Re: [boost] Should we refine boost::timer

From: Beman Dawes (bdawes_at_[hidden])
Date: 2003-05-20 20:46:31

At 06:44 AM 5/20/2003, Robin Hu wrote:

 >   I once mentioned boost::timer behaviors different in different
 >   platform in this maillist, and Beman said that's the problem of
 >   microsoft himself, because it's he who failed to conform to the
 >   standard. (Hi Beman, am I right? ).

Yep, the C standard says "The clock function returns the implementation's 
best approximation to the processor time used by the program..."

I wouldn't say wall-clock time (as returned by Windows C libraries) is the 
"best approximation", since the platform does support processor time 
functions.

 >   But I think such measurement is used so often, that we shouldnt leave
 >   this problem alone. I'd prefer to refine timer to process_timer and
 >   walltime_timer, and each behaviors the same on different platforms.

I agree with you that we should really have separate timers for processing 
time and wall-clock time.

But I would also like to nail down some of the other specifications. As 
Toon points out, the behavior of clock() as regards wraparound is really 
not reliable.

 >   Here's my simple implementation of new timers, which should work on
 >   both linux and windows nt. I hope you'll intested in this idea and
 >   give some comments. Thanks.

Let's look more seriously at the Boost Date-Time library first. I don't 
want to make any changes to Boost timer without first understanding the 
pros and cons of basing timers on Date-Time.

Do keep pestering me if there isn't more progress on these issues in the 
next week or two!

Thanks,

--Beman