Re: [coreutils] [PATCH 2/2] stat: print timestamps to full resolution

[Eric Blake]

On 10/01/2010 09:03 AM, Eric Blake wrote:
> > > time_t can be a float on weird platforms I think?
> >
> > If you use nstrftime, no one can complain ;-)
>
> Correct - time_t need not be integral, but do we have any proof of a
> system using a floating-point time_t?

My understanding of POSIX is that time_t was permitted to be floating 
point to allow for subsecond resolution in early versions of the 
standard, in earlier versions of the standard where there was no other 
required support for subsecond resolution (that is, by being 
weasel-wordy about whether time_t was an integer, it was possible to do 
a weirdnix implementation that improved QoI by using floating-point 
time_t).  But now that POSIX 2008 uniformly requires subsecond 
resolution via struct timespec, it makes less sense to have a 
floating-point time_t with a fractional value alongside the tv_nsec 
fractional value.

I think I'm going to write a POSIX bug requesting that time_t be 
tightened to integral in light of the fact that subsecond support is now 
uniformly guaranteed via tv_nsec, and given the argument that existing 
platforms never took advantage of float time_t.

--
Eric Blake   address@hidden    +1-801-349-2682
Libvirt virtualization library http://libvirt.org