f-tds operations fail when dataset has NaN for missing

[karlmsmith]

Reported by @AnsleyManke on 18 Dec 2014 19:20 UTC
John Maurer reported this.

When we define the transformed variable, the operation consists of defining the variable and inheriting the attributes from the dataset variable.

let/d=1 salt_1_transformed=salt[d=1,z=100:500@AVE]
set att/like=salt salt_1_transformed

If SALT has missing_value flag of NaN, this is incorrect. Ferret isn't going to save the data with NaN's.

Migrated-From: http://dunkel.pmel.noaa.gov/trac/ferret/ticket/2223

[karlmsmith]

Comment by @AnsleyManke on 18 Dec 2014 19:30 UTC
There is a fix needed in the F-TDS scripts that run. We should get the attributes from the dataset variable but then reset the missing-data flag:

let/d=1 salt_1_transformed=salt[d=1,z=100:500@AVE]
set att/like=salt salt_1_transformed

set var/bad=-1.e34 salt_1_transformed

Also, in Ferret there's a mistake in what it does with datasets that have NaN as their missing_value and/or _FillValue attribute.

In Ferret 15 years ago when we implemented NaN as missing values, the method was to substitute the Ferret missing-value flag for NaN as the data is read. The routine mis-defined the Ferret flag, as 1.e-34 instead of -1.e34. This leads to consistent results purely within Ferret, but it's much better to use Ferret's default missing-value flag.

[karlmsmith]

Comment by @AnsleyManke on 18 Dec 2014 19:36 UTC
This is fixed in the LAS iosp script get_datavar.jnl, and in Ferret routines tm_switch_nan.c and cd_get_1_dep_var.F

[karlmsmith]

Comment by @AnsleyManke on 19 Dec 2014 00:03 UTC
Rework this, undoing the changes, once #2224 is completed.