Re: [Quickfix-developers] process_sleep has unitialized pointer

[Joerg T. <Joe...@ma...>]

Oren Miller wrote:
> Actually, if we aren't making use of that parameter, we can just use 
> NULL. The only time that the structure is written out if a signal has 
> been delivered to the process, thus interrupting the call and allowing 
> us to continue with the remainder. I don't think this is a situation we 
> need to really concern ourselves with, and it certainly isn't handling 
> this case right now.

I would suggest to do it right now. nanosleep() is the POSIX compliant 
way to wait and it is designed to make restarts after a signal easy:

	while( -1 == nanosleep( &time_to_sleep, &remainder ) ) {
		time_to_sleep = remainder;
         }

> If the structure is set to NULL, the system call 
> will not try to write to the structure. So perhaps we should just change 
> the method to the following.
> 
> void process_sleep( double s )
> { QF_STACK_PUSH(process_sleep)
> 
> #ifdef _MSC_VER
> Sleep( ( long ) s * 1000 );
> #else
> timespec time;

	timespec time, remainder;

> double intpart;
> time.tv_nsec = (int)modf(s, &intpart);
> time.tv_nsec *= 1000000000;

Am I stupid? If I cast the fractional part returned from modf() to int, 
wouldn't it get 0? I would expect something like:

	time.tv_nsec = (long) ((modf(s, &intpart) * 1e6);

> time.tv_sec = (int)intpart;

	time.tv_sec = (time_t) intpart;

> nanosleep(&time, NULL);

	while ( -1 == nanosleep( &time, &remainder ) ) {
	    time = remainder;
	}

> #endif
> 
> QF_STACK_POP
> }

-- 
Joerg Thoennes
                                http://macd.com
Tel.: +49 (0)241 44597-24      Macdonald Associates GmbH
Fax : +49 (0)241 44597-10      Lothringer Str. 52, D-52070 Aachen