bug#9000: patch for higher-resolution time stamps

[Eli Zaretskii]

> Date: Sat, 23 Jun 2012 15:45:26 -0700
> From: Paul Eggert <eggert@cs.ucla.edu>
> CC: 9000@debbugs.gnu.org
> 
> On 06/23/2012 12:42 PM, Eli Zaretskii wrote:
> > Never mind, I fixed that myself.
> 
> Unfortunately the patch that you installed can break
> Emacs on hosts where time_t is unsigned, because
> it assigns time_t values to intmax_t variables,
> and this does not always work correctly when
> INTMAX_MAX < TIME_T_MAX.

We could use (UINTMAX_MAX / 2) instead, would that work?

> Which version of GCC are you running?

3.4.2

> What is INTMAX_MAX defined to, on your platform?

It is defined to INT64_MAX (i.e., 9223372036854775807LL).  The problem
is that EMACS_SECS returns a time_t, a 32-bit value, whereas
INTMAX_MAX is a 64-bit value.

> Does it work to pacify your old GCC if you
> replace this:
> 
>       intmax_t secs = EMACS_SECS (t);
> 
>       wait_reading_process_output (min (secs, INTMAX_MAX),
>                                    EMACS_NSECS (t), 0, 0, Qnil, NULL, 0);
> 
> with this?
> 
>       intmax_t max = INTMAX_MAX;
> 
>       wait_reading_process_output (min (secs, max),
>                                    EMACS_NSECS (t), 0, 0, Qnil, NULL, 0);

You don't show what is 'secs' in this snippet.  The key to avoiding
the warning is to have both arguments of 'min' be 64-bit values.