Message-ID: <431E7655.1020204 at rftp.com>
Date: Tue, 06 Sep 2005 22:10:45 -0700
From: Robert Roessler <robertr at rftp.com>
Organization: Robert's High-performance Software
MIME-Version: 1.0
To: lablgtk at math.nagoya-u.ac.jp
Subject: Re: Timeout (windows/linux comparison)
References: <431D8AA6.20102 at laposte.net>
In-Reply-To: <431D8AA6.20102 at laposte.net>
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Transfer-Encoding: 7bit

Matthieu Dubuget wrote:

> I'm trying to draw fast curves on screen using a drawing area from lablgtk2.
> I experienced very low frequencies under windows.
> 
> I did a very small (and dirty) program, and ran it in bytecode version on
> windows and linux in order to understand better my problem.
> 
> It appears that on windows :
> - the timeout is different from the expected one
> - there is a maximum (at about 60 Hz) of the frequency at which I can
> have a timeout work.

I just did my own quick experimentation on my Windows XP system - my
results differ a bit from yours.

Like you, I used Unix.gettimeofday - and I noted right away that using
printf as the timer "ticks" came in really messed up the results. :)

So I set slots in an array with the gettimeofday values and print the
array at the end of the run.

I observed that GLib timeouts could be used down to 10 milliseconds
(100Hz) reliably and without difficulty.

Since it is not clear if this granularity is due to the timeout
facility implementation or to gettimeofday itself, I then tried to run
timeouts with smaller intervals and use GLib GTimers for higher
accuracy in the time *measurement*... and found that GTimers do not
appear to be wrapped by LablGTK (at least as of the last CVS snapshot
of 7/1). :(

So, either these need to be wrapped - or someone more knowledgeable
can steer me to a higher-precision timing facility already available
to OCaml programmers (my non-authoritative quick check of the Hump did
not yield anything suitable).

Robert Roessler
robertr@rftp.com
http://www.rftp.com