bug#9000: patch for higher-resolution time stamps

[Eli Zaretskii <eliz@gnu.org>]

> Date: Sat, 23 Jun 2012 11:39:40 +0300
> From: Eli Zaretskii <eliz@gnu.org>
> Cc: 9000@debbugs.gnu.org
> 
> > Date: Fri, 22 Jun 2012 14:21:50 -0700
> > From: Paul Eggert <eggert@cs.ucla.edu>
> > CC: Eli Zaretskii <eliz@gnu.org>
> > 
> > No further comment and the patch seems ripe
> > so I installed it into the trunk as bzr 108687
> > and am marking this bug as done.  I'll CC: this
> > to Eli as it affects the Microsoft ports.
> 
> This badly broke the MS-Windows build.  I'm working on fixing it.

I think I fixed this in revision 108698, MS-Windows users please see
if there are any leftovers.  In particular, any feature that uses
system time or file time or 'select' (i.e. subprocesses) is suspect.
I only did some minimal testing.

Paul, please note that I needed changes in 3 gnulib headers:
utimens.h, timespec.h, and stat-time.h.  The change is to include
sys/time.h, because that's where the Windows build defines
'struct timespec' (time.h cannot be used for that).  I hope you can
make these changes in upstream gnulib.

I'm OK with any other solution to this conundrum, which will be
compatible with the following conditions:

 . do not assume that 'struct timespec' is defined on time.h, or on
   any other MS-Windows system header
 . do not use gnulib's time.h (as that requires Sed to edit time.in.h)

It is OK to add headers to Emacs's nt/inc/ directory in order to solve
this, provided that they do not shadow system headers (since MSVC does
not have include_next or its equivalent).

Finally, I see compiler warnings due to the changes in revno 108687:

  dispnew.c: In function `Fsleep_for':
  dispnew.c:5960: warning: comparison is always true due to limited range of data type
  dispnew.c: In function `sit_for':
  dispnew.c:6008: warning: comparison is always true due to limited range of data type

  keyboard.c: In function `kbd_buffer_get_event':
  keyboard.c:3861: warning: comparison is always true due to limited range of data type

  process.c: In function `Faccept_process_output':
  process.c:3999: warning: comparison is always true due to limited range of data type

They all come from snippets such as this one:

    min (EMACS_SECS (t), INTMAX_MAX)

I think any host with a signed 32-bit time_t will see this warning.
Can this annoyance be fixed, please?