[Bug libstdc++/60421] std::this_thread::sleep_for doesn't sleep for all arguments

["redi at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=60421

Jonathan Wakely <redi at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|5.0                         |6.0

--- Comment #7 from Jonathan Wakely <redi at gcc dot gnu.org> ---
You're right, sorry, I ended up reducing the scope of the fix and it only deals
with PR58038, see https://gcc.gnu.org/ml/libstdc++/2015-03/msg00078.html for
some commentary.

At one point I added:

  if (__s < chrono::seconds::zero() || __s < __rtime)
    __s = chrono::seconds::max();

which would detect some cases where the duration_cast to chrono::seconds
overflows, but that overflow is still undefined behaviour so I didn't include
that in the commit. I want to fix it properly to avoid any undefined behaviour
(probably converting to duration<long double> and comparing to
duration<time_t>::max()).

> I mean when I sleep for UINT64_MAX hours/years/millenia, you can't possibly
> wrap that into a single nanosleep call due to the limitations of the type
> time_t of the tv_sec parameter of the first argument to nanosleep. One
> obviously can not get around using loop.

To be honest, I'm not very concerned about the failure to sleep for 290 billion
years, if we sleep for duration<time_t>::max() instead of
duration<uint64_t>::max() and don't loop you'll never know the difference.

> Additionally, the nanosleep code is also missing proper EINTR handling,
> which again could break the sleep.

Yes, see the mailing list post above.

I'll deal with this for gcc 6.