[Olsr-dev] Timing Bug in the Scheduler, with Patch

[Hannes Gredler]

nice catch, many thanks !

added to CVS and HG.

/hannes

Andreas Jacobs wrote:
> Hi,
> 
> I've found a bug in the scheduler that makes the scheduler poll less 
> often than it should.
> 
> Reason:
> After each iteration of the scheduler's main loop the scheduler 
> calculates, how much time this iteration took, and then sleeps for the 
> rest of the scheduler interval. The bug lies in the calculation of the 
> elapsed time, which is converted from a tick count to a 'struct timeval' 
> which consists of a seconds value and a microseconds value. The seconds 
> are filled in correctly, but the microseconds field is filled a 
> milliseconds value instead of a microseconds value.
> 
> Consequently the elapsed time is always underestimated and afterwards 
> the scheduler sleeps too long.
> 
> Patch:
> - against recent mercurial revision (0fc82e1fb7ee)
> - introduces a new variable milliseconds to avoid calulation the 
> milliseconds twice, as it is currently
> - corrects the line which sets the 'tv_usec' (microseconds) field of the 
> 'struct timeval time_used'
> 
> Regards
> Andreas
> 
> 
>