[gmx-developers] Bug in analysis programs with frame skipping.

[Berk Hess]

ramon at juguete.quim.ucm.es wrote:

>This is to report that affects all analysis programs when the options -dt
>is used. This has been noticed with g_angle.
>
>When a double precision version of these programs is used, 
>and a single precision file is analyzed, randoms frame are skipped.
>
>The problem comes from the function bRmod, that checks whether the timestep
>is a multiple of the user suplied dt. This function has a tolerance that
>depends on the current precision of the build, instead of the precision of
>the source file.
>
>The solution would be to add an additional parameter to bRmod, indicating
>the precision of the data. As a temporary solution, I suggest using the 
>tolerance of single precission.
>  
>
I have implemented a real fix in CVS.
The precision of the trajectory is now passed to bRmod.
I just realized the precision for bRmod should not be double
with single precision code, as t is passed in single precision.

In general it should not be necessary to use double precision
analysis tools on single precision trajectories though.
I have changed a lot of the real variables in the analysis
tools to double precision, so the analysis is always performed
at the highest accuracy. But there are probably several
tools where this still needs to be done.

Berk.