Houston, we have a problem

[William Gropp]

At 01:07 PM 8/4/2003 -0500, Jianwei Li wrote:


>         John,
>
>         Thanks for actively testing this new release!
>
> > Jianwei,
> >
> > I finally loaded up 0.8.9 and ran with it.  My run of the Fortran test
> > code appears to put out exactly the same results that you got, except
> > that my read/write times are all zero?  This also leads to a couple of
>
>         You mean both header I/O (define) time and data I/O time are zero?
>         Is that possible while your data is actually written out?
>         //I'll try another run later to see what's going on...

The version of the Fortran test that I saw stored the MPI_Wtime values in 
reals.  This can result in zero differences, because the MPI_Wtime values 
are doubles and may not be normallized to the beginning of the run (e.g., 
it may give seconds since the beginning of the Unix epoch).


> > INF values being output (/0).  My netCDF output appears to be the same
> > as yours, but it does not match the C test code output?  The C output
> > starts 0.001, 0.002, 0.003, ...  The Fortran output starts 65.794,
> > 65.795, 65.796, ...  Could be a problem with my test code?  I was kind
>
>         This is definitely due to difference of your C test and F test:)
>         I checked it, in the Get_Field tri-loop, your F index start
>         from 1, while your C index start from 0, being multipiled by 256
>         makes them so different:)

I've modified the test that John provided (and checked it in) to handle 
both the time values and the initialization of the data.

Bill