Xapian discuss - Aug 2006 - Omega is fast, but not THAT fast

>Search took -125.376129 seconds
I double checked with a handheld stopwatch, and at no
point did the hands spin backwards.

Known problem?

On Sun, Aug 06, 2006 at 10:45:09PM -0700, Jeff Breidenbach
wrote:
> >Search took -125.376129 seconds
> 
> I double checked with a handheld stopwatch, and at no
> point did the hands spin backwards.
> 
> Known problem?
Not one I have seen.  Do you have a reproducible test case?

-- 
Richard

On Mon, Aug 07, 2006 at 08:49:56AM +0100, richard@lemurconsulting.com wrote:
> > >Search took -125.376129 seconds
> > 
> > I double checked with a handheld stopwatch, and at no
> > point did the hands spin backwards.
> > 
> > Known problem?
> 
> Not one I have seen.  Do you have a reproducible test case?
Also, can you say whether the following are set or unset in config.h:

HAVE_GETTIMEOFDAY
FTIME_RETURNS_VOID
HAVE_FTIME

It looks to me like the code is wrong if all three are unset.

James

-- 
/--------------------------------------------------------------------------\
  James Aylett                                                  xapian.org
  james@tartarus.org                               uncertaintydivision.org

On 8/6/06, Jeff Breidenbach <breidenbach@gmail.com>
wrote:
> >Search took -125.376129 seconds
>
> I double checked with a handheld stopwatch, and at no
> point did the hands spin backwards.
>
> Known problem?
Here's the code that sets that up in 0.9.6.  It's in query.cc.  It
depends on what platform you're running the code on.  Probably you
have gettimeofday() I'm guessing...

#ifdef HAVE_GETTIMEOFDAY
            if (gettimeofday(&tv, 0) == 0) {
                sec = tv.tv_sec - sec;
                usec = tv.tv_usec - usec;
                if (usec < 0) {
                    --sec;
                    usec += 1000000;
                }
            } else {
                usec = -1;
            }
#elif defined(FTIME_RETURNS_VOID)
            struct timeb tp;
            ftime(&tp);
            sec = tp.time - sec;
            usec = tp.millitm * 1000 - usec;
            if (usec < 0) {
                --sec;
                usec += 1000000;
            }
#elif defined(HAVE_FTIME)
            struct timeb tp;
            if (ftime(&tp) == 0) {
                sec = tp.time - sec;
                usec = tp.millitm * 1000 - usec;
                if (usec < 0) {
                    --sec;
                    usec += 1000000;
                }
            } else {
                usec = -1;
            }
#else
            usec = time(NULL);
            if (usec != -1) {
                sec = sec - usec;
                usec = 0;
            }
#endif

It actually sets sec and usec earlier but the code is similar.

By coincidence I've been looking into some issues with system time
going backwards recently.  Is there anyway you could run this for 30
seconds:
"while true ; do date ; done | uniq -c".  Probably your system clock
is fine but I'm paranoid right now.

regards,
dan carpenter