Re: virtual variable in LAS

[Jonathan Callahan <callahan@xxxxxxxxxxxxx>]

Zhangfan,

I'm glad it worked so easily!

> >         <u_v units="M/S" name="u v diff"
> > url="file:/mnt/sde1/datasets/adjoint/
> > Vave.des>
> 
> The url value here is faked in order to fool las parser?
> I reasoned that the *.jnl script would preempt *.xml with regard
> to which real varaibles are to be used.

Yes, the LAS parser must find a URL attribute even though we're going to
override it later in the Ferret script.
 
> > let/title="u and v Difference"/units="M/S" u_v = Uave[d=1] - Vave[d=2]
> 
> What does "d=1" and "d=2" mean?

The qualifier [d=N] means use the N'th dataset open.  You can see the
order of datasets with SHOW DATA.

> > ! We must define var$2_grid for use in differencing.
> > ! This grid must reference a file variable: one that actually
> > ! exists in dataset 'dset' as opposed to a virtual variable
> > ! created with an expression involving other variables.
> > define grid/like=Uave[d=1] var1_grid_
> 
> Does var1_grid_ defines a default grid for
> whatever variables to be manipulated below?

In order to be ready for comparison by regridding, Ferret must define a
grid (associated with the first variable) that the second variable will
be regridded to.  The name for this grid is "var1_grid_".  The whole
business of passing in a number to generate [d=1] or [d=2] is a problem
because your modified u_v_diff_init.jnl is actually opening two
datasets.  So any comparison variable should be associated with [d=3].

This [d=#] was originally done because
[d="potentially_long_dataset_name"] occasionally ran int problems by
overflowing the Ferret command line buffer.  I believe that recent
versions of Ferret have greatly increased the command line buffer and
you should probably try changing your std_initialize.jnl file (and
anything based on that) to say:

define grid/like=$4[d=($dset)] var$2_grid_

This will get around the problem mentioned above and *should* work.  But
test it out to be sure.


-- Jonathan