently requires linking with some compatibility
> library.  Is there really no knob to use gettimeofday or clock_gettime instead
> of ftime?

If the header was still there but empty I could do something at link time, but
since it doesn't exist at all I have to provide my own local headers.  Also if you
have a list of interfaces glibc plans to remove we can also feed this on to them
to avoid in the next SPEC release.  https://www.spec.org/cpuv8/

Most of the SPEC2017 benchmarks do the right thing with a include guard,
but xalancbmk doesn't, which was presumable taken directly from the
2006 version which also fails (along with h264 in 2006).

I will contact the committee but it seems highly unlikely that they would
Issue a fix for 2006.

Regards,
Tamar

> 
> Thanks,
> Florian
> --
> Red Hat GmbH, https://de.redhat.com/ , Registered seat: Grasbrunn,
> Commercial register: Amtsgericht Muenchen, HRB 153243, Managing
> Directors: Charles Cachera, Brian Klemm, Laurie Krebs, Michael O'Neill