All,
While I agree with most of what is said in this digest, I would take small
issue with this one statement by Harvey Davies:
> Regarding valid_range. The generic conventions clearly state that
> the type must match that of the variable.
I disagree with "clearly". They refer to "the type" of the variable,
when a packed variable clearly has two types, often referred to as "internal"
and "external", but more obviously termed "unpacked" and "packed". The
statement referred to would have been clear, even to a dolt such as myself,
if one of those four words had been inserted between "the" and "type". Thus,
as a person who wrote many datasets that have a valid_range matching the
unpacked type of the variable, I resent being told my data files are
"erroneous" or "non-standard". The standard, due to its vagueness, was
adhered to, despite my data files being in apparent contradiction to the
author's intentions. In the future, I will be using "unpacked_valid_range"
and "valid_range", as I did with my NCEP_DOE AMIP II Reanalysis 2 files.
I'm sorry if my older files broke some software, but my software never had
the slightest problem filtering with "missing_value" rather than valid_range.
I've never understood the insistence that this is not possible or too
difficult, with packed data or unpacked data (threshold comparisons work,
and are simplicity itself to code up).
-Hoop
