Project Home
Project Home		 Documents
Documents		 Wiki
Wiki		 Discussion Forums
Discussions		 Project Information
Project Info
QNX Operating System/Discussion/OSTech/pthread_cond_timedwait example in docs/List of Posts
Forum Topic - pthread_cond_timedwait example in docs: (4 Items)
   
Update
Ken Schumm
03/24/2009 12:04 PM
post25116
pthread_cond_timedwait example in docs   
I spent some time chasing why a pthread_timedwait() call occasionally returned EINVAL in 3.4. It was my bug, but if the 
given example in the docs was a bit more detailed it would have saved some time so here's a suggestion.

The given example timed wait call sets up a timer like this:

struct timespec to;

clock_gettime(CLOCK_REALTIME, &to);
to.tv_sec += 5;
.
.
pthread_cond_timedwait(...,&to)

That's all well and good for periods measured in seconds.

However, for periods measured in fractions of a second, when the tv_nsec field overflows 1000000000 (one billion) 
pthread_cond_timedwait() returns EINVAL. This is documented in the 3.4 release notes but not in the affected examples.

It might be helpful if the given examples pointed out this issue. Something like this perhaps:

#define BILLION 1000000000
struct timespec to;

clock_gettime(CLOCK_REALTIME,&to);
to.tv_sec += 5;
to.tv_nsec += 5000000 ;
if(to.tv_nsec > BILLION) {
    to.tv_sec++ ;
    to.tv_nsec -= BILLION ;
}

Just a suggestion. The other functions affected by this EINVAL change in the release notes might benefit from this sort 
of clarification as well.
Steve Reid
03/24/2009 12:09 PM
post25117
RE: pthread_cond_timedwait example in docs   
Thanks for the suggestion. I'll create a doc PR.

Steve Reid (stever@qnx.com)
Technical Editor
QNX Software Systems
Shiv Nagarajan(deleted)
03/24/2009 12:10 PM
post25118
Re: pthread_cond_timedwait example in docs   
You could use the timespec2nsec and nsec2timespec functions from libc to 
go from nsecs to a timespec and back

thanks
shiv
Tue Mar 24 12:10:43 EDT 2009

 --> According to Ken Schumm <--
	I spent some time chasing why a pthread_timedwait() call occasionally
	returned EINVAL in 3.4. It was my bug, but if the given example in the
	docs was a bit more detailed it would have saved some time so here's a
	suggestion.
	
	The given example timed wait call sets up a timer like this:
	
	struct timespec to;
	
	clock_gettime(CLOCK_REALTIME, &to);
	to.tv_sec += 5;
	.
	.
	pthread_cond_timedwait(...,&to)
	
	That's all well and good for periods measured in seconds.
	
	However, for periods measured in fractions of a second, when the tv_nsec
	field overflows 1000000000 (one billion) pthread_cond_timedwait()
	returns EINVAL. This is documented in the 3.4 release notes but not in
	the affected examples.
	
	It might be helpful if the given examples pointed out this issue.
	Something like this perhaps:
	
	#define BILLION 1000000000
	struct timespec to;
	
	clock_gettime(CLOCK_REALTIME,&to);
	to.tv_sec += 5;
	to.tv_nsec += 5000000 ;
	if(to.tv_nsec > BILLION) {
	    to.tv_sec++ ;
	    to.tv_nsec -= BILLION ;
	}
	
	Just a suggestion. The other functions affected by this EINVAL change in
	the release notes might benefit from this sort of clarification as well.
	
	
	_______________________________________________
	OSTech
	http://community.qnx.com/sf/go/post25116
	

-- 
****
Shiv Nagarajan,
Kernel Developer, QNX Software Systems,
Ottawa, Canada
****
Ken Schumm
03/24/2009 12:15 PM
post25120
Re: pthread_cond_timedwait example in docs   
Yes, those would be a better way to go.

I just realized my example was also buggy since it only accounted for nsec periods of <= 2 billion and it would still 
have the same problem if nsec was > 2 billion and < ffffffff.


> You could use the timespec2nsec and nsec2timespec functions from libc to 
> go from nsecs to a timespec and back
> 
> thanks
> shiv
> Tue Mar 24 12:10:43 EDT 2009
> 
>  --> According to Ken Schumm <--
> 	I spent some time chasing why a pthread_timedwait() call occasionally
> 	returned EINVAL in 3.4. It was my bug, but if the given example in the
> 	docs was a bit more detailed it would have saved some time so here's a
> 	suggestion.
> 	
> 	The given example timed wait call sets up a timer like this:
> 	
> 	struct timespec to;
> 	
> 	clock_gettime(CLOCK_REALTIME, &to);
> 	to.tv_sec += 5;
> 	.
> 	.
> 	pthread_cond_timedwait(...,&to)
> 	
> 	That's all well and good for periods measured in seconds.
> 	
> 	However, for periods measured in fractions of a second, when the tv_nsec
> 	field overflows 1000000000 (one billion) pthread_cond_timedwait()
> 	returns EINVAL. This is documented in the 3.4 release notes but not in
> 	the affected examples.
> 	
> 	It might be helpful if the given examples pointed out this issue.
> 	Something like this perhaps:
> 	
> 	#define BILLION 1000000000
> 	struct timespec to;
> 	
> 	clock_gettime(CLOCK_REALTIME,&to);
> 	to.tv_sec += 5;
> 	to.tv_nsec += 5000000 ;
> 	if(to.tv_nsec > BILLION) {
> 	    to.tv_sec++ ;
> 	    to.tv_nsec -= BILLION ;
> 	}
> 	
> 	Just a suggestion. The other functions affected by this EINVAL change in
> 	the release notes might benefit from this sort of clarification as well.
> 	
> 	
> 	_______________________________________________
> 	OSTech
> 	http://community.qnx.com/sf/go/post25116
> 	
> 
> -- 
> ****
> Shiv Nagarajan,
> Kernel Developer, QNX Software Systems,
> Ottawa, Canada
> ****

Return


  A subsidiary of BlackBerry             All content ©2004-2013, QNX Software Systems