[vlc-devel] commit: Workaround braindead QT implementation of ResizeToContents ( Marian Ďurkovič )

[Marian Ďurkovič]

On Tue, 22 Sep 2009 09:03:14 +0200, Rémi Denis-Courmont wrote
> On Tue, 22 Sep 2009 08:02:07 +0200, "Marian Ďurkovič" <md at bts.sk> wrote:
> > Just to give an idea how "efficient" is the QT4's ResizeToContents, here
> > are test results on the same EPG data with different QT4 methods:
> > 
> > 1) default (text cut at right window edge)   7 msec
> > 2) the "right" thing (ResizeToContents)     76 msec
> > 3) my latest workaroud (columnwidth=20000)   7 msec
> > 
> > Results were obtained by putting
> >    msg_Err (p_intf, "update %lld",  mdate());
> 
> Don't do that!
> 
> First, msg_Generic() is not very fast, so you don't want to include 
> it. You should fetch the two timestamps and print them outside of the measurement.
> 
> Second, mdate() uses the monotonic clock. You want to use the thread 
> CPU time clock instead here. With a typical desktop OS scheduler at 
> 100 Hz frequency, any duration over a few milliseconds is likely to be 
> totally off due to preemption.

OK, I've rewritten it using:

    barrier();
    clock_gettime( CLOCK_THREAD_CPUTIME_ID, &t1 );

Yes, msg_Err takes about 1 msec, but otherwise the results on my system are
almost the same:

1)  5.50 msec
2) 77.33 msec
3)  5.64 msec

   With kind regards,

        M.